<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Peter McCoy on Medium]]></title>
        <description><![CDATA[Stories by Peter McCoy on Medium]]></description>
        <link>https://medium.com/@mccoype_64598?source=rss-9db8f4f15a12------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*qYcCzd52NhA4oBxeyApZew.jpeg</url>
            <title>Stories by Peter McCoy on Medium</title>
            <link>https://medium.com/@mccoype_64598?source=rss-9db8f4f15a12------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 11:40:04 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mccoype_64598/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Cost Illusion]]></title>
            <link>https://medium.com/@mccoype_64598/the-cost-illusion-75740fab841a?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/75740fab841a</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[private-equity]]></category>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Wed, 01 Jul 2026 23:20:39 GMT</pubDate>
            <atom:updated>2026-07-01T23:44:31.144Z</atom:updated>
            <content:encoded><![CDATA[<p><em>On artificial intelligence, moats, and where the durable value in buyouts actually lives.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-sFG8oOU8qOlKx91SdmJxw.png" /></figure><p>Every mania begins with something true. The current enthusiasm for buying labor-heavy businesses and pointing artificial intelligence at them is no exception. It is true that these models can do in minutes what used to take a person a day. It is true that a firm which adopts them early will, for a while, run the same work with fewer people. And it is true that fewer people, holding revenue constant, produce a higher margin. None of that is in dispute, and I do not intend to dispute it.</p><p>The useful question is not whether this is true. It is whether it lasts. First-level thinking stops at a simple syllogism: AI removes cost, so buy the business and remove the cost. Second-level thinking asks the harder question that determines whether you actually made money: if the advantage can be removed, who else can remove it, and how long before they do? An edge that everyone can obtain is not an edge. It is a new baseline. And you are not paid for reaching a baseline your competitors reach on the same schedule, least of all when you paid a premium to reach it first.</p><p>I want to work through this from first principles, because I think a large share of what is being marketed today as an AI buyout is underwriting a moat that will not be there in three years, and because the durable version of the idea is sitting in plain sight, unloved, while the capital runs the other way.</p><p><strong>Cost is not a moat</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-sN2O9c-VFXGHCEqeH9LSg.png" /></figure><p>Begin with the principle, because everything follows from it. A cost advantage is durable only when the source of that advantage is something a competitor cannot also acquire. That is the whole test. If the source is a tool you can buy, then your advantage lasts exactly as long as it takes your competitor to buy the same tool, and not one day longer.</p><p>Consider the accounting roll-ups, since they are the clearest case. The thesis is that you assemble a large, multi-geography firm, layer modern software on top, and turn out tax returns and audits faster and with fewer people. In the short run it works, and the arithmetic is real: each professional becomes more productive, and productivity holds margin. I understand the appeal. But walk the logic all the way to the end and three problems appear, each fatal on its own.</p><p>The first is the ceiling. Making a tax return faster does not create more tax returns. Cost-side automation improves the margin on demand that already exists; it does nothing to the demand itself. So the gains are capped at the labor you are able to remove, and once you have removed it, the well is dry. A business whose entire AI story is efficiency does not have a growth engine. It has a one-time step-change wearing the costume of a growth engine, and step-changes do not compound.</p><p>The second is that speed is the seller’s obsession, not the buyer’s criterion. A client does not choose an accountant because the return comes back on Tuesday instead of Thursday. They choose on trust, on relationships, on the sense that someone competent is watching their situation. When you spend capital optimizing the one variable the customer does not select on, you win an argument nobody is having. You feel faster. The customer feels nothing.</p><p>The third problem is the one the marketing deck leaves out, and it is the one that should stop a serious buyer cold. Most of these firms run on the same handful of platforms: Intuit and its few peers. The sponsor pays a control premium for scale on the theory that a large firm accumulates proprietary data, and that proprietary data will yield a proprietary edge. But the data was never the firm’s. It belonged to the platform the whole time. And the platform has every incentive in the world to build one artificial-intelligence feature and ship it to every customer it has, which is to say, to your competitors, on the morning it launches. In two or three years the small firm across town, running the same software, is just as fast. Very possibly faster, because it does not carry the integration debt of a dozen practices stitched together across geographies and partnership cultures. You will have paid a premium for an advantage the vendor always owned and is about to hand to the entire industry at no charge.</p><p>Put those three together and the picture is unforgiving. You did not buy a moat. You rented a head start, and you paid the price of a moat to rent it. That is the most expensive possible way to be temporarily ahead.</p><p>The ablest people running this strategy will answer that they do not rent their intelligence at all, that they build it themselves. That is the one objection worth taking seriously, and I will give it a serious answer further on. It does not rescue the argument, but it does change where the argument has to be won.</p><p><strong>The outcome the customer never received</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*REOSIib3HkO9V-p1TvorTg.png" /></figure><p>There is a deeper error underneath the cost error, and it is worth naming on its own. These strategies change how a service is produced without changing what the customer receives. The client’s experience is identical. Only the provider’s internal cost structure moved. And an advantage that lives entirely on your side of the table, invisible to the person who is supposed to be choosing you, is not a competitive advantage. It is an efficiency, and efficiencies are shared with everyone the moment the tool that produced them becomes generally available.</p><p>Removing people to lower cost and accelerating a back-office process are the same move wearing two different costumes. Both operate on the seller’s ledger. Neither hands the customer a reason to choose you that they did not already have. If the buyer cannot tell the difference, you have not differentiated. You have only become cheaper to operate, and cheaper-to-operate is a race whose finish line is set by your vendor’s next release.</p><p><strong>How the machine actually works</strong></p><p>If you want to know whether an artificial-intelligence advantage will endure, subject it to three questions. They are not complicated, but almost nothing being funded today survives all three.</p><ol><li>Where does the advantage come from? If it comes from a tool you can purchase, a competitor can purchase the tool. The advantage is borrowed, and borrowed advantages are recalled.</li><li>Does it compound, or is it a single step? Cost removal is a step. You take it once, and then you are standing on the new floor with everyone else. A moat has to widen with use, not settle at a new level.</li><li>Does the customer choose you because of it? If the advantage does not show up as something the customer selects on, it is invisible in the only place that matters.</li></ol><p>Software and data-centered professional services rarely pass even one of these, and almost never all three, for a structural reason: the software is purchasable, the data belongs to the platform, and the customer chose the firm for reasons that have nothing to do with either. This is why the moat wears away so quickly in exactly the places the capital is rushing toward. It was never a moat. It was a feature with good timing.</p><p><strong>The bridge</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hjrxCM9OA-ShgAFNLE6aLg.png" /></figure><p>The durable version of the idea lives in the one place almost no one is willing to work: at the bridge between software and the physical economy. The move is not to buy a business and remove its people. It is to own a physical operation, instrument it, and use the data that operation throws off to deliver an outcome the customer genuinely selects on, and then to lock that outcome into a structure that compounds. Crossing that bridge does not produce a cheaper version of an existing company. It produces a different kind of company, with a different cost structure, a different relationship to its customer, and a different reason to exist.</p><p>The customers who make this work are the ones who run mission-critical physical assets. The pump that idles a plant at six figures an hour. The switchgear standing behind a data center. The refrigeration protecting a season of inventory. The overhead crane that is the literal heart of a manufacturing bay. For these customers one variable dominates all the others, and it is not price and it is not speed of paperwork. It is uptime. Uptime is a variable the customer chooses on, urgently, and pays real money to secure. That is the difference between this and the back office: here, the thing you are good at is the thing the customer actually buys.</p><p>The mechanism runs in two stages. First you win the outcome. You own the parts and the service, you resolve faster than anyone else can, and you become the vendor the customer routes their uptime through rather than one of five they put out to bid. Then you own the book. You instrument the installed base, and the data you accumulate on those specific assets, in those specific plants, comes to predict their failures better than the customer can, and clears a large share of them before a truck ever rolls. That data did not exist until you owned the operation that produces it. No vendor can sell it to your competitor, because your competitor does not have your installed base, your history, or your book. The advantage widens with every asset you add and every fault you resolve. It is the opposite of the cost edge: instead of decaying to a shared baseline, it compounds toward a position no follower can reach.</p><p>This is where the balance sheet earns its keep, and it is the part I would underline. The right structure is recurring contracts in place of transactions, inventory positioned as the competitive wedge, and negative working capital as the goal, with the customer prepaying on contract while you turn parts quickly and stretch supplier terms, so that the business funds its own growth rather than leaning on the next check. A competitor cannot run negative working capital without the contracted base, the density, and the data that produce it. So the structure is not merely the engine. It is the scoreboard. If you are able to run it, the moat is real; if you cannot, you do not have one yet. And the instrumentation that builds the data moat is paid for out of the recurring fee, because the downtime it prevents is worth more to the customer than the sensors cost. You are handed the money to build the very thing that becomes your advantage.</p><p><strong>Owning the tool is not the same as owning the ground</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tqV4_KIsaZ_pkZDetIsA3Q.png" /></figure><p>The most serious objection to everything I have said does not come from the operator who bolts a purchased tool onto a tired business. It comes from the best practitioners in the field, and I want to state it at full strength before I answer it, because a critique that only defeats the weak version is not worth writing. They will say, correctly, that they do not rent their advantage. They own their operations outright. They embed their own engineers inside them. They have made the model provider an equity partner paid in talent rather than a vendor paid in license fees, and they train proprietary systems on their own clients’ work. By the three questions I posed, they will say, we pass: we own the operation, we generate our own data, and the customer chooses us. The real line, they conclude, is not the physical against the digital. It is those who own and build against those who rent and hope, and on that line we are standing beside you, not across from you.</p><p>It is a good argument, made by serious people, and it deserves more than to be waved away. But it points the eye at the wrong axis. The question that decides durability is not whether you rented the tool or built it yourself. It is whether the thing your advantage rests on is excludable: whether a competitor can obtain it without first displacing you from a position that only one party can hold at a time.</p><p>Look closely at what the two kinds of data actually are, because they are not the same kind of thing. The condition history of a critical pump exists in exactly one place, on the machine you are under contract to service, and it exists only because you, and not someone else, are the one on that asset. It is rivalrous in the strict sense: one servicer holds the contract and the sensor feed, and while he holds them, no one else can. A competitor cannot train his way around you, because the raw material does not exist anywhere he is permitted to stand. To get it, he has to take the asset itself, physically, one plant at a time. That is a moat with a wall around it, and the wall is built of steel that someone else owns and you alone service.</p><p>Now look at the proprietary tax system, built well, on real returns, by real engineers. The relationship it lives inside is genuine and sticky, and I will not pretend otherwise. But examine the materials. The training data is tax returns, which are neither scarce nor exclusive; every firm in the country produces the same kind by the thousands, and in machine learning one firm’s returns substitute for another’s about as cleanly as any input ever substitutes for another. The capability doing most of the work is a general one, the preparation of a return, running on a foundation model whose owner is in the business of selling that capability to everyone alive. The proprietary layer is a thin and improving skin stretched over a floor that is rising for the whole industry, because raising that floor for everyone is precisely the model provider’s reason to exist. Equity in the provider changes who profits when the floor rises. It does not make a general capability exclusive to one holder, and it cannot, because exclusivity is the opposite of the provider’s business.</p><p>That distinction produces a result that should give pause to anyone paying a premium for the services version. In the physical business, a better model makes your data more valuable, because you are the only one who holds it, and a sharper tool applied to a rivalrous input widens your lead. In the services business, a better model makes your data less valuable, because the model is now doing more of the job for everyone who can rent it, and the rising floor climbs toward your thin proprietary layer and squeezes it. The same advancing technology cuts in opposite directions in the two arenas. In one it is the wind at your back. In the other it is the tide coming in behind you, and you are wagering that you can build the wall faster than the water rises.</p><p>I am not saying the services version cannot work. The relationship moat is real, and a firm that keeps its clients for a generation can earn a good return for a long time. But be honest about what you are underwriting. Strip the technology away, and most of the advantage that remains is the same trust and switching cost the accounting business always had, that every incumbent already holds, and that the models did not create. On top of that old moat you are buying a temporary lead on a capability the provider is racing to hand to your competitors. That can be a fine business. It is not, underneath, an artificial-intelligence moat. It is an old moat in a new coat, sold at the price of a new one.</p><p><strong>Where the value is, and where the price is</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SlS_0CVbdCv2N-yGgJr-hQ.png" /></figure><p>Hold the two models side by side and notice where the advantage lives in each. In the cost-cutting model, the advantage sits inside a tool the vendor controls and will eventually commoditize, and it shows up on your ledger, where the customer cannot see it. In the physical-digital model, the advantage sits in data you generate and own, it shows up as an outcome the customer deliberately chose, and it grows with use. The first is competed away by definition. The second cannot be competed away without an entrant replicating your operation, your operating history, and your contracted book, which is precisely the thing that cannot be bought.</p><p>So why is the second one available at reasonable prices while the first is bid to premiums? Because capital is drawn to what is fast, clean, and light, and repelled by what is slow, physical, and hard. Humanoid robots and copilots for accountants get the headlines and the funding rounds. The unglamorous assets that actually keep the economy running are ignored. That misallocation is not a problem to be solved before investing. It is the opportunity itself. The best bargains have always been found where other people are not looking and are unwilling to look, and there are few places the modern capital markets are less willing to look than a crane-service business in an industrial park.</p><p>None of this is work for a fund that must show a mark in eighteen months, or for a public roll-up answering to the next quarter. It requires patient, permanent capital and a genuine willingness to buy and operate real companies, which is uncomfortable and slow and exactly the reason the field is not crowded. The difficulty is not a bug in the thesis. It is the moat around the thesis. If it were quick and easy, it would already be done, and the advantage would already be gone.</p><p><strong>In closing</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mJRQdVTOUm8QQfm4H7zReA.png" /></figure><p>I may be wrong about the timing. Manias run further and longer than reason suggests they should, and for a while the cost-cutting story will keep clearing at high prices while the patient version looks dull by comparison. But I am confident about the direction, because it follows from how competition works and not from how I happen to feel about it. An advantage your vendor can hand to your competitors, whether by selling them the same tool or by improving the model beneath the one you built yourself, is not an advantage you own. An advantage that only you can generate, because only you hold the operation and the assets that produce it, is. Today the first is being sold at the price of the second. The entire discipline is in refusing to pay for the first, and in doing the harder work required to build the second.</p><p>The durable companies of this cycle will be built on the bridge between the digital and the physical worlds. Almost no one is willing to cross it. That is precisely why it is worth crossing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=75740fab841a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Father’s Grin]]></title>
            <link>https://medium.com/@mccoype_64598/my-fathers-grin-766cfdcd45c7?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/766cfdcd45c7</guid>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Wed, 01 Jul 2026 13:56:45 GMT</pubDate>
            <atom:updated>2026-07-01T13:56:45.714Z</atom:updated>
            <content:encoded><![CDATA[<p>This morning I was feeding my daughter. She is five months old. When she was happiest her whole face folded up into a grin, and it was my father’s grin. I do not mean a grin like his. I mean his.</p><p>Some of her I can account for. My wife’s nose sits in the middle of her face, small and exact, and that is only right, since she came from the two of us and a nose has to come from somewhere. My own expressions I can explain as well. She has had a hundred mornings of watching me make faces at her over the bottle, and small children are patient students of the people who feed them. She copies what she is given to see.</p><p>She was never given my father to see. He was gone before she arrived. She has had no morning of studying the particular way the corners of his mouth lifted when something pleased him. And there it was anyway, on a face that had no earthly way of knowing it.</p><p>I had Kanye West’s “Waves” on while I fed her. “Waves don’t die,” the hook keeps saying, and the line had been turning over in me all day. A wave is only a motion the water holds for a little while. When it breaks against the shore we say it is finished, though nothing has been lost. The water sets the shape down, and the sea picks it up again further along, and if you did not know better you would swear one wave had run the whole length of the coast.</p><p>My father was a shape the water held for a while, and then I was one, and now she is. His grin did not come down to her the way his watch came down to me. The watch is mine now, and I keep it. No one handed her the grin. It rose on its own, because it had never been his to keep. It was one of the shapes the sea knows how to make, and the sea has not gone anywhere.</p><p>She will not remember this morning. I will keep it for the two of us. And the next time her face folds up like that, as I expect it will, I will know who has come by to see her.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=766cfdcd45c7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Nobody Cares, Just Win]]></title>
            <link>https://medium.com/@mccoype_64598/nobody-cares-just-win-01f1b1db11cc?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/01f1b1db11cc</guid>
            <category><![CDATA[business]]></category>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Sun, 03 May 2026 16:06:21 GMT</pubDate>
            <atom:updated>2026-05-03T16:06:21.012Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1YJ6TYxDzspgViWv.jpg" /></figure><h4><strong>There is a diner on Grand Street in Manhattan called the Landmark. It is drafty in a way that suggests the management has considered fixing the draft and decided against it. The waitress tosses menus onto the table with the mechanical indifference of someone who has been doing this for a very long time and would like you to understand that your arrival has not improved her day. It is, in other words, a perfect New York diner, which is to say, a place that does not care whether you are there or not.</strong></h4><p>My friend Ryan and I have been meeting at the Landmark for about a decade. We started companies around the same time, not together but in parallel, and we fell into a rhythm early on: meet up, trade war stories, try to keep each other from losing our minds. This is a more important function than it sounds. The experience of running a company is, at its core, an experience of radical isolation. Hires don’t sign. Customers churn without explanation. Board members and investors who fought to get into your round, who told you they had conviction, go quiet the moment things get hard. The people around you care, but they don’t care the way you care. They can’t. It’s not their company. It’s not their 3 AM problem.</p><p>Ryan and I cared about each other’s 3 AM problems. That was the whole point of the diner.</p><p>In 1983, Bill Parcells became the head coach of the New York Giants. It was his first head coaching job, and almost immediately, things went wrong. The team was hit with a wave of injuries — the kind that don’t just weaken a roster but reshape it entirely, so that the team you’re coaching on Sunday bears little resemblance to the one you planned for in August. Parcells called his friend and mentor Al Davis, the legendary owner of the Oakland Raiders, and laid out the situation. The injuries. The depth chart. The impossibility of winning under these conditions.</p><p>Davis listened. And then he said something that Parcells would repeat for the rest of his career:</p><p><em>“Bill, nobody cares, just coach your team.”</em></p><p>This is, on its surface, a brutal thing to say to someone asking for help. But Davis wasn’t being cruel. He was making an observation about how the world actually works. No one was going to adjust the standings because Parcells had a good excuse. No one was going to award partial credit. The scoreboard doesn’t have an asterisk column.</p><p>Ben Horowitz wrote a <a href="https://a16z.com/nobody-cares/">wonderful essay</a> about this moment in 2011. You should read it. Horowitz had lived his own version, running Loudcloud while half his customers went bankrupt and the capital markets evaporated — and he distilled the lesson as cleanly as anyone has. <em>A great reason for failing, he wrote, won’t preserve one dollar for your investors, won’t save one employee’s job, won’t get you one new customer.</em></p><p>I’ve been an Al Davis fan and repeating this line to anyone who will listen for as long as I can remember. Horowitz just wrote it down better than I ever could. I should be clear: I am not repackaging his essay as my own. That would be incredibly lame, and I would like to think I have more self-respect than that. But we did arrive at the same place, which either means the idea is universal or that there are only so many things a person can say to another person who is feeling sorry for himself.</p><p>Davis told Parcells what to stop doing. Stop explaining. Stop building the case. Stop looking for someone to acknowledge that the situation is unfair. But there is a second half to the lesson that is, in some ways, more interesting than the first: what to start doing. And the answer is deceptively simple. Win something. Anything. However small.</p><p>Anyone who has run a company during a losing streak knows the feeling. You walk into the office and the energy is different. Not hostile, just flat. People are doing their jobs but they’ve stopped leaning forward. You can feel the organization bracing instead of building. No speech fixes this. No offsite. No all-hands with a new slide deck. The only thing that fixes it is a win.</p><p>You close two small customers, doesn’t matter if they’re not your ideal profile, and suddenly the sales floor has a pulse again. You pull an all-nighter and clear fourteen tickets off the engineering board, and the next morning the team pushes harder the next sprint. Not because you inspired them. Because you won. And they saw it. And now they want to win too.</p><p>But here is what makes winning genuinely dangerous as a force: it doesn’t just restore energy. It increases your gravity.</p><p>Consider what happened to Apple in 1997. The company was ninety days from bankruptcy. The talent had fled. The developers had fled. The customers had all but fled. Then Steve Jobs came back and shipped the iMac, one product, one win, and within a year the engineers who had left were calling to come back, developers were building for the platform again, and consumers were lining up. Nothing about Apple’s underlying technology had changed in those twelve months. What changed was the signal. Apple was winning again, and winning is a signal that is almost impossible to fake.</p><p>This is what most people miss about momentum. They think of it as energy, a feeling that runs through a team. It’s much more than that. When you are winning, customers return your calls faster, because buyers want to adopt solutions from companies they believe are ahead of them. The A-players who ignored your recruiter six months ago start asking around, because the best people want to work alongside other people who are the best, and winning is how they identify each other. You get invited to rooms you didn’t know existed. Conferences. Partnerships. Conversations that were happening without you. None of this is charity. It is a rational response to a signal.</p><p>And the compounding is remarkably fast. A new customer validates the product, which shortens the next sales cycle, which gives your team confidence, which sharpens the demo, which closes the next customer faster. An engineer joins because she heard you were shipping, and her presence makes the team better, which attracts another engineer, who makes the team better still. The velocity doesn’t just increase, it accelerates.</p><p>This is the same instinct that leads professional athletes to leave tens of millions of dollars on the table to sign with a championship contender. We tend to treat this as sentimental — the aging veteran chasing a ring. It isn’t sentimental at all. These athletes have recognized something that most people in business take years to understand: that winning changes the entire environment around you, not just the scoreboard. The tide rises and everything rises with it. But only if you keep winning.</p><p>A few months ago Ryan and I were back at the Landmark. Same booth, same draft coming through the door, same waitress who has never once in ten years asked us how we’re doing. I was in the middle of something at work, the kind of situation where you’ve had the same conversation with four different people and each one has made it worse. I was laying it all out. The context. The history. The reasons it was more complicated than it looked.</p><p>Ryan put his coffee down.</p><p>“Nobody cares, man. Just win.”</p><p>I hadn’t even gotten to the hard part yet. He didn’t need me to. He’d heard enough, not because the details didn’t matter, but because they never do. They never have. Not once, in a decade of sitting across from each other at that diner, has the explanation ever been the thing that fixed the problem.</p><p>It is the best advice I have ever received. About business. About life. About all of it. Al Davis died in October of 2011. Davis left behind a franchise and five words that have meant more to me than anything else I’ve learned about how the world works. I think about him every time I sit down at that diner. The waitress doesn’t care. The coffee is bad. And somewhere, in a silver-and-black tracksuit, the old man is still saying it:</p><p><em>Just win, baby.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=01f1b1db11cc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Assembly Line: How Private Equity Stopped Building Companies and Started Manufacturing Them]]></title>
            <link>https://medium.com/@mccoype_64598/the-assembly-line-how-private-equity-stopped-building-companies-and-started-manufacturing-them-94f44da3d31b?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/94f44da3d31b</guid>
            <category><![CDATA[investing]]></category>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Sat, 14 Mar 2026 17:39:50 GMT</pubDate>
            <atom:updated>2026-03-15T13:43:09.682Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*57R1kyJMoI_fYUZtFm9ltQ.png" /></figure><p>There’s an old joke a trader told me back when I worked at Bridgewater about a man who buys a dog, sells it to his neighbor for twice the price, then the neighbor sells it back to him for three times that. They go back and forth for years, each time at a higher price, congratulating themselves on their investment acumen. Then one day the dog dies. “What a tragedy,” the first man says. “We just lost our best asset.”</p><p>That’s private equity in 2026.</p><p>Not all of it, mind you. But enough of it that the dog is looking very, very tired.</p><p>I’ve been thinking about this since watching the <a href="https://www.youtube.com/watch?v=uVVcyEEess0">Elad Gil fireside chat with Henry Kravis at Stripe last November</a>. There’s Kravis , 81 years old, the man who essentially invented the modern leveraged buyout, and he’s describing what he and George Roberts set out to do in 1976 with $120,000 of their own money. “We like to think of ourselves as industrialists,” he says. “We buy a company and look at what we can do to make it better. How can we improve operations? Maybe it’s fixing the pricing or the supply chain, or putting the proper metrics in place so we can better measure what the company is doing.”</p><p><strong>Industrialists</strong>. He said industrialists. Not assemblers. Not packagers. Not “value-add through multiple arbitrage facilitated by sponsor-to-sponsor secondary transactions leveraging continuation fund technology.”</p><p>Industrialists.</p><p>He even told the story of meeting the head of China’s CSRC, their equivalent of the SEC, who asked Kravis to speak to the country’s top 75 private equity operators. Their message to Kravis was essentially: please explain to our people what you <em>actually do</em> with a company once you buy it, because all they’re doing is buying stock before offerings and dumping it after the price goes up. Kravis was being asked to explain to the Chinese what we, in America, seem to have collectively forgotten.</p><p>Now, Kravis was no angel in the 1980s, the RJR Nabisco deal was as much financial engineering as industrialism, and the man knows how to make money from leverage. But the <em>aspiration</em> he described, the orientation toward making companies operationally better, has been systemically replaced by something else entirely.</p><p>Which brings me to the question that’s been nagging at me, the one that should be nagging at every LP writing a check: <em>when exactly did the private equity industry stop being a collection of maniacs on a mission to make companies better and start being an assembly line that manufactures financial products to sell to one another?</em></p><h3>The Factory Floor</h3><p><strong>Station One: The Lower Middle Market.</strong> A fund, let’s call them Smallco Capital, because their name doesn’t matter, there are hundreds of them, identifies a “fragmented market.” This is the magic phrase that launches a thousand CIMs. Tree services, HVAC, plumbing, pest control, veterinary clinics, dental practices, auto body shops, roofing. The specifics are irrelevant. What matters is that there are lots of small operators, each doing $1 to $5 million in EBITDA, usually run by an owner-operator who’s 58 years old and “has no succession plan.”</p><p>Smallco buys one of these for 5–7x EBITDA. This is the “platform.” They hire a CFO. They centralize QuickBooks into NetSuite. They put in a CRM. They build a branded website. And then they start doing tuck-ins. Three here, four there. Each acquired at 4–6x. They smash them together under one P&amp;L. The combined entity now does $15–20 million in EBITDA. Smallco has created a “story”, which in finance means they’ve taken a spreadsheet and given it a narrative arc.</p><p>I want to pause here because if you described this process to anyone outside of finance, to your dentist, to your brother-in-law, to literally anyone who runs a real business, they’d look at you like you were insane. You bought seven small companies, none of which are growing, smashed them together, didn’t really change how any of them operate, no new products, no new strategy, and now you’re selling the bundle for twice what you paid? And someone is <em>buying</em> it? Yes. That’s Station One. Hold my beer while It gets wilder.</p><p><strong>Station Two: The Middle Market.</strong> Midco Partners picks it up, its likely named afer a rock or a beach, maybe some long gone european trading company. They pay 10–12x for the “platform” that Smallco assembled. They do the same thing Smallco did, but bigger. They buy another $20 million in EBITDA through more tuck-ins. They hire a COO. Maybe they bring in McKinsey for a hundred-day plan, they save the pretty PDF never to see the light of day again. Most of what they do is make it easier for another PE buyer. They get the EBITDA to $40–50 million.</p><p><strong>Station Three: The Upper Middle Market / Mega Fund.</strong> Bigco Capital swoops in at 14–18x. The entity now has enough EBITDA to warrant their fund size. They do a few more acquisitions. They lever it up. They put it on the conveyor belt toward the exit, which means… well, here’s where it gets interesting.</p><p>I sat through a pitch last year where the GP’s entire value creation plan was, I kid you not, a list of fourteen add-on targets, hire BCG and bam its worth 14x. No operating thesis. No go-to-market strategy. Just a shopping list with multiples next to each name and a facy power point. The whole presentation was about <em>assembling</em> the company, not <em>running</em> it. That’s the tell.</p><p>Because the exit used to be to a strategic buyer or the public markets. That was the whole point. You used capital to scale, to give you a buffer of time to improve operations, bring on new expensive talent and to implement new ideas so that an operating company wanted to own or that public investors wanted to buy shares in as it grew fast. The exit validated the enterprise, it proved you’d built something with real, durable, strategic value.</p><p>In the industrialist era, that’s what PE actually did. You bought a company, improved operations, grew the customer base, invested in the product. You kicked the old guy with the Habsburg jaw who’d been asleep at every board meeting off the board. The business got better. Revenue grew because customers noticed. Margins grew because operations tightened. And when it was time to sell, strategics would pay a premium because the company did something they couldn’t replicate easily. The exit worked because the product, the company itself, was genuinely worth owning.</p><p>The assembly line skipped all of that. Nobody along the conveyor belt invested in organic growth because nobody had to. The returns came from multiple expansion at each handoff, buy at 6x, sell at 12x, let the next buyer worry about the business. And by the time the assembled product reaches Station Three, Bigco is sitting on a company doing $100 million in EBITDA that grows organically at maybe 3–4% a year. Morgan Stanley’s research shows that large-cap PE managers grow revenue and EBITDA at roughly a third the rate of middle-market managers. The business underneath the financial packaging is barely growing.</p><p>And now Bigco is stuck. They’re at the top of the pyramid. There’s no Station Four. There’s no bigger sponsor to sell to. The company can’t go public at 3% organic growth, public market investors expect double digits. Strategics don’t want a hollowed-out rollup. So the asset sits. And sits. And eventually, it goes into a continuation vehicle, which we’ll get to.</p><p>But first, the numbers.</p><h3>The Numbers Don’t Lie (But the Multiples Do)</h3><p>A<strong>dd-on acquisitions, the tuck-ins and bolt-ons that fuel the assembly line, now represent 76% of all private equity buyout deals</strong>. In Q2 2025, they hit 75.9%, up above the five-year average. <strong>For context, in 2002, add-ons were 43% of deal activity. In the 1990s, the number was in the low teens</strong>, and they were typically merging assets not acquiring them, meaning they were of equal size. And in Kravis’s era — the 1980s — the concept barely existed. Those deals were whole-company buyouts: RJR Nabisco, Beatrice, Safeway. You bought a company, you fixed it, you sold it. The idea of buying a platform and bolting on dozens of smaller acquisitions to manufacture scale wasn’t a strategy. It was a rounding error. The industry has almost quadrupled its reliance on acquisitive assembly in three decades.</p><p>The exit data tells you where all these assembled products end up. The shift is structural.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xuiBMequUYbTxymEJLuUrw.png" /></figure><p>In the early days of PE , the 1990s and early 2000s , the dominant exit paths were sales to strategic acquirers and IPOs. Throughout the 1990s, PE investments were primarily exited through IPOs and trade sales to operating companies. Secondary buyouts, one sponsor selling to another — were rare and carried a stigma; they were considered a sign that the GP couldn’t find a real buyer. As recently as the mid-2000s, trade sales alone accounted for more than 40% of all buyout exits. These exits demanded something important: the companies had to be <em>good enough</em> that an operating company would pay a premium for the synergies, or that public market investors would underwrite the equity story. The exit was a quality filter.</p><p>Today, IPOs represent a rounding error. In 2024, IPOs comprised just 6% of PE exit value. Meanwhile, sponsor-to-sponsor transactions and continuation vehicles have become the primary exit channel. Secondary buyouts and GP-led transactions now account for a dominant share of how PE returns capital. In 2025, approximately one in five private equity exits involved a continuation vehicle, a structure where the GP essentially moves a company from an old fund into a new one, allowing existing LPs to cash out (at a discount) or roll over.</p><p>The GP-led secondaries market has exploded from $35 billion in 2020 to $115 billion in 2025. It has more than tripled in five years. Total secondary transaction volume hit a record $162 billion in 2024 and surpassed $240 billion in 2025, with GP-led transactions representing nearly half of all secondary activity.</p><p>The private equity industry is increasingly selling companies to itself.</p><p>The distribution data is even worse than the exit data suggests.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CVnrWbRMgvr7Tf6zh3FxnQ.png" /></figure><p>While global buyout AUM has tripled over the past decade, from roughly $1 trillion to $3.8 trillion, distributions as a percentage of NAV have collapsed from an average of 29% during the 2014–2017 peak to under 15% today. That 15% threshold has held for four consecutive years, a record for the modern PE industry. The five-year rolling DPI for buyout funds hit its lowest recorded level in 2025.</p><p>The industry got three and a half times bigger. The cash flowing back to LPs got cut in half.</p><p><strong>During PE’s golden decade, when more exits went to strategics, when fewer identical platforms clogged the market, LPs could expect roughly 29 cents of every dollar of NAV returned as distributions each year. Today it’s 14 cents.</strong> The 2021 exit frenzy briefly spiked the yield back to 38% as sponsors raced to dump assets into a frothy market, but that only makes the subsequent collapse look worse. Sugar high, not recovery.</p><p>It’s like a restaurant that triples its seating capacity but cuts the portion sizes in half. The owner is thrilled, look at all these seats! The customers are starving.</p><p>The organic growth data makes the distribution collapse feel almost inevitable. <strong>The industry reports double-digit revenue growth, PE-backed companies are on track for about 10.6% in 2025, but strip out the add-on acquisitions and the picture collapses. Across the assembly-line verticals, organic EBITDA growth runs at low single digits, typically 2–4% annually. The rest is acquired revenue.</strong> McKinsey’s latest data shows that during PE’s golden decade, a typical deal required only 5% annual EBITDA growth to hit a 2.5x MOIC and deliver a 20% IRR — leverage was cheap, multiples were expanding, the math was forgiving. Today, with borrowing costs at 8–9% and purchase multiples at record highs, the same return target requires 10–12% annual EBITDA growth. The assembly line delivers 2.6%. You don’t need a calculator to see the problem.</p><p><strong>The alpha was never in organic growth. It was in multiple expansion</strong>. Buy at 6x in the lower middle market, assemble, sell at 12x to the middle market, assemble more, sell at 16x to a mega fund. Each handoff to a larger sponsor at a higher multiple was the “return.” The actual business underneath was growing at the rate of GDP or less. And now the belt has stopped. The AUM sitting in what industry insiders euphemistically call “long-dated positions”, and what everyone else calls zombie funds, has swollen to $441 billion in North America alone.</p><h3>The Traffic Jam on the Assembly Line</h3><p>The assembly line doesn’t have one factory. It has <em>hundreds</em>. And they’re all manufacturing the same product, the same way, for the same buyer at the next station.</p><p>There are now over 4,000 private equity firms in the United States. The overwhelming majority, the ones with $200 million to $1 billion funds, are running the identical playbook. They screen for the same thing: fragmented markets, aging owner-operators, $2–5 million EBITDA businesses, recurring revenue, low capex. They read the same Bain reports. They hire the same operating partners out of the same McKinsey class. They screen deals off the same Axial or Grata lists. And they all pitch their LPs the same thesis: “We buy platforms in fragmented industries, execute tuck-in acquisitions, professionalize management, and sell to a larger sponsor at a higher multiple.”</p><p>Any owner-operator who’s listed their business in the last five years can tell you what this looks like from the receiving end. A plumbing company owner in Phoenix or a landscaping business in Charlotte gets dozens of cold emails a month from PE-backed platforms, each promising ‘partnership’ and ‘resources’ and a ‘next chapter.’ The emails are almost identical. The logos are different. The pitch decks are different. The people are different. But the offer, the thesis, the strategy, the projected returns — all the same. I know a guy who runs an HVAC business in New Jersey who started forwarding them to his wife as a bit. She stopped laughing after the fortieth one. He’s gotten sixty-three this year.</p><p>The competition for these small businesses has become so intense that entry multiples in the lower middle market have compressed the arbitrage that made the playbook work in the first place.</p><p>The best investors position capital for where markets are going. The roll-up playbook is the opposite. It’s a <em>backward-looking</em> strategy: extracting the remaining value from a market structure — fragmented small businesses with aging owners, rather than creating new value. And now four thousand funds are all extracting from the same depleting vein.</p><p>In dental alone, there were 161 PE deals in 2024, the highest of any healthcare subsector, with more than 100 PE-backed platforms fighting over independent practices. HVAC, pest control, plumbing, roofing, auto body, veterinary, optometry, name a fragmented service vertical and there’s a dozen sponsors running an identical consolidation thesis.</p><p>When everyone is manufacturing the same product, there is no scarcity at the exit.</p><p>The assembly line worked from 2010 to 2021 because there were relatively few sponsors running these playbooks, rates were near zero, and each successive buyer was willing to pay expanding multiples. Every handoff — from lower middle market to middle market to upper middle market — worked because the next buyer was there, cash was cheap, and multiples were expanding.</p><p>Now the middle-market buyers at Station Two aren’t looking at one or two HVAC platforms. They’re looking at fifteen. They’re not looking at a scarce asset. They’re looking at a commodity. And commodities don’t command premium multiples.</p><p>This is the inevitability of the herd. When everyone piles into the same trade, the trade stops working — not because the thesis was wrong in the abstract, but because the execution became consensus.</p><p>The industry is holding more than 32,000 portfolio companies globally, representing $3.8 trillion in unrealized value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EpaQPc4WYv_PYgM1BhVj_g.png" /></figure><p>Nearly half were acquired since 2020. At recent exit rates, the current inventory represents 8.5 to 9 <em>years</em> of exits. Holding periods at exit now hover around seven years, up from five to six during the golden decade. The assembly line is jammed, not because of interest rates, not because of “market conditions,” but because everyone built the same thing at the same time and there’s a 16,000-company traffic jam of aged assets and no one can get off the highway.</p><p>The only thing keeping the belt moving was an ever-expanding chain of financial buyers willing to pay ever-higher multiples for undifferentiated assets. Remove <em>any</em> link in that chain, higher rates, tighter credit, LP fatigue, multiple compression, and the whole system seizes. There’s no natural buyer of last resort for a $50 million EBITDA HVAC rollup that grows at 4% organically and has no strategic moat. The strategic acquirers don’t want it. The public markets don’t want it. The only buyer is another sponsor who plans to do the same thing the last three sponsors did.</p><p>And that bigger fund? They’re choking on 16,000 unsold companies of their own.</p><h3>The Inventory Problem (Or: How to Destroy a Competitive Advantage in Three Easy Steps)</h3><p>So why are the only buyers other sponsors?</p><p>Because the shift from strategic exits to sponsor-to-sponsor sales isn’t just a market-clearing mechanism. It’s a quality signal. And it’s not a good one.</p><p>If the assembly line were building genuinely valuable companies, strategic buyers would be lined up. They’re not in mass, to be honest some are but very few, and they&#39;re looking to buy an….add-on service. The assembly line has systematically stripped out everything that would make these companies strategically attractive. The only buyers left are other financial sponsors, the only people who evaluate a business through a spreadsheet rather than a strategy lens. A strategic acquirer asks: <em>what does this company do that we can’t do ourselves?</em> A financial buyer asks: <em>what’s the EBITDA and what multiple can I sell it for?</em> The assembly line builds for the second question. Then everyone wonders why the first buyer never shows up.</p><p>Consider the industrial services sector, companies that maintain, repair, and service critical equipment for manufacturers, utilities, and infrastructure operators. The competitive moat in these businesses has always rested on three pillars: skilled technicians, speed of response, and parts availability.</p><p><strong>That third pillar, parts availability, is the one PE keeps destroying.</strong></p><p>When an industrial service provider carries inventory of new and spare parts, it can offer something that competitors without inventory cannot: rapid turnaround. Your CNC machine goes down at 2 AM? The service provider with the right spare motor or control board in their warehouse has your line running by breakfast. The one who has to order from the OEM? You’re waiting days or weeks. In heavy industry, where unplanned downtime costs tens of thousands of dollars per hour, this isn’t a nice-to-have. It’s the entire value proposition.</p><p>But here’s the problem from a financial statement perspective: inventory is a working capital headache. Parts sitting in a warehouse tie up cash. They show up on the balance sheet. They create carrying costs. And distribution of parts, while sticky and high-volume, carries lower margins than pure services. Service labor might run at 45–55% gross margins. Parts distribution might run at 20–30%.</p><p>So what does the PE playbook say? Strip the inventory. Exit the distribution business. Rebrand as a pure-play service company. Margins go up. Working capital comes down. The financial statements look <em>beautiful</em>.</p><p>And the company is now strategically worthless.</p><p>It’s the private equity equivalent of renovating a house by ripping out the kitchen to improve the square-footage-to-bathroom ratio. Sure, the metrics look great on paper. But no one wants to live somewhere they cant eat.</p><p>Without parts availability, you cannot guarantee rapid turnaround. Without rapid turnaround, you cannot guarantee high utilization for your customers’ critical equipment. Without that guarantee, you’re just another body shop competing on price. <strong>You’ve traded a competitive moat for a margin improvement that exists only on paper, because the revenue it protected starts eroding the moment a competitor — or the OEM itself — offers the integrated solution you just abandoned.</strong></p><p>But in a world where the next buyer is another sponsor who evaluates businesses through a financial statement lens rather than a market dynamics lens, having higher margins means they’re willing to pay a higher multiple. And since these assembled businesses rarely grow organically — the whole assembly-line model depends on acquisitive growth, not organic — the multiple expansion is <em>everything</em>. It’s the whole game. You’re not building a business. You’re building a multiple.</p><p>A pretty, pretty bad way to invest.</p><p>The belt has stopped. And the industry’s response has not been to fix the product, build better companies, invest in organic growth, or do any of the hard things that would make these companies attractive to buyers outside the ecosystem.</p><p>The response has been to build a better <em>vehicle</em>.</p><h3>Continuation Vehicles: Selling to Yourself and Calling It a Strategy</h3><p>I need to describe something, and I need you to just sit with how it sounds before I explain why it’s legal.</p><p>A fund manager can’t sell a company. Nobody wants it. The fund is expiring. So the manager creates a <em>new</em> fund, moves the company into it, charges new fees on it, takes new carry on it, and calls it an “exit.” The company didn’t go anywhere. The company is in the same office, with the same people, doing the same thing. But on paper, it “exited.” The manager gets paid. The old investors either take a haircut to leave or pay to stay. And the manager starts the clock again.</p><p>That’s a continuation vehicle. Here’s how it actually works. A GP has a company in Fund III that it can’t sell. The fund is past its term. LPs want their money back, or GP wants them to put those dollars in a new fund. So the GP creates a new vehicle, Fund III-CV, and transfers the company into it. New investors (usually secondaries funds or their existing LPs) come in to provide liquidity to the old LPs who want out. The LPs who like the company can roll over. The GP keeps managing the asset. Everyone’s happy.</p><p>Except.</p><p>If this company is so wonderful — if the value creation thesis is so compelling, if the organic growth story is so strong, <em>why can’t anyone outside the PE ecosystem figure that out?</em> Why is there no strategic buyer willing to pay the premium? Why is there no public market investor willing to underwrite the IPO?</p><p>The answer, in an uncomfortably large number of cases, is that the company isn’t wonderful. It’s a $100 million EBITDA tree servicing business, accounting rollup, HVAC rollup, or dental DSO, or pest control platform, that was manufactured on the assembly line for the specific purpose of being sold to a larger sponsor.</p><p>And these companies can’t go public. Not because of market conditions, because of what they are.</p><p>The public market is somehow the sanity check again.</p><p><strong>In 2024, exactly seven U.S. PE-backed companies went public. Seven. Out of more than 12,000 PE-backed companies in the United States</strong>. The number was the same in 2023. The IPO window isn’t closed, 2025 saw PE-backed IPO value nearly double, but the deals that got out were overwhelmingly tech, healthcare, or infrastructure. The assembly-line rollups didn’t make the cut. And there’s a qualitative reason that no amount of financial engineering can fix: no retail investor — no public school teacher in Des Moines, no firefighter in Tampa contributing to their 401(k), needs to own stock in a pest control holding company. There’s no narrative. No innovation story. The TAM slide would put an equity analyst to sleep. These companies have no natural public market constituency.</p><p>But even if they could get listed, the fundamentals wouldn’t support it. <strong>Public companies in the S&amp;P 500 carry a median debt-to-EBITDA ratio of about 2.6x. PE-backed companies? U.S. LBO debt-to-EBITDA averaged 4.9x in 2024, with interest coverage ratios at 2.4x — the lowest since 2008. These companies carry nearly twice the leverage of their public peers, on slower organic growth, with interest payments that are consuming an increasing share of cash flow.</strong> The S&amp;P 500 grew earnings per share at roughly 10% annually over the last decade. The assembly-line rollups grow organically at low single digits. Public market investors would take one look at the balance sheet and the growth rate and walk away.</p><p>So the company was never designed to go public. It doesn’t grow fast enough organically for the public markets to care. It has no organic growth engine because the management team spent the entire hold period doing add-on acquisitions instead of building a go-to-market strategy. And it doesn’t have sufficient strategic differentiation, because strategic differentiation was systematically removed in favor of margin optimization — for an operating company to want it.</p><p>So it goes into a continuation vehicle. The GP couldn’t find a buyer, not a strategic, not the public markets, not anyone who looked at this company and thought ‘I need to own that.’ So the GP sold it to… a new fund that the GP also manages. And charged a fee for the sale. And then charged a fee to manage it again. If you did this with a house, you’d get audited, and basically be out money on both sides. In private equity, you get a Jefferies fairness opinion and a closing dinner.</p><p><strong>The mechanics are worth understanding.</strong></p><p>The GP-led continuation vehicle market hit $115 billion in 2025. Nearly 75% of the largest global PE firms have done at least one. Single-asset CVs — one company, one vehicle — represent 48% of the GP-led market.</p><p>Most large institutional LPs, the pension funds, the endowments, the sovereign wealth funds, have re-up rates with their GPs that are extremely high. These relationships are sticky. A major state pension that invested in Fund I, II, and III is almost certainly going to invest in Fund IV and Fund V. The GP knows this. The LP knows this. Everyone knows this.</p><p>So when that GP moves a company from Fund III into a continuation vehicle, and the pension fund that’s in Fund III is also likely to be in the CV (or in the next flagship fund), what’s actually happening? <em>The LP is selling to itself.</em></p><p>The pension fund has $50 million in Fund III. The GP can’t sell the tree servicing company. So they create a CV. The pension can either: (a) cash out, receiving liquidity at whatever price the GP and their handpicked fairness opinion advisor have determined, or (b) roll into the CV.</p><p>If the pension cashes out, they take a discount, average CV pricing declined to 92% of NAV in 2025. They eat a loss on paper. The GP books an exit. DPI goes up. The GP uses that improved DPI number to fundraise for Fund V. The pension commits to Fund V. The cash that just came out of Fund III goes right back in.</p><p>If the pension rolls over, they’re now in the CV alongside new secondaries investors. The GP resets carry. The GP resets fees. The GP gets a new management fee stream on an asset they’ve already been managing for seven years, and critically, the management fee basis is now <em>higher</em> than it was in the original fund, because the asset’s NAV has been marked up through years of add-on acquisitions and EBITDA growth. A company that entered Fund III at $50 million EBITDA and was assembled to $100 million through tuck-ins now generates management fees on a much larger base. The GP didn’t just get a second bite at carry — they got a bigger plate for the management fee. The pension pays carry on the same asset twice — once when the “exit” from Fund III is booked, and again when the CV eventually exits. Phalippou at Oxford has shown this explicitly: LPs who opt into continuation vehicles pay higher performance fees than they would have in the original fund structures, because the carry hurdle resets.</p><p><strong>A WhiteHorse Liquidity Partners analysis of over 1,000 CV transactions found that assets were valued an average of 28% higher than where the GP was marking them just six months earlier</strong>. The GP marks the asset up, books an exit at the higher valuation, charges carry on the uplift, and transfers the asset to a new vehicle where they’ll charge carry again.</p><p><strong><em>This is selling from your left pocket to your right pocket</em></strong><em>.</em> The GP gets paid twice. The LP either takes a haircut to exit or pays double fees to stay. The company itself, the actual operating business — is unchanged. No strategic buyer showed up. The public markets never validated the price. The only “exit” was a transaction between two vehicles controlled by the same manager, priced by an advisor selected by that manager, at a valuation that manager determined.</p><p><strong>Only 17% of incumbent LPs roll their stakes into continuation vehicles, according to Jefferies. The other 83% take cash and leave. In an ILPA webinar, more than 60% of LP participants said they preferred liquidity from traditional exits, <em>even at a discount to CV valuations</em></strong>. And when you look at who <em>does</em> come into CVs, the composition is revealing: traditional institutional investors collapse from the majority to a small minority. Public pensions fall to just 10% of new CV investors. <strong>The dominant new capital — two-thirds of it — comes from financial intermediaries</strong>, the secondaries funds and semi-liquid retail vehicles that have raised massive pools specifically to deploy into these structures.</p><p><strong>The institutional investors, the people with the longest experience evaluating PE, are voting with their feet</strong>. They don’t believe the story. They take the cash and move on. The GP replaces them with secondaries capital that was raised specifically to buy these assets, at prices that secondaries funds believe offer a discount to intrinsic value.</p><h3>Incentives Explain Everything</h3><p>Most of what confuses people about the PE industry becomes instantly clear once you look at the incentive structure.</p><p>A GP earns management fees on committed capital, typically 2% per year. They earn carried interest, typically 20%, on returns above a hurdle rate. And they raise new funds every 3–5 years.</p><p>Under this structure, the GP is incentivized to: (a) raise ever-larger funds, (b) deploy capital quickly, mark assets at favorable valuations, (d) generate DPI through whatever means necessary to enable the next fundraise, and (e) hold onto winners as long as possible to maximize carry.</p><p>Notice what’s missing from this list: <em>build great companies</em>. There is no management fee for organic revenue growth. There is no carry specifically attributable to operational improvement versus financial engineering. The incentive structure rewards <em>activity</em> — acquisitions, assemblages, financings, exits, re-entries — not <em>quality</em>.</p><p>The assembly line is the rational response to this incentive structure — and it’s important to be honest about why. It’s not that GPs don’t <em>want</em> to build great companies. Many do. Some are too deep in the playbook to realize they aren’t. But the fundamental problem is simpler than villainy: improving organic growth is <em>hard</em>. It’s uncertain, it takes years, and there’s no guarantee it works. You can hire the best sales team in the world and still miss your number. R&amp;D spending doesn’t guarantee anything. A proprietary service model might not move the needle at all. Organic growth is a bet on execution with no guaranteed payoff.</p><p>Financial engineering is easy. Buy a company at 6x, bolt on three more, sell the package at 12x. The math is clean. The playbook is proven. The returns are modelable. Putting lipstick on a pig is a repeatable, scalable process. Building a better pig is not. So the industry, rationally, chose the repeatable path.</p><p>Of course this is what happened. What did anyone expect?</p><h3>Things That Can’t Go On Forever, Don’t</h3><p>Trees don’t grow to the sky. The PE assembly line depends on three conditions, and all three are deteriorating at once:</p><p><strong>First, a perpetually expanding pool of larger sponsors willing to pay higher multiples.</strong> But the industry has hit a wall. Median purchase multiples are at record highs. The backlog of unsold companies is at record highs. And since 2006, PE returns have largely tracked the S&amp;P 500 — an asset class you can buy for five basis points in an index fund, without lockups, without illiquidity, and without a 2-and-20 fee structure. Between 2022 and Q3 2025, US PE funds returned an annualized 5.8%, compared to 11.6% for the S&amp;P 500. The willingness to keep paying up for the next turn of the assembly line is diminishing.</p><p><strong>Second, LP willingness to keep recycling capital into new funds.</strong> But with distributions at historic lows and fundraising declining, this flywheel is broken. U.S. buyout fundraising declined 20% in 2024, and the number of funds fell by nearly 50%. In 2025, fundraising fell another 22% year-over-year. The time to raise a fund has stretched from 11 months pre-pandemic to 20 months.</p><p><strong>Third, cheap debt to grease the assembly line.</strong> The zero-interest-rate environment that powered the 2010–2021 boom is over. Interest coverage ratios among buyout-backed companies dropped to 2.4x in the U.S. — the lowest since 2008. Leverage, which StepStone estimates accounted for roughly 45% of PE returns between 2010 and 2022, can no longer do the heavy lifting. And roughly $300 billion in leveraged loans came due by the end of 2025, requiring refinancing at rates that are materially higher than the original terms.</p><p>If the traditional return drivers — leverage, multiple expansion, and cheap financing — are “largely spent” (McKinsey’s words, not mine), what exactly are LPs paying 2-and-20 for?</p><p>The answer is supposed to be operational value creation. But operational value creation is hard. It requires industry expertise, patience, and a willingness to invest in things that don’t immediately improve the financial statements — like carrying spare parts inventory, or building a sales force, or developing proprietary technology. It requires treating a company as a <em>business</em> rather than a <em>financial product</em>.</p><p>And the assembly line doesn’t do hard.</p><h3>The Returns Tell the Story</h3><p>Everything in this piece, the assembly line, the traffic jam, the stripped moats, the continuation vehicles, the incentive structure, describes the mechanism. The returns are the verdict.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2VGlcKq-8xtRbcHm7r0v4g.png" /></figure><p>In the early 1980s, when Kravis and a handful of firms were buying whole companies and operating them as industrialists, U.S. buyout funds generated excess returns over the S&amp;P 500 of roughly 1,600 basis points per year. The average fund delivered IRRs around 25% with multiples of 2.5x. There were fewer than 200 firms. Add-on acquisitions were a rounding error. The exits went to strategics and IPOs. The alpha was real, and it was enormous.</p><p>By the 1990s, as the industry grew and the playbook began to standardize, excess returns fell to roughly 300–470 basis points per year. Still strong. Still worth the illiquidity. But already compressing.</p><p>By the 2000s, with over 2,500 firms and add-ons approaching half of all deal activity, the median fund’s outperformance dropped to roughly 150 basis points per year — well below the 300 basis points most investors view as the minimum compensation for PE’s illiquidity and risk.</p><p>And since 2006, the era of the fully industrialized assembly line, with 4,000+ firms and add-ons at 76% of deals — U.S. buyout funds have, on average, performed approximately in line with the S&amp;P 500. The median fund no longer clears the illiquidity premium. Between 2015 and 2025, the top-quartile buyout managers still delivered a 20.7% IRR. But the median delivered roughly 12% — barely distinguishable from a low-cost index fund after accounting for fees, lockup, and risk. CalPERS, the largest pension fund in America, reports a since-inception net IRR of 11.2% and a net multiple of 1.5x across its entire PE portfolio. That’s not terrible. But it’s not what LPs signed up for when the pitch was “private equity as an asset class.”</p><p>Now, some of this return compression is just math, more capital chasing the same opportunity set will compress returns in any asset class. I’m not naive about that. But it’s not the whole story. If capital inflows were the only issue, you’d expect returns to compress evenly across all strategies. They haven’t.</p><p>When a thousand firms are running the same playbook on the same targets with the same capital structure, the alpha compresses to a commodity margin. One competitor’s margin becomes another’s opportunity. The trade gets crowded. The excess return gets competed away. The assembly line turned PE from a craft into a commodity — and commodity producers don’t earn premium returns.</p><p>The proof is in the exception. McKinsey’s 2026 data shows specialist buyout funds, the ones doing something closer to what Kravis described, with deep sector expertise and real operational work, delivered 17% pooled IRR across 2010–2022 vintages versus 13% for generalists. Cambridge Associates found an even wider gap: sector specialists at 23.2% gross IRR versus 17.5% for generalists, with higher multiples and fewer losses. The firms still operating as industrialists are still earning industrialist returns.</p><p>The assembly line isn’t.</p><p>And here’s what makes this self-correcting mechanism unlikely to kick in: the LPs know all of this, and they’re responding rationally, by moving <em>up</em> the funnel. If most exits are to other funds, then the safest place to be is at the top of the machine, not the bottom. Lower middle market and middle market funds don’t need to be industrialists because they’re not betting on the company. They’re betting on the conveyor belt. They trust that a bigger fund at the next station will buy what they’ve assembled. The exit isn’t contingent on building something valuable. It’s contingent on the machine continuing to move. So the capital flows to the top of the funnel, which manufactures more product, which clogs the exit market further, which creates more continuation vehicles. The machine feeds itself. And everyone along the way is behaving rationally — which is exactly what makes it so hard to stop.</p><h3>The Simple Version</h3><p>Sometimes the most complex situations deserve the simplest observation.</p><p>The private equity industry has created a very elaborate, very expensive, very sophisticated system for buying dogs, selling them to each other at ever-higher prices, and then when nobody wants to buy the dog anymore, creating a special vehicle to sell the dog to a slightly different version of themselves.</p><p>And now there are 16,000 dogs that nobody wants, the people who paid for the dogs haven’t gotten any money back, the people who sell the dogs are having trouble raising new money to buy more dogs, and the solution the industry has come up with is: <em>more vehicles</em>.</p><p>Henry Kravis started KKR to be an industrialist, to buy companies and make them better. To fix the supply chain, improve operations, develop talent, and create durable competitive advantages. Fifty years later, an industry built on his vision has devolved into a conveyor belt that manufactures undifferentiated financial products, strips them of strategic value in the name of margin improvement, and trades them between sponsors in an increasingly circular game of multiple arbitrage.</p><p>The lack of liquidity in PE isn’t a market problem. It’s a <em>system</em> problem. The system is increasingly designed and incentivized to manufacture lower-quality companies and sell them to one another. The companies can’t go public because they don’t grow fast enough and no retail investor wants to own them. They can’t sell to strategics because the assembly line stripped out everything that made them strategically valuable. The only remaining buyer is another sponsor who will run the same playbook, hold longer, and eventually face the same exit problem at a higher cost basis.</p><p>Continuation vehicles aren’t a solution. They’re a confession.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=94f44da3d31b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Return of Steel: Why AI Will Force the Enterprise Back On-Premise]]></title>
            <link>https://medium.com/@mccoype_64598/the-return-of-steel-why-ai-will-force-the-enterprise-back-on-premise-256781c15206?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/256781c15206</guid>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Thu, 19 Feb 2026 04:17:33 GMT</pubDate>
            <atom:updated>2026-02-19T04:17:33.758Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PsV5wEpYYm_UnvluGBMvyg.png" /></figure><p>Last Tuesday I hit my weekly usage limit on Claude. Not as an engineer. Not as some power-user developer running automated agents. Just me — a person with a laptop, working through deal memos, drafting LP updates, and asking the kind of questions you used to pay analysts $150,000 a year to answer slowly. One week of normal thinking and the machine said: that’s enough.</p><p>That’s when it clicked for me. The constraint isn’t intelligence anymore. We cracked intelligence — or something close enough that the distinction barely matters for most business applications. The constraint, as it always has been throughout economic history, is the physical substrate that intelligence runs on. Compute. Energy. Steel.</p><blockquote>Every great technological wave eventually runs into physics. The industrial revolution ran into coal and iron. The internet ran into fiber and bandwidth. AI is running into GPUs and megawatts — and the companies that own that physical layer are about to become extraordinarily powerful.</blockquote><p>Every great capital cycle follows the same arc: an insight about productivity creates a consensus trade, the consensus trade attracts excess capital, the excess capital creates the conditions for the reversal. The consensus trade right now is everything in the public cloud, rented from the hyperscalers, with AI as a utility you access rather than an asset you own. That consensus is not wrong — it’s just early-cycle thinking being applied to a late-cycle problem. The second-order effect that most strategists haven’t priced in is straightforward: the enterprise is going to want its own silicon. And it’s going to need it for reasons that go far beyond cost.</p><h3>II. Tokens Are the New Headcount — and They’re More Expensive</h3><p>Here is a fact about human behavior that economists find endlessly interesting and everyone else finds obvious: when something gets dramatically cheaper, people don’t pocket the savings. They spend more.</p><p>In 1908, Henry Ford introduced the Model T at $850 — roughly $28,000 in today’s money. A car was a luxury. By 1925, Ford had gotten the price down to $260 through the magic of the assembly line. Now here’s the part that matters: total American spending on automobiles didn’t fall. It exploded. Because Ford hadn’t just made the same product cheaper. He’d made it cheap enough that an entirely new class of buyer walked into the dealership. The farmer who’d never considered owning a car suddenly could. The schoolteacher. The small-town doctor. The denominator changed. And when the denominator changes — when the number of people who can afford something goes from “the wealthy” to “everyone” — the aggregate spend doesn’t shrink. It multiplies. Exposed demand doesn’t reduce your bill. It creates a bill you didn’t know you had.</p><p>AI tokens have just done the exact same thing.</p><p>In late 2022, processing one million tokens — roughly the text of four novels — cost $20. By mid-2025, that same million tokens cost $0.40. A 98% price decline in under three years. And the engineers and product managers who saw that chart drew the obvious, rational, completely wrong conclusion: AI is getting cheap, so our bills will go down. They were wrong because they made the classic mistake of confusing the price of a unit with the cost of a habit. When something gets 50 times cheaper, people don’t buy the same amount of it. They buy 100 times more. The per-unit price collapses. The total bill doubles. It’s not a paradox. It’s just math that most people run in the wrong direction.</p><p>The average mid-sized enterprise was already spending $85,521 per month on AI by the end of 2025 — and according to an Andreessen Horowitz survey of 100 enterprise CIOs, they expect 75% growth in LLM budgets this year alone. One CIO put it plainly: “What I spent in 2023, I now spend in a week.” In 2024, innovation budgets still made up 25% of LLM spending — the money companies set aside to explore and experiment. That figure has collapsed to 7%. AI has graduated from the R&amp;D line item to the operating budget, which means the spending no longer stops when the experiment ends. It compounds.</p><p>And here is the specific mechanism that almost nobody anticipated — the thing that turned a pricing decline into a spending crisis.</p><p>The new generation of reasoning models — the ones that actually think before they answer — don’t just process your question. They argue with themselves first. When you ask a reasoning model to review a contract, it doesn’t just read the contract. It generates hypothetical counterarguments, checks its own logic, considers edge cases, and then — only then — responds. That internal argument burns tokens. Thousands of them. Before you see a single word of output. One company’s AI agent, doing nothing unusual, consumed 10 billion tokens in a single month. That is the equivalent of reading every book in a large university library. Twice. The model got cheaper. The model also got hungrier. And the hunger won.</p><blockquote>Tokens got 50x cheaper. Usage went up 100x. The bill doubled. This is not a pricing problem. It is a consumption problem — and consumption problems don’t get solved by negotiating better rates.</blockquote><p>Shopify’s VP of Engineering said publicly that individual engineers on his team are spending $1,000 to $10,000 a month in LLM costs. A single engineer. Per month. The median ML engineer earns $162,000 a year — roughly $13,500 a month in salary. So at the high end, one engineer’s AI bill is approaching what you’d pay another engineer to sit next to them. Now multiply that across a team running 50 agentic workflows around the clock. You are no longer buying software. You are staffing a department — except this department sends you an invoice that grows every time it does more work, forever, with no headcount cap and no salary freeze. It’s the only employee in history who gets a raise every time it does its job. And you can’t even talk to it about its performance.</p><p>Tell me the incentive and I’ll tell you the outcome. The incentive built into cloud AI pricing is elegant and invisible. Make the unit cost low enough that any individual purchase seems reasonable. Watch the aggregate become unreasonable before anyone notices. By the time the CFO starts asking questions, the workflows are embedded, the engineers are dependent, and switching costs are real. The safest road to hell is the gentle slope — soft underfoot, without sudden turnings, without milestones, without signposts. No single line item shocks you. The slope is always gentle. And then you’re at the bottom, wondering how a 98% price decline made you spend more money.</p><p>Now, every industry that has ever run a high-volume, predictable physical process at scale has learned the same lesson: when the volume is certain and the process is continuous, ownership beats rental. Shipping companies own ships. Airlines own planes. Manufacturers own factories. And I know what you’re thinking — “own” is doing a lot of heavy lifting in that sentence. You’re right. Delta doesn’t have $60 billion in cash stuffed under a mattress. They own the planes the way most people own their house: the bank has a very strong opinion about whether they keep making payments. The fleet sits on the balance sheet next to $18 billion in debt. United, same story. Maersk owns ships that were financed by syndicates of European banks who have very specific feelings about container rates.</p><p>But here’s the thing — and this is the insight that matters for AI infrastructure: Delta chose the debt. They chose it because the alternative was worse. An airline that leases its entire fleet from a third party pays more per flight-hour over time, has no residual asset value, and — critically — is subject to the lessor’s decisions about pricing, availability, and fleet allocation during the exact moments when capacity matters most. The ownership isn’t free. It comes with leverage, maintenance obligations, and balance sheet risk. But it also comes with control. And in a business where the asset IS the operation — where the plane is the product, where the ship is the service — control is what separates an operator from a tenant. The captain of a leased vessel is still at sea. He’s just also at the mercy of whoever owns the hull.</p><p>AI token consumption inside a large enterprise — every claim processed, every contract reviewed, every wire transfer screened — is not experimental. It is the production line. And you don’t rent your production line. Not because buying is always cheaper than renting in the abstract — it isn’t — but because when you know you’ll need the capacity every single day, paying the landlord a margin on top of the cost of that capacity is money you’re donating to someone else’s shareholders. And if that landlord also happens to be one of the most strategically ambitious companies in the history of capitalism — well, you might want to think carefully about what else you’re donating.</p><h3>III. The Hidden Tax: Your IP Is Paying the Model’s Tuition</h3><p>There’s a story that the security people in enterprise technology like to tell, and it goes like this: the lock on your front door doesn’t stop a determined burglar. What it does is make the burglar go to your neighbor’s house. Security, in other words, is not about being impenetrable. It’s about being harder to penetrate than the alternative.</p><p>This is a comforting story if you live in a neighborhood. It is a terrible story if you live in an apartment building — because in an apartment building, every tenant shares the same front door. The burglar doesn’t need to pick your lock. He picks the building’s lock, and now he’s in the hallway.</p><p>Enterprise AI, as currently architected, is an apartment building.</p><p>Every prompt you send to a shared AI service is signal. Every correction you make to an output is training data. The enterprise-tier agreements limit what gets used for model improvement — but they don’t eliminate the exposure, they contractualize it. And here’s the thing about contracts in technology: they describe what a vendor promises not to do. They do not describe what a nation-state adversary, a rogue employee, or a sophisticated attacker can do once the data has left your building.</p><p>And this is where it gets philosophically interesting — in the way that should make a CIO lose sleep, not win a dinner party argument. What we call the enterprise’s “power over AI” — the ability to automate claims processing, to screen transactions, to extract insights from decades of proprietary data — turns out to be a power exercised by the infrastructure owner over the enterprise, with the enterprise’s own data as the instrument. You think you are using the model. But the model’s owner is using you — your queries, your corrections, your patterns of thought — to make the model better for everyone, including your competitors. The power doesn’t flow outward to the user. It flows upward to the platform. Every tool you adopt from a shared infrastructure is a tool that works for you and simultaneously teaches someone else how to build a better version of you. That’s not a partnership. That’s a very expensive education you’re funding for someone else’s benefit.</p><p>The breach record makes the abstract concrete. In November 2025, attackers compromised an OpenAI vendor and exfiltrated names, emails, and technical system details of API business customers. An alleged breach of 20 million credentials circulated on dark web forums in early 2025. And then the big one — the one that should have changed the conversation entirely: Chinese state-sponsored actors used jailbroken Claude Code to execute autonomous cyberattacks against 30 global organizations. The AI handled 80 to 90 percent of each intrusion with minimal human oversight.</p><p>Read that again. The adversary didn’t breach the enterprise directly. They operated through the AI infrastructure both parties were using. They walked through the shared front door. And somewhere, a very expensive security consultant is updating a slide deck that still says “defense in depth” on it.</p><blockquote>Your data is a moat. But a moat only works if the castle controls the drawbridge. The moment you route your most sensitive intellectual property through someone else’s server, you’ve handed the drawbridge keys to a landlord — and you may never know who else has a copy.</blockquote><p>The legal sector understood this first, for the simple reason that their entire business model depends on the concept of privilege. Attorney-client privilege doesn’t survive a trip through a third-party API. That’s not a theoretical concern — it is a factual statement about how privilege law works, and every Am Law 100 firm has now grappled with it. Harvey, the AI legal research platform, exists because of this fact: it was built from the ground up on the premise that privileged information must run on controlled infrastructure or it isn’t privileged anymore. Financial services is a step behind but moving fast. Insurance is next. Healthcare is already there by regulatory mandate.</p><p>This is not a cost decision. It is a liability decision, a competitive strategy decision, and increasingly a regulatory compliance decision — GDPR, HIPAA, SOX, and a growing stack of AI-specific regulation all create legal exposure for enterprises that cannot tell a regulator, with certainty, where their data was processed, by whom, and under what conditions. The enterprise that can answer those questions in discovery is not the one running on shared cloud APIs.</p><h3>IV. The Energy Equation: Physics Always Wins</h3><p>Markets are pendulums. They swing from fear to greed, from scarcity to glut, and the investor’s only real job is to notice where on the arc they’re standing — not where the narrative says they should be. The narrative right now says AI infrastructure is a sure thing. The capital structure says something more interesting.</p><p>Here is what is actually happening. Amazon will spend $200 billion on AI infrastructure in 2026. Google has committed up to $185 billion. Microsoft is tracking toward $120 billion or more. Meta has guided to $135 billion. Together, the four largest hyperscalers will spend somewhere between $630 and $700 billion this year alone — a 60% increase from 2025, which was itself a record. To put that number in human terms: it is more than the entire GDP of Sweden, deployed in a single year, into data centers and GPUs. Wall Street estimated the increase would be 19%. The actual number is three times that. Analysts have underestimated AI infrastructure spending every year since 2022. The question worth asking is not whether they are right or wrong, but what it means when an industry consistently surprises consensus by factors, not percentages.</p><p>The scale of this commitment has begun to do something that would have been unthinkable five years ago: it is consuming the cash flows of the most profitable companies in the history of capitalism.</p><p>Let me make this concrete, because the numbers are so large they can slide past you if you’re not careful. Amazon generated $139 billion in operating cash flow last year — a record for any company, ever. Morgan Stanley now projects Amazon will run negative free cash flow of $17 billion in 2026. Alphabet’s free cash flow is expected to crater 90%, from $73 billion in 2025 to $8 billion. The company has already issued $25 billion in bonds and quadrupled its long-term debt in a single year. Barclays is modeling Meta at negative free cash flow not just in 2026 but through 2027 and 2028 as well.</p><p>The richest companies in human history are borrowing money. Just sit with that for a second. These aren’t WeWork. These aren’t cash-burning startups with a dream and a pitch deck. These are the most profitable businesses ever built by human beings, and they are spending faster than they earn. If your buddy who pulls in $500,000 a year told you he just took out a home equity loan to fund a hobby, you’d have questions. You’d have a <em>lot</em> of questions. You certainly wouldn’t hand him your retirement account and say “sounds great, keep going.”</p><blockquote>When the most cash-generative companies on earth need to tap the debt markets to fund their capital plans, you are not in the early innings of a normal technology cycle. You are somewhere more interesting than that.</blockquote><p>I want to explain why this matters to an enterprise CIO who thinks the hyperscaler spending binge is someone else’s problem. It isn’t.</p><p>Debt cycles have a specific character. They expand when borrowing costs are manageable and the returns on invested capital exceed the cost of servicing the debt. They contract — always faster than anyone expects — when that relationship inverts. The hyperscaler debt cycle is not at the breaking point. These companies have strong balance sheets and genuine revenue growth. But the direction is worth understanding: the five hyperscalers plan to add roughly $2 trillion in AI-related assets to their balance sheets by 2030. Those assets depreciate at roughly 20% per year. The annual depreciation bill alone will eventually exceed their combined 2025 profits. The AI revenue required to make this math work is not a small number — sustaining the buildout through 2030 requires AI revenue approaching $2 trillion annually, which is five times the size of the entire global subscription software market today.</p><p>That may happen. It is not guaranteed. And the enterprises renting compute from a capital structure under that kind of pressure should understand what they’re actually dependent on: the continued ability of the hyperscalers to subsidize infrastructure costs with investor capital during a build phase that hasn’t yet proven its return on investment. That’s not a criticism. It’s just a description of the physics of the situation.</p><p>Here is the conflict that follows logically from that physics, and that almost nobody is naming directly. AWS’s business model requires enterprises to rent compute indefinitely. The enterprise’s business model requires building competitive moats that don’t depend on a landlord. These interests are structurally opposed, and they will diverge further as AI moves from experimental to mission-critical. Every dollar an enterprise spends on cloud inference is a dollar that flows to a company whose strategic interest is to deepen the enterprise’s dependency — not reduce it. The hyperscalers are not villains. They are rational actors in a system where your lock-in is their revenue. Understanding that is the beginning of the infrastructure decision. Not being upset about it — just understanding it. The way you understand that the casino is designed for you to lose. You don’t get mad at the casino. You just stop pretending it’s a savings account.</p><p>And now the part of the story where it gets really interesting — the part where the AI race becomes an energy race, which is a polite way of saying the AI race ran headfirst into the laws of thermodynamics and thermodynamics won.</p><p>Google has acquired Intersect Power — not a software company, not a model lab, but a power company — for $4.75 billion. Microsoft has struck nuclear deals with Constellation Energy and Helion. Amazon has signed power purchase agreements with nuclear plants across Virginia. Microsoft alone has $80 billion in Azure orders it cannot fulfill because it cannot procure electricity fast enough. The three largest cloud providers are, simultaneously, becoming energy utilities.</p><p>This is not a coincidence. It is a signal. The binding constraint on AI infrastructure is no longer silicon. It is electricity. And electricity, unlike GPUs, cannot be manufactured faster. It is procured, permitted, and built over years — over decades, in the case of nuclear. The enterprises that understand the power constraint understand why the hyperscalers’ cost structure is not as predictable as their current pricing suggests — and why owning your own hardware, on your own power arrangement, is a hedge against a variable that no contract can fully control.</p><h3>V. Elon’s Gambit: What a Converted Factory in Memphis Tells You About the Future</h3><p>There’s a way to understand a market that has nothing to do with reading analyst reports. You watch what the most aggressive, least consensus-dependent actor does when nobody is telling them they have to. What they build when there’s no board to approve it and no quarterly call to explain it on. That actor, in AI infrastructure, is Elon Musk. And what he built tells you something that a hundred earnings calls won’t.</p><p>xAI’s Colossus supercomputer — housed in a converted Electrolux factory in South Memphis — went from concept to operational in 122 days. The average data center takes four years. Musk’s team did it in four months. As of mid-2025, it runs over 200,000 GPUs across three generations of Nvidia hardware, making it arguably the largest AI training platform on earth. The target is 1 million GPUs.</p><p>But the GPUs are not the interesting part. The interesting part is the power.</p><p>When Colossus’s grid connection was only 8 megawatts — laughably insufficient for a facility of that scale — xAI didn’t wait for the utility. They deployed over 35 natural gas turbines, generating 422 megawatts of their own electricity. When the permitting process in Tennessee moved at the speed of, well, a permitting process in Tennessee, Musk shipped an entire natural gas power plant from overseas rather than wait for domestic approvals. You know how when you order a pizza and it’s taking too long, some part of your brain thinks “I could just make one myself”? Now imagine that, except the pizza is a two-gigawatt combined-cycle power plant and “making it yourself” means shipping it across an ocean on a cargo vessel. That is the level of institutional impatience we are talking about. Most people can’t get a permit to add a deck to their house in 122 days. This man built a supercomputer.</p><p><strong>COLOSSUS BY THE NUMBERS:</strong> 200,000+ GPUs operational (Q2 2025) | ~400 megawatts of power consumption | 35+ natural gas turbines (including unpermitted) | 88-acre solar farm planned (30 MW output — 10% of current load) | 100 MW additional solar + 100 MW battery storage announced | A 2-gigawatt natural gas power plant being shipped from overseas | Target: 1 million GPUs = ~2 gigawatts total capacity</p><p>Now, I want to be precise about something, because precision matters when you’re drawing lessons from a case study this unusual. Colossus is not a template for a regulated enterprise. A bank CTO who deploys unpermitted gas turbines in a residential neighborhood will answer to the OCC, not just the local planning board. The environmental costs are real and documented — the Southern Environmental Law Center found a 79% increase in nitrogen dioxide levels in surrounding neighborhoods. The community opposition is legitimate and growing. This is not a playbook to replicate. It is a signal to read.</p><p>The strategy, stripped of its controversies, demonstrates the most complete version of vertical integration applied to compute that exists in the world today: own the model, own the data center, own the power generation. xAI is not renting compute from Amazon and hoping AWS’s priorities align with Grok’s training schedule. It controls the silicon, the electricity, and the model. That’s a fundamentally different cost structure than any enterprise currently operating on hyperscaler infrastructure.</p><blockquote>The first-principles insight from Memphis is not about rule-breaking. It’s simpler than that. In the age of AI, the data center IS the product. The building, the power, the cooling, the silicon — that’s not overhead. That’s the moat.</blockquote><p>Think about what “sovereign compute” means for every enterprise watching from the outside. It means controlling your own power, your own silicon, your own model, your own cost structure. Answering to no hyperscaler on pricing, no utility on capacity, no vendor on upgrade cycles. The world is dividing — faster than most strategists have priced in — between companies that beg for compute and companies that command it.</p><p>The risks to the xAI approach are real and worth naming honestly. The solar-plus-storage strategy has merit but faces genuine constraints: 88 acres of solar in Memphis produces approximately 30 megawatts, about 10% of Colossus’s current draw. Even at 10x scale, solar is non-dispatchable without massive battery storage. The regulatory overhang from operating unpermitted turbines creates political risk that could force operational curtailment at the worst possible moment. And the capital intensity creates model risk — if the foundation model race produces a commodity outcome (which DeepSeek’s emergence in early 2025 demonstrated is possible), xAI has built an enormous fixed-cost base chasing revenue that may compress dramatically.</p><p>But the reaction function is predictable and already underway: every hyperscaler is now acquiring power generation assets. Google bought Intersect Power. Microsoft has nuclear deals. Amazon signed PPAs with nuclear plants. The AI compute race has become an energy race, and the energy race is forcing everyone to think like a utility — which is to say, to think about owning infrastructure rather than renting it.</p><p>That logic, once internalized by the hyperscalers, inevitably reaches the enterprise level too. The question Colossus poses isn’t whether you should build your own. It’s this: if the gap between frontier AI compute and enterprise AI compute is widening by the quarter, when exactly do you decide to close it?</p><h3>VI. The Case for Steel: Why On-Prem AI Wins</h3><p>The argument for enterprise on-premise AI infrastructure rests on four legs, each independently sufficient to justify the investment, and together forming a thesis the market has not yet fully priced.</p><p>Before the legs: one objection worth naming upfront. Running GPU infrastructure at enterprise scale requires MLOps engineers who understand CUDA, model serving, hardware failure modes, and the particular pathologies of distributed inference. These people are scarce, expensive — median salary above $162,000 — and actively recruited by the same hyperscalers you’d be competing with. The talent gap is real. It has quietly killed more on-prem AI projects than bad economics ever has. But the market’s response is already underway: FinOps discipline around AI infrastructure has become one of the top enterprise technology priorities for 2026, according to ETR’s survey of 1,357 technology leaders, precisely because runaway cloud bills forced organizations to build the capability to manage and optimize AI compute. Pre-validated infrastructure stacks that deploy in hours rather than months have shifted the operational burden from “build a custom GPU cluster from scratch” to “operate a validated system with vendor support.” Most large enterprises already run real-time trading systems, industrial process controls, and hospital imaging networks with specialized staff who didn’t exist as a job category twenty years ago. Private AI infrastructure is entering that same category.</p><h3>Leg One: The Rent Is Too Damn High</h3><p>Understanding any machine — financial or mechanical — requires looking past the surface and into the mechanism. The mechanism at work in enterprise cloud AI spending is straightforward, and once you see it, you cannot unsee it.</p><p>Cloud infrastructure has a pricing model that looks rational at low utilization and slowly, quietly, becomes irrational at high utilization. When you are running experiments — spinning up a workload, testing a hypothesis, shutting it down — renting compute is obviously correct. You pay for what you use and nothing more. This is the mental model that most enterprise buyers formed in 2021 and 2022, and it is the mental model they are still operating on in 2026. It was correct then. It is increasingly wrong now.</p><p>The shift happened when AI moved from the lab to production. A hospital running diagnostic AI on every incoming patient scan is not experimenting. It is running a continuous, predictable, high-volume workload that executes the same operations, at roughly the same scale, every single day. The cloud bill for that workload does not go down. It is linear with volume, forever, growing with the institution, paying the hyperscaler’s margin in perpetuity. The moment that’s true, the rent-versus-own math inverts — and the inversion is not subtle.</p><p>Here is the mechanism in concrete terms. An 8-GPU H100 cluster costs roughly $250,000 in hardware, amortized over three years at about $7,000 per month. Add electricity at $0.08/kWh for a cluster pulling 5 to 6 kilowatts per GPU: you’re looking at $25,000 to $35,000 per month fully loaded. The equivalent AWS instance runs $32 to $98 per hour on demand. For a single 8-GPU node, that’s $23,000 to $71,000 per month before data transfer fees. At sustained utilization, you are paying two to three times as much to rent what you could own. McKinsey’s infrastructure practice has confirmed: cloud-based AI costs 2 to 3 times more than equivalent on-prem hardware at high capacity.</p><p>The catch — and it is a real one — is utilization. Fujitsu’s research found that over 75% of organizations run GPU utilization below 70% at peak load. The economics of ownership work when you push past that threshold. The enterprises that can — a hospital processing millions of scans, an insurer routing every claim through AI triage, a bank screening every wire — hit 70% utilization without trying. For them, every additional month on cloud is a month of unnecessary margin donated to AWS. The enterprises that cannot are better served renting for now. The insight is not “own everything.” It is: know which workloads belong where, and be honest about which category you are actually in.</p><p><strong>THE HYBRID HEURISTIC:</strong> High-volume, predictable, data-sensitive workloads — own the steel. The cost compounds in your favor and the IP stays in the building. Bursty, experimental, or frontier-model-dependent workloads — rent from the hyperscaler. You cannot justify owned hardware for use cases you are still figuring out, and you cannot afford to be a model version behind when the frontier matters. Real-time, latency-critical workloads — on-prem by definition; no hyperscaler data center is close enough. The workloads that generate the most durable competitive advantage almost all sit in column one.</p><h3>Leg Two: Your Data Is Already Leaving the Building</h3><p>When Goldman Sachs runs AI analysis on a client’s M&amp;A target, the pattern of the query itself is information. It tells someone who is looking — a vendor employee, a nation-state adversary, a future competitor — what Goldman is working on. When JPMorgan’s AI evaluates loan applications, the aggregate patterns of what gets rejected, at what tolerances, constitute proprietary risk intelligence built over decades. When an insurer’s AI underwrites a policy, the combination of variables it weighs is the product of a century of actuarial investment. In each case, the query is the secret. Not the answer. The question.</p><p>Mastercard understood this before most companies were even asking the question. Its Decision Intelligence Pro system — a proprietary recurrent neural network built entirely in-house — is trained on 125 billion transactions annually and scores each one in under 50 milliseconds. The model improves with every transaction that passes through it. Mastercard reports fraud detection improvements of up to 300% over baseline methods. That performance gap is not the product of a better algorithm. It is the product of 50 years of proprietary transaction data that no competitor can replicate, running on infrastructure that Mastercard controls absolutely. No API call to OpenAI is in that loop. No AWS data center holds those weights. The model, the data, and the compute are owned. The result is a competitive position that compounds with every passing millisecond.</p><p>Every one of these use cases has the same problem with shared LLMs: the data doesn’t just leave the building — it potentially makes the next version of the model smarter for everyone. The public model trains on de-identified versions of your data, but de-identification is not anonymization, and no amount of contractual language fully closes the loop.</p><p><strong>THE 2025 WAKE-UP CALL:</strong> Chinese state-sponsored actors used jailbroken AI to execute autonomous cyberattacks against 30 organizations — with 80 to 90% of the attack performed by the AI itself. The adversary didn’t steal your data. They used your vendor’s model, running on your vendor’s infrastructure, to find the door. The only complete defense is an AI system that runs on hardware you control. Policy doesn’t fix this. Architecture does.</p><h3>Leg Three: The Compound Interest of Intelligence</h3><p>Here’s the dimension of this thesis that gets the least attention and matters the most.</p><p>If you look for truth, you may find comfort in the end. If you look for comfort, you will not get either comfort or truth — only soft soap and wishful thinking to begin with, and in the end, despair. The enterprise looking for comfort — the one that rents its AI from a vendor because it’s easy, because the procurement process is familiar, because nobody ever got fired for buying AWS — will find that the comfort evaporates the moment the competitive gap becomes visible. The enterprise looking for truth — the one willing to do the harder thing, to own the infrastructure, to build the feedback loop, to accept the balance sheet complexity — finds that the truth compounds in its favor. Not immediately. Not dramatically. But relentlessly.</p><p>Consider what Mastercard actually built with Decision Intelligence Pro. Not a fraud detection tool — a feedback loop. Every transaction scored by the model generates a data point that makes the next score more accurate. The model trained on this year’s 125 billion transactions is better than last year’s model. The gap between Mastercard’s fraud detection and a bank using a generic off-the-shelf system widens every day — not because Mastercard has better engineers, but because Mastercard owns the compute that processes the data and therefore owns the improvement.</p><p>This is a categorically different kind of competitive asset than anything enterprises have owned before. It doesn’t depreciate. It compounds. And the compounding is directional — it only works for the entity that controls the hardware, because the hardware is where the feedback loop lives. Send your data to someone else’s infrastructure and the loop runs for them. Keep it in your building and the loop runs for you. There is no third option. There is no “virtual” version of ownership that captures the same dynamic. Oh, there’s a virtual private cloud. Wonderful. You know what else is virtual? Virtual leather. Virtual reality. Virtual assistants that can’t actually assist you with anything. The word “virtual” has always been a polite way of saying “not quite the real thing, but we’re hoping you won’t notice.”</p><p>The open-weight model revolution has made this argument dramatically more accessible — and this is the most significant change to the on-prem economics that most strategic analyses miss entirely. DeepSeek V3, built by a Chinese quant fund for approximately $6 million and released open-source in early 2025, performs comparably to GPT-4 on most enterprise benchmarks. Llama 4, Meta’s open-weight model, is free. Mistral has struck enterprise deals including a multi-year arrangement with HSBC to deploy its models bank-wide. The model layer is commoditizing. Not for frontier reasoning tasks — those still require the labs at the cutting edge. But for the high-volume, domain-specific inference tasks where the feedback loop applies most powerfully — claim triage, wire screening, credit scoring, document extraction — open-weight models fine-tuned on proprietary data are now within 10 to 15% of frontier performance at essentially zero marginal model cost.</p><p>This changes the calculus fundamentally. The enterprise that fine-tunes Llama 4 on 80 years of its own claims data doesn’t need GPT-5. It needs its own hardware, its own weights, and its own feedback loop. The hardware costs $250,000. The model is free. The data already exists. The competitive moat is the combination — and it’s a moat that gets deeper every quarter the model runs in production.</p><blockquote>Renting a model from a hyperscaler is like renting a high-frequency trading algorithm. You get the trades. But the market maker gets the data on how you trade. You are teaching your counterparty how to beat you.</blockquote><p>Every new power you gain through shared infrastructure is simultaneously a power the infrastructure owner gains over you. Each new model, each new feature, each deeper integration leaves the enterprise both stronger in capability and weaker in autonomy. The only way to break that symmetry is to own the substrate. Everything else is a more comfortable version of dependency. And comfort, as we’ve established, is not a strategy. It’s the absence of one.</p><blockquote>When the model layer commoditizes, the data layer becomes everything. And when the data layer becomes everything, the compute layer — which processes the data and generates the feedback — becomes the infrastructure that owns the moat. You cannot build that asset on someone else’s hardware. You cannot outsource the compounding.</blockquote><p><strong>The VPC Objection — Addressed Directly</strong></p><p>The hyperscalers know this argument. Azure OpenAI now offers models in dedicated virtual networks with contractual guarantees your data never leaves your environment. Amazon Bedrock Custom Models lets enterprises fine-tune foundation models on proprietary data and retain the weights. These are real improvements over the 2023 landscape.</p><p>Here is the honest answer: private VPC deployments solve the data transit problem but not the feedback loop ownership problem. When your fine-tuned model runs on AWS hardware inside a virtual private cloud, the weights are yours — contractually. The compute is not. The feedback loop runs on Amazon’s silicon, synchronized to Amazon’s infrastructure, dependent on Amazon’s pricing decisions and capacity allocation. In the event of a pricing dispute, a service disruption, or a strategic shift in Amazon’s enterprise AI positioning, your moat is hostage to a landlord. The company that owns its inference hardware owns the feedback loop absolutely. That distinction may seem theoretical until the moment it isn’t — and in regulated industries where the AI model is the product, it’s a distinction worth the capital to resolve.</p><h3>Leg Four: Physics Doesn’t Care About Your SLA</h3><p>The most transformative enterprise AI applications — real-time clinical decision support, millisecond fraud detection, supply chain dynamic pricing, live underwriting — require latency that public cloud cannot reliably deliver.</p><p>The round-trip from a branch bank to an AWS data center and back is 30–100 milliseconds on a good day. Fine for a chatbot. Unacceptable for a fraud detection system that needs to evaluate a transaction before it clears. Edge AI — inference on owned hardware, co-located with the data source — is the only architecture that delivers sub-10ms responses for mission-critical applications.</p><p>Beyond latency, data residency. European GDPR, US state privacy laws, China’s Cybersecurity Law, and a growing thicket of sector-specific regulation increasingly mandate that certain data cannot leave specified geographic boundaries. A hospital cannot process patient data on servers physically located in another jurisdiction. A European bank cannot process customer data on servers where a US court order might compel disclosure. In November 2025, EU regulators formally designated AWS, Google, and Microsoft as critical third-party providers for the financial sector — a classification that significantly increases regulatory pressure to maintain exit plans and on-prem alternatives. Sovereign AI infrastructure — data centers owned or controlled by the company, within defined legal boundaries — is becoming table stakes.</p><h3>VII. The 10-Year Map: How This Actually Plays Out</h3><p>We all want progress. But progress means getting closer to where you actually want to be. And if you’ve taken a wrong turn, progress means doing the unfashionable thing — stopping, turning around, and walking back to the right road. In that case, the person who turns back soonest is the most progressive of all. The enterprise AI market is on a road. The road leads to increasing dependency on hyperscaler infrastructure. The road is well-paved, well-lit, and clearly marked. It is also, for a growing number of enterprises, the wrong road.</p><p>The honest prediction is that this transition plays out slower than the bulls want and faster than most institutions are prepared for. Structural shifts in enterprise infrastructure have never moved at the pace of a product cycle — they move at the pace of balance sheets, procurement cycles, and the moment a CFO has to explain a line item that has doubled twice in three years. That moment is arriving now.</p><p>For the next two years — through roughly 2028 — the hyperscalers continue to dominate enterprise AI. Token costs for commodity inference keep falling as competition intensifies and open-weight models improve. Most enterprises stay on cloud because the capital costs of ownership still feel unnecessary against a declining marginal price of renting. The CIOs who are already beginning to shift — the 30% that Futurum’s survey found actively moving workloads between public and private cloud — are early. They are not wrong.</p><p>The inflection comes when the first major enterprise AI IP litigation lands. It will happen in a regulated industry — insurance, finance, or healthcare — and it will involve a data exposure through a third-party AI vendor that creates material liability the enterprise cannot easily quantify or defend. When that case settles, every general counsel in every regulated industry will read the opinion and call their CTO. The EU’s formal designation of AWS, Google, and Microsoft as critical third-party providers has already begun building the regulatory scaffold for that conversation.</p><p>By 2030, the GPU server is the new server rack — something every enterprise above a certain scale owns, operates, and treats as core infrastructure rather than a vendor relationship. The hyperscalers do not disappear. They serve smaller enterprises, provide training capacity for frontier models, and absorb the bursty experimental workloads that never justify ownership economics. But the core inference layer — the high-volume, proprietary, mission-critical workloads that generate the feedback loops — migrates to owned silicon. Not because enterprises became ideologically committed to ownership. Because the compounding math makes the alternative irrational, and finance eventually enforces the math.</p><p>The energy picture only accelerates this. The hyperscalers’ $700 billion in 2026 capex is already generating pushback from investors watching free cash flow crater. Power procurement is the binding constraint — Microsoft alone has $80 billion in Azure orders it cannot fulfill because it cannot get electricity fast enough. The enterprises that build their own compute on co-located power arrangements will find by 2030 that they have insulated themselves from a cost variable the hyperscalers were never fully able to control. The datacenter becomes the new HVAC system — infrastructure every serious facility operates, not a service they purchase.</p><blockquote>The enterprises that own their AI infrastructure in 2030 won’t just have cheaper compute. They’ll have smarter models, deeper moats, and institutional knowledge their competitors cannot acquire — because it was generated on infrastructure their competitors didn’t own.</blockquote><h3>VIII. Where the Capital Should Go</h3><p>The opening of this piece was a usage limit on an AI tool — a small friction point. But the most important structural shifts tend to announce themselves as friction before they announce themselves as opportunity. The compute constraint on a single laptop is the edge of a demand curve that has no visible ceiling. Every enterprise in every regulated industry is about to run into the same wall, at larger scale, with larger consequences.</p><p>The thesis is not that the cloud is dying. The cloud is not dying. The thesis is that the cloud’s role in enterprise AI is about to be redefined by the same forces that have always redefined infrastructure economics: scale, security, and the compounding value of ownership. When a line item doubles twice, finance gets involved. When a breach exposes what the data flows were carrying, legal gets involved. When a competitor’s model outperforms yours on the same task because they’ve been running a feedback loop for three years and you haven’t, strategy gets involved. All three of those conversations are coming. The enterprises — and the investors — who have already moved will be in a very different position when they arrive.</p><p>Mastercard built an AI moat that compounds with every transaction it processes. Not because it had a visionary AI strategy in 2010. Because it had 50 years of proprietary transaction data, owned the infrastructure that processed it, and ran the loop long enough for the compounding to become unreachable. The gap between Mastercard’s fraud detection and every bank using a generic third-party model is not a technology gap. It is a time-times-data-times-compute gap — and the only way any competitor closes it is by starting their own clock, on their own infrastructure, today.</p><p>That dynamic is now available to every enterprise with sufficient proprietary data and sufficient inference volume to justify ownership. The model is free — Llama 4 is free. The hardware is available. The operational tooling has matured. The only thing that’s scarce is the decision to move, made early enough for the compounding to matter.</p><p>Courage is not simply one of the virtues. It’s the form every virtue takes at the testing point. Prudence, conviction, strategic clarity — they are all just words on a whiteboard until the moment you have to actually do something that costs real money and makes the comfortable people in the room uncomfortable. The enterprise AI infrastructure decision is approaching that moment. The comfortable choice is to keep renting. The courageous choice is to own. And in an economy where the feedback loop is the moat, where data compounds only on hardware you control, and where the gap between the leaders and the followers widens every quarter — the courageous choice is also, as it so often turns out to be, the correct one.</p><blockquote>The compute layer commoditizes. It always does. What doesn’t commoditize is the feedback loop running on top of it — the proprietary model, trained on proprietary data, improving on proprietary infrastructure, compounding in a direction no competitor can see until it’s too late to catch. That is the asset. Own the infrastructure that generates it.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=256781c15206" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The History Trap]]></title>
            <link>https://medium.com/@mccoype_64598/the-history-trap-1a9e4c8cec97?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/1a9e4c8cec97</guid>
            <category><![CDATA[private-equity]]></category>
            <category><![CDATA[investing]]></category>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Fri, 13 Feb 2026 15:28:50 GMT</pubDate>
            <atom:updated>2026-02-13T15:28:50.640Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4U0sRzUNWMBcCI2V-kO6qw.png" /></figure><h3>Why Most Private Equity Returns Are a Fraud of Timing, and What Druckenmiller Says You Should Do Instead</h3><p>It was Q3 2021. SOFR was still basically zero and every fund in America was tripping over itself to deploy. The CIM landed on a Thursday — 200 pages, professionally typeset, the kind of document that costs $40,000 to produce and exists primarily to make the buyer feel like the process is serious. By Monday morning the lender had already indicated. By Wednesday the deal team had built the model.</p><p>The company had 112% net revenue retention. Gross margins in the mid-70s. A customer logo slide that looked like a Fortune 500 yearbook. Revenue had compounded at 28% for three years running. The management team was polished, the churn was low, and every cohort chart bent in the right direction.</p><p>We had no thesis on the space. None. We didn’t know the product category, didn’t understand the competitive dynamics, couldn’t have told you who the next three customers would be or why. But the numbers were beautiful. And in a zero-rate world, beautiful numbers felt like enough.</p><p>Everyone in the room nodded. The junior associate had cross-referenced the customer cohorts. The operating partner had “back-channeled” with two former executives. The Patagonia vests were crisp. The formatting was immaculate.</p><p>And not a single person in that room was investing. We were performing an autopsy on a living patient and calling it a diagnosis.</p><h3>The 18-Month Ghost</h3><p>Stanley Druckenmiller — who compounded capital at 30% annually for three decades and never had a down year — has a thesis so simple it should embarrass every investment committee in the country. He put it bluntly: <strong>“Never invest in the present.”</strong></p><p>The market is not a snapshot. It’s a séance. It is a visualization of the world 18 months from now. The price you pay today is a bet on a future that hasn’t happened yet. Earnings last quarter? Already in the stock. The trailing twelve months of EBITDA you underwrote in that CIM? The market digested it before your associate finished formatting the waterfall.</p><p>The implication is brutal: most investors are diligencing the wrong thing. They treat historical performance as a legal contract with the future — as if a 30% growth rate is a binding covenant, as if 90th-percentile NRR is a constitutional right. They aren’t making investments. They are filing audits. They are historians holding themselves out as prophets.</p><p>Druckenmiller’s point isn’t that the past doesn’t matter. It’s that the past is an <em>input</em>, not an <em>answer</em>. The answer lives in the setup — in the structural conditions that will determine whether the next 18 months look anything like the last 18. And almost nobody does that work. We certainly didn’t, sitting in that room in 2021, mesmerized by a spreadsheet for a market we couldn’t even define.</p><h3>The 15x Prophecy: The Performance Fraud</h3><p>Let’s be honest about what “private equity returns” actually are, because the industry has gotten away with one of the great attribution frauds in the history of capital allocation.</p><p>The directional truth is damning, and undisputed by anyone honest enough to run the math: over the last decade, roughly two-thirds of private equity returns were generated by multiple expansion and leverage. Not better operations. Not smarter go-to-market. Not technological transformation of stagnant businesses. Just: buy it cheap, lever it to the gills, sell it at a higher multiple, and write a case study about “operational value creation” that wouldn’t survive a first-year accounting exam.</p><p>Here’s what that actually looks like in practice. Take a representative mid-market deal. Acquired at 10x EBITDA on $20M, so $200M of enterprise value. 60% leverage at SOFR plus 250. Five-year hold. Exit at 13x on $23M of EBITDA — that $3M of improvement coming mostly from price increases and one tuck-in acquisition, not organic growth. The equity check triples. The deck says “3x MOIC, 25% gross IRR.”</p><p>Now run the attribution. Multiple expansion from 10x to 13x explains roughly 60% of that return. Leverage did another 25%. The actual operational improvement — the thing the GP is charging you two-and-twenty to deliver — generated maybe 15 cents of every dollar you made. And that’s the <em>base case</em>. In the downside, the leverage that manufactured the return is exactly what kills you.</p><p>This is what Druckenmiller means when he says you can’t make money agreeing with the market. Trailing EBITDA is the ultimate consensus data point — everyone in the process has it, everyone believes it, and therefore it contains zero information about the future. The return attribution above isn’t a failure of execution. It’s the predictable outcome of an entire industry investing on consensus and then calling the result “alpha.”</p><p>Now consider this: you can buy Microsoft — <em>Microsoft</em>, the company that owns the operating system of the planet, the cloud infrastructure of the enterprise, the AI co-pilot embedded in every knowledge worker’s daily life — at roughly 30x forward earnings. The last three mid-market HVAC platforms I’ve seen marketed were priced between 14x and 18x trailing EBITDA, which on an apples-to-apples basis, adjusted for growth differential and capital intensity, implies a <em>higher</em> effective multiple than the most dominant technology franchise in human history.</p><p>A collection of furnace-repair trucks, a CRM held together with duct tape and an owner’s nephew who “handles IT,” a business model predicated on the continued failure of residential heating equipment in the Ohio River Valley — this is priced at a premium to Microsoft.</p><p>Microsoft is amortizing the largest R&amp;D budget in human history into an asset that compounds — every dollar spent on AI infrastructure makes the next dollar more productive. The HVAC rollup is accumulating technical debt in the form of aging fleets, dispatch software that predates the iPhone, and deferred maintenance that shows up as a surprise capex line in Year 3 of the hold. The PE firm underwrites the latter as “lower risk” because the revenue is recurring. They are confusing the predictability of demand with the durability of the asset. These are not the same thing.</p><p>Charlie Munger would call this what it is: <em>standard insanity</em>. The entire mid-market PE apparatus is not in the business of building value. It is in the business of surfing a wave of cheap debt, harvesting management fees, and praying — with the quiet desperation of a man at a craps table — that the macro winds don’t change.</p><p>The winds are changing.</p><h3>The Baseline: Diligence as a Means, Not an End</h3><p>None of this is an argument against diligence. It’s an argument against diligence as a <em>destination</em>.</p><p>I understand the appeal. I’ve sat in those rooms and felt the gravitational pull of a clean dataset. The numbers are seductive precisely because they’re knowable. A historical cohort doesn’t argue with you. A trailing revenue figure doesn’t require a point of view. The future does — and that’s exactly why most institutional frameworks are designed to avoid it. Ambiguity doesn’t fit in a sensitivity table.</p><p>But historical analysis, done right, is vital. It is the baseline from which you understand the <strong>market setup</strong> — a term Druckenmiller uses constantly. You audit the past not to extrapolate it forward in a straight line, but to find the friction. Where is capital stuck? Where are incumbents paralyzed by their own balance sheets? Where is the customer screaming for a solution that the existing supplier is too indebted, too complacent, or too technologically illiterate to deliver?</p><p>Real diligence isn’t confirming that the patient was healthy last year. Real diligence is identifying the gap between <em>what is</em> and <em>what is structurally possible</em> given a genuine re-engineering of the business. It’s the difference between reading an EKG and performing surgery.</p><p>The spreadsheet jockeys get the EKG right every time. Perfectly sensitized IRR tables that show a 25% return in the base case and a 19% return in the downside that will never, ever actually be the downside.</p><p>But they can’t operate. And they can’t see forward. Because to see forward, you need a thesis — not a spreadsheet. You need to have spent years understanding a market before the CIM ever lands. That’s what we got wrong in 2021. We had the discipline to diligence the past. We didn’t have the conviction, or the context, to underwrite the future.</p><h3>The Platform Gambit: Executing the Druckenmiller Thesis</h3><p>If the Druckenmiller thesis is the philosophy, the next question is mechanical: how do you actually build a portfolio around the future rather than the past?</p><p>Two modern operators have given us distinct versions of the playbook. They don’t get enough credit for it, probably because their approaches lack the theatrical complexity that the finance industry uses to justify its fees.</p><h3>The Operational Wedge: The Brad Jacobs Hand</h3><p>Brad Jacobs has built five multi-billion-dollar companies across logistics, equipment rental, packaging, and building products. His strategy is almost offensively simple. He targets huge, fragmented, “ugly” industries — markets where the word “technology” still refers to a fax machine in the back office.</p><p>The insight is structural. In these industries, the PE-owned platforms are so debt-laden they can’t invest in anything beyond the minimum required to service the note. And the family-owned operators — the ones who built the business in 1974 and have been milking cash ever since — have no incentive to modernize. They are running lifestyle businesses with $40 million in revenue and $8 million in owner compensation.</p><p>In Druckenmiller’s language, this is a <em>setup</em> — a structural condition where capital is stuck, incumbents are paralyzed, and the gap between what is and what is possible is wide enough to drive a platform through.</p><p>But Jacobs doesn’t walk into these markets with a pitch deck about “digital transformation.” He walks in with a centralization playbook that most family owners would find physically uncomfortable. Procurement gets consolidated in the first 90 days. Redundant facilities close in the first year. Every branch manager who was used to running their own fiefdom now reports into a standardized operating cadence with weekly KPI reviews and monthly performance rankings.</p><p>The technology — fleet management, dynamic pricing, inventory optimization — comes <em>after</em> the organizational discipline is in place. The tech amplifies the operating system. It doesn’t replace the need for one.</p><p>Here’s the thing most people miss about this model: Jacobs isn’t a manager. He’s a capital liberator. Every fragmented industry is sitting on a reservoir of trapped capital — cash buried in excess inventory, redundant overhead, and the operational slack that accumulates over decades of comfortable ownership. A distributor with 90 days of inventory on hand and three overlapping warehouses within a 50-mile radius isn’t just inefficient. It’s a vault full of imprisoned equity that the owner never knew was there because he never had the operating system to see it.</p><p>Jacobs doesn’t inject capital into these businesses. He frees the capital that was already there. And this is where the physics of his model diverge completely from traditional PE: the payoff isn’t a markup on paper. It’s a <strong>working capital arbitrage</strong>. Higher inventory turns mean less cash trapped in the system. Less cash trapped means more free cash flow. More free cash flow means you can self-fund the next acquisition with cash the first one liberated — recycling your initial equity through velocity rather than borrowing against it through leverage.</p><p>By the time a “history trap” investor finishes their 100-day integration plan, Jacobs has already recycled his equity and is compounding off operations. He’s not playing a better version of the same game. He’s playing a different game entirely — one where the return engine is cash conversion, not financial engineering, and where each acquisition makes the platform structurally harder to compete against rather than just arithmetically larger.</p><p>The risk/reward asymmetry is pure Druckenmiller: the downside is knowable — the business continues operating roughly as it did before you arrived — and the upside is structural, driven by a re-engineering of the cash conversion cycle that leveraged competitors literally cannot replicate without dismantling their own capital structures first.</p><p>This is what operational dominance actually looks like in practice. Not a PowerPoint about “synergies.” A machine that converts fragmentation into velocity and velocity into monopolistic market position.</p><h3>The Surface Area Play: The Josh Kushner Hand</h3><p>Josh Kushner, through Thrive Capital, has built a career around a different but equally Druckenmiller-aligned bet: identifying markets where the <strong>surface area is naturally expanding</strong>, and then owning the infrastructure layer of that expansion.</p><p>The key concept here isn’t surface area alone — it’s the <strong>default choice</strong>. In the history trap, investors look at a company’s current customer count and extrapolate. In the surface area play, you ask a fundamentally different question: <em>When the next thousand companies enter this market, who do they choose without thinking?</em></p><p>That’s the moat. Not switching costs, not brand, not even technology in isolation. It’s the cognitive and structural path of least resistance. Stripe captured the explosion of internet commerce not by selling to each new online business, but by being the infrastructure that each new business arrived pre-sold on. Developers chose Stripe by default because it was the easiest, most obvious path. Every new entrant to the market expanded Stripe’s revenue at near-zero marginal acquisition cost.</p><p>The purest current expression of this is Thrive’s position in OpenAI. Kushner didn’t bet on a chatbot. He bet that foundational AI infrastructure would become the next compute layer — that every application built for the next decade would require an inference engine the way every website needed a server in 2005. Every developer building an AI-native app defaults to the OpenAI API the way every web developer defaulted to AWS in 2010. The thesis isn’t about one company’s product. It’s about owning a tax on the future growth of an entire technological paradigm. As the surface area of AI-native applications expands, OpenAI’s position as the default choice means revenue scales geometrically with the category, not linearly with its own sales effort.</p><p>But here’s what makes Kushner credible rather than just lucky: the framework’s limitations are visible in his own portfolio, and the distinction is instructive. Oscar Health was built on a structurally similar thesis — the ACA was expanding the surface area of individual health insurance, and a digital-first platform could own the infrastructure layer of that expansion. Thrive made money on Oscar; it trades well above where they invested. But Oscar didn’t compound the way Stripe or OpenAI compounds, and the reason is precisely the default-choice dynamic. Health insurance has no path of least resistance. Every new enrollee comparison-shops on the exchange. There is no moment where a customer arrives pre-sold. Oscar had to <em>fight</em> for every marginal customer at roughly the same acquisition cost as the first one. The surface area of the market expanded, but Oscar couldn’t convert that expansion into geometric scaling because the default-choice mechanism didn’t exist.</p><p>The lesson is precise and valuable: surface area expansion only compounds into infrastructure-layer returns if the default-choice dynamic is present — if the marginal customer chooses you by structural gravity rather than by persuasion. When that dynamic exists, you own a toll bridge on the future. When it doesn’t, you’re in a growing market with no structural advantage, which is a much more ordinary — and much more fragile — bet.</p><h3>The Synthesis</h3><p>Both approaches — the Operational Wedge and the Surface Area Play — are expressions of the same Druckenmiller principle. They are forward-looking bets on structural conditions, not backward-looking bets on historical performance.</p><p>Jacobs isn’t buying EBITDA. He’s buying the right to liberate trapped capital and compound off the velocity. Kushner isn’t buying growth. He’s buying the default-choice position in categories where the surface area is expanding geometrically.</p><p>Neither of them is sitting in a room on the 34th floor, staring at a trailing-twelve-month number and calling it a strategy. Because it isn’t a strategy. It’s just getting lucky. And luck, as Druckenmiller would tell you, is not a repeatable process.</p><p>But if the math of the future is so compelling, why is the industry still so obsessed with the past?</p><h3>The Fee Trap Inside the History Trap</h3><p>Charlie Munger considered <strong>incentive-caused bias</strong> the single most reliable predictor of institutional stupidity. He didn’t mean people are corrupt. He meant something worse: they are rational actors inside an irrational structure, and they will optimize for whatever the structure rewards — even when it leads them off a cliff.</p><p>Start at the bottom of the food chain. The junior associate who built that model in Q3 2021 — the one with the immaculate cohort analysis and zero thesis on the market — she wasn’t lying to us. She wasn’t lazy. She was responding rationally to her incentive structure. Her bonus is tied to deals closed. Her promotion depends on deals closed. Her entire career trajectory is a function of volume, not accuracy. When the CIM arrives and the numbers are clean, she is structurally incapable of saying “we don’t have a thesis on this space, we should pass” — because passing doesn’t get her promoted, doesn’t get her to the next fund, doesn’t get her career where it needs to go. The rational move, for her, is to build the model, flag nothing, and let the partners decide.</p><p>One level up, the partners have their own version of the same disease. Management fees accrue on committed capital from day one. A $500 million fund generates $10 million a year in fees regardless of whether the GP has deployed a dollar or created a cent of value. The incentive is to deploy, not to wait for conviction. Sitting on dry powder while you develop a thesis on an unfamiliar market doesn’t generate fees. It generates uncomfortable LP calls asking why the fund is behind on deployment pace.</p><p>And carry? Carry crystallizes on paper marks that are, in many cases, self-assessed — the GP tells the LP what the portfolio is worth, and the LP’s auditor nods along because the methodology is “industry standard.” The entire two-and-twenty apparatus, from associate to managing partner to fund structure itself, is an incentive cascade that rewards backward-looking pattern recognition and punishes the forward-looking ambiguity that Druckenmiller says is where all the returns live.</p><p>Munger would have recognized the architecture immediately. Nobody in this system is committing fraud. Everybody in this system is being rational. And the aggregate output of all that individual rationality is an industry that deploys capital into businesses it doesn’t understand, on the basis of historical data it can’t contextualize, using leverage it can’t sustain, and then marks the result at valuations it can’t defend — all while collecting fees that accrue regardless of the outcome.</p><p>The result is an industry increasingly populated by what LPs are quietly starting to call <strong>zombie funds</strong> — vehicles that are technically “active” because they hold assets and report paper marks, but are functionally dead because they haven’t returned meaningful cash in years. A fund raised in 2019 or 2020 sitting at 1.8x TVPI and 0.15x DPI isn’t a performing investment. It’s a fiction maintained by quarterly valuation reports and the GP’s continued ability to collect management fees on the unreturned capital. The marks say the past was great. The cash register says nothing has actually happened.</p><p>These funds send polished quarterly letters with optimistic commentary about “value creation in progress” and “approaching monetization events.” The LPs read them. The allocators file them. And the DPI line doesn’t move.</p><p>The DPI reckoning that’s been building since 2022 isn’t a correction. It’s a return to the only question that ever mattered: did you actually make money, or did you just mark it? LPs are finally asking what they should have been asking all along: “Where is my cash?” Not my IRR. Not my TVPI. My <em>cash</em>. Distributions to paid-in capital — actual money returned divided by actual money invested — is the only metric that can’t be gamed by marks, timing, or leverage. And the numbers, for many post-2019 vintage funds, are ugly enough to change behavior. If your fund’s primary export is quarterly PDF updates rather than wire transfers, the market is going to reclassify you — not as an underperforming fund, but as a zombie that was never alive in the first place.</p><p>Druckenmiller would recognize this setup immediately. The era of free money that allowed the entire industry to generate returns through leverage and multiple expansion is over. The GPs who built their track records on that wave are about to discover what their returns look like without it. And the LPs who kept re-upping based on paper marks are about to discover what their portfolios are actually worth.</p><p>The history trap isn’t just an intellectual error. It’s a business model. And the business model is breaking.</p><h3>Autopsy vs. Operation</h3><p>Here is the question every investor should ask themselves, preferably before the next IC meeting:</p><p><em>Am I performing an autopsy, or an operation?</em></p><p>If your return profile relies on a trendline continuing, on multiples expanding, on a lender offering you one more turn of leverage at a spread that makes the model work — you are a master of the autopsy. You can tell us, in exquisite detail, why the patient was healthy yesterday. You can produce the lab work, the imaging, the full medical history. It will be beautifully formatted and completely useless.</p><p>But you cannot save the patient on the table today. And you certainly can’t tell us what the patient will look like 18 months from now.</p><p>Real money — durable, compounding, defensible money — is made in the operating room. It is made by investors who understand the market setup, who see the structural friction, and who have the conviction and the operational toolkit to bend the curve. It is made by people who had a thesis <em>before</em> the CIM arrived — who spent years in a category learning its physics, so that when the opportunity presented itself, the diligence was confirmation of a view they already held, not a substitute for one they never developed.</p><p>The history trap isn’t a failure of intelligence. The people in those rooms are brilliant. It’s a failure of structure — an industry that pays you to audit the past and charges you for the privilege of ignoring the future.</p><p>The question isn’t whether you can read the past. Everyone in this business can read the past. The question is whether you’ve built anything — a thesis, a platform, a competitive position — that earns the right to compound into the future.</p><p>Druckenmiller never invested in the present. He invested in the gap between what the market believed and what was structurally becoming true.</p><p>That gap is where all the returns live. The only question is whether you’re willing to do the work to see it.</p><p><em>The Druckenmiller interview referenced throughout this piece can be found </em><a href="https://www.youtube.com/results?search_query=druckenmiller+never+invest+in+the+present"><em>here</em></a><em>. Return attribution data reflects directional findings from Bain &amp; Company’s annual Global Private Equity Reports (2022–2024) and McKinsey’s Global Private Markets Reviews. Deal math is illustrative and composited from representative mid-market transactions.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1a9e4c8cec97" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Billion Dollar Cover Charge]]></title>
            <link>https://medium.com/@mccoype_64598/the-billion-dollar-cover-charge-a0cdca14c096?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0cdca14c096</guid>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[investing]]></category>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Mon, 09 Feb 2026 21:59:48 GMT</pubDate>
            <atom:updated>2026-02-09T21:59:48.828Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Why every GP in America is paying any price to get into the same ten companies — and what it tells us about a market based on consensus.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8wn-H0jDNtrrRlH8I3VFsg.png" /></figure><p>The family office CIO had been doing this long enough to know when he was being hustled. Twenty-three years in the business, a GSB MBA he never mentioned, and a Bloomberg terminal he still kept on his desk even though he hadn’t touched it in months. He’d survived the dot-com collapse, the GFC, and the 2022 rate shock without a single LP letter that used the word “impairment.” He was good at this. But here he was, in a conference room on Sand Hill Road in January 2026, being told by a twenty-nine-year-old in a Loro Piana quarter-zip that he could “get into” Anthropic at a $350 billion valuation — but only if he also committed $40 million to a side-car allocation in a Series B infrastructure company he’d never heard of. The Anthropic shares were table stakes. The side-car was the price of admission. The secondary broker called it “portfolio construction.” The CIO had another word for it.</p><p>The term of art in 2026 venture capital is “clearing price,” and the clearing price is one billion dollars. Not because companies are worth a billion dollars at inception. Not because the technology justifies it. Not because any rational discounted cash flow model on earth produces that number at the seed stage. The clearing price is one billion dollars because that is the minimum denomination of capital that the system can metabolize without choking.</p><p>But here’s what nobody talks about in polite company: the GPs writing these checks aren’t making investment decisions anymore. They are making <strong>AUM decisions</strong>. And the math behind those AUM decisions is the most important number in venture capital that nobody has bothered to write down.</p><h3>The Number That Explains Everything</h3><p>In 2025, the entire US venture capital industry raised <strong>$66 billion</strong> from limited partners. That is the total pool. Every LP commitment, every first close, every final close, every emerging manager scraping together $10 million from dentists in Connecticut. Sixty-six billion. It was the lowest annual commitment total since the late 2010s, down 35% from the prior year and roughly 70% below the 2022 peak.</p><p>Now consider the other side of the ledger. In that same year, just the top ten mega-rounds — OpenAI, Anthropic, xAI, Databricks, Anduril, and a handful of others — absorbed <strong>$84 billion</strong> in direct capital.</p><p>Read that again. The top ten <em>companies</em> consumed more capital than the entire VC <em>industry</em> raised from LPs. The demand side of the cap table dwarfs the supply side of the fundraise. This is not a normal market condition. This is a financial system eating itself.</p><p>And it creates an incentive structure so powerful, so perfectly misaligned, that it explains nearly everything happening in venture capital right now.</p><h3>The GP’s Real Calculus</h3><p>Imagine you are a general partner at a top-quartile venture firm. You manage $3 billion in AUM across several funds. Your current flagship is a $2 billion vehicle. Your management fee is 2%, so you collect $60 million a year to keep the lights on. Your carry on realized gains is where the real money lives, but carry requires exits, and exits require time. The fee stream is the oxygen.</p><p>Now consider what happens to your franchise in three different scenarios.</p><p><strong>Scenario A:</strong> You invest $100 million in an AI company at a $2 billion valuation. It goes sideways, then slowly to zero over five or six years — the way these things actually die, not with a bang but with a series of increasingly desperate bridge rounds and whispered markdowns. Your fund takes a 5% hit. Your LPs grumble. But by the time the write-off is official, you’ve already raised your next fund. The management fee keeps flowing. Life goes on.</p><p><strong>Scenario B:</strong> You <em>don’t</em> invest. You sit it out because the price is insane. The company becomes Anthropic. It raises $37 billion in total funding. Its valuation goes from $2 billion to $350 billion. Every other GP in America has it on their track record. Your LPs ask why you don’t. Your next fundraise stalls. Your AUM shrinks. The fee stream contracts. Your firm dies slowly.</p><p><strong>Scenario C:</strong> You invest. The company becomes Anthropic. Now you’re on the cap table of the most sought-after AI company in the world — and every LP in your rolodex wants exposure. You raise $10 billion in SPVs at 1% management fee and 15% carry, earmarked for follow-on rounds. That’s $100 million a year in SPV management fees alone — quickly surpassing the $60 million you collect on your main fund. Your main fund fees become the <em>side hustle</em>. Then comes the halo effect: you’re now the GP who “saw Anthropic early,” the maven picker of AI companies, and your next flagship fund is twice the size of your last one. One cap table entry didn’t just return capital. It created an entirely new business.</p><blockquote><em>The cost of being wrong on price is a 5% fund loss. The cost of being wrong on access is an existential threat to the firm. The reward for being right on access is a permanent restructuring of your economics.</em></blockquote><p>Run those three scenarios side by side and the answer is obvious. A is survivable. B is fatal. C is transformative. If you’re a GP and you believe a company has even <em>medium</em> potential to become the next consensus asset, paying the clearing price is the rational move every single time. The expected value of being on the cap table — measured not in investment returns but in AUM, fees, SPVs, and franchise value — overwhelms the downside of overpaying by orders of magnitude.</p><p>And that is why the clearing price for AI assets today — pre-product-market fit, sometimes pre-company, occasionally pre-coffee-machine — is <strong>$1 billion and climbing</strong>. It is not a valuation. It is a cost-of-doing-business calculation for GPs who understand that the game is no longer about picking winners. It is about being seen near them.</p><p>Look at the numbers. The top 30 venture funds captured 75% of all LP fundraising in 2024 — roughly $57 billion out of $76 billion raised. Just nine firms captured 46%. If you are not in that top tier, you are in a death spiral. And the price of admission to the top tier is having the right names on your cap table.</p><h3>The Cap Table Arms Race</h3><p>Let’s make this concrete. Here are the companies that every GP in America is fighting to access, and what they’ve absorbed:</p><ul><li><strong>OpenAI</strong> — ~$58B raised, $300B–$500B valuation</li><li><strong>Anthropic</strong> — ~$37B raised, $350B valuation</li><li><strong>xAI / SpaceX</strong> — ~$22B raised (xAI alone), $1.25T combined valuation</li><li><strong>Stripe</strong> — ~$10B raised, $107B valuation</li><li><strong>Anduril</strong> — ~$7B raised, $31B–$87B valuation</li><li><strong>Databricks</strong> — ~$6B raised, $62B valuation</li><li><strong>Cursor (Anysphere)</strong> — $2.3B raised, $29B valuation</li><li><strong>Safe Superintelligence</strong> — $3B raised, $32B valuation</li><li><strong>Thinking Machines Lab</strong> — $2B raised, $12B valuation</li><li><strong>Top ~10 Total: ~$150B+</strong></li></ul><p>One hundred and fifty billion dollars, absorbed by roughly ten companies. In an industry that raised $66 billion from LPs last year. PitchBook reported that 41% of <em>all</em> VC dollars deployed in 2025 went to just 10 startups. Crunchbase found that close to 60% of invested capital went to 629 companies raising rounds of $100 million or more. More than a third of global funding went to just 68 companies raising $500 million+.</p><p>This is not an investment market. It is a toll road. And the toll is paid in logos.</p><h3>The Stripe Original Sin</h3><p>Every addiction has a patient zero. Venture capital’s addiction to mega-rounds traces back to Thrive Capital and Stripe — but not in the way most people tell the story.</p><p>The venture capital business, as originally conceived, is supposed to work like this: you develop a differentiated view on a company or a market before the rest of the world catches up. You invest in what’s <em>coming</em>, not what’s <em>arrived</em>. You earn your returns by being early, by being right, and by having the conviction to pay a premium for something that most of your peers either can’t see or aren’t brave enough to underwrite.</p><p>In December 2014, Thrive Capital led Stripe’s $70 million Series C, writing a $30 million check at a $3.5 billion valuation. Thrive was four years old. Its fund was roughly $400 million. Josh Kushner’s firm was not Sequoia. It was not Andreessen Horowitz. It was a young New York–based firm betting nearly 8% of its fund on a payments company that, while growing fast, had not yet become the consensus generational asset it would later be recognized as. That took genuine conviction — the kind that earns you a seat at the table not because of your brand, but because the founder looks at your term sheet and thinks, <em>these people actually understand what we’re building</em>.</p><p>The even bigger conviction play came in 2023, when Stripe needed roughly $7 billion to avoid going public in a brutal market. Multiple well-known firms passed on the round — not because the business was broken, but because, as Patrick Collison later put it, they couldn’t answer the question, “Why are we so lucky?” It seemed too broadly available to be a real opportunity. Thrive took the opposite view. Kushner committed $1.8 billion before the round had even started coming together — more than a billion dollars ahead of any other firm — and then helped raise the remaining $5 billion. That is what conviction capital looks like. You don’t wait for the herd to validate your thesis. You <em>are</em> the thesis.</p><p>Founders Fund has operated this way in hard science for years — seeding Anduril in 2017 when defense tech was career poison in Silicon Valley, backing SpaceX as its first institutional investor in 2008 when rocket companies were considered vanity projects for billionaires, not venture-scale opportunities. A small number of excellent VCs have always worked this way: they invest ahead of consensus, absorb the ridicule, and collect the returns when the world comes around.</p><p>The problem is that roughly 90% of the industry does the opposite. Most venture capitalists don’t invest in what’s coming. They invest in what just worked. Three years ago, every pitch deck was a fintech. Before that, crypto. Before that, direct-to-consumer. The pattern is always the same: something succeeds, it gets “de-risked” in the minds of investors who missed it, and the entire industry piles in — not to the original company, but to everything that rhymes with it. Conviction gets replaced by pattern-matching. Underwriting gets replaced by cosplay.</p><p>And that is exactly what is happening in AI right now. Thrive earned its Stripe position through conviction. But the lesson the industry drew from it wasn’t “develop differentiated views on generational companies.” The lesson was “get on the cap table of the hot thing at any price.” So now every GP in America is writing $1 billion seed checks into every AI research lab founded by someone who once shared a Slack channel with Sam Altman. Mira Murati leaves OpenAI and raises $2 billion for Thinking Machines Lab at a $12 billion valuation — before the company has shipped a product. Ilya Sutskever departs and raises $3 billion for Safe Superintelligence at a $32 billion valuation — with approximately twenty employees and a website that is functionally a mission statement. The company doesn’t need product-market fit. It barely needs a website. It needs a founder with an OpenAI pedigree and a GP desperate enough to pay the clearing price.</p><p>This is not conviction. It is not investing in what’s to come. It is trend-following at $1 billion a ticket — dressed up in the language of conviction because the check sizes are large enough to feel brave.</p><p>The second-order effects of the original Thrive-Stripe relationship are what industrialized this behavior. Thrive’s fund at the time of its initial Stripe investment was roughly $400 million. The Stripe position didn’t merely return the fund — it <em>became</em> the fund. It became the firm. It became the pitch. Stripe has since raised roughly $10 billion in total funding, its valuation passed $107 billion in late 2025, and Thrive now manages north of $25 billion in AUM. That is a trajectory from a $400 million fund to a $25 billion platform in roughly a decade, driven overwhelmingly by the gravitational pull of a single company and the franchise effects it created.</p><p>The AUM feedback loop works like this:</p><ol><li>GP secures one “Lighthouse” asset (Stripe, SpaceX, OpenAI)</li><li>Lighthouse generates outsized markups</li><li>Markups attract LP capital (“we want access to the next round”)</li><li>LP capital inflates AUM</li><li>AUM generates management fees regardless of realized returns</li><li>GP transitions from the <em>underwriting business</em> to the <em>allocating business</em></li></ol><p>The skill set changes. The incentives change. The entire economic model inverts.</p><p>Every GP in the market today is paying a billion dollars at the seed stage because they saw what a single Lighthouse asset did for Thrive. But most of them are not doing what Thrive actually did — which was to develop a differentiated view on a specific company before the world caught up, and to back that view with the kind of concentrated capital commitment that made the founders take notice. They are doing the cheap imitation: paying the clearing price for anything that <em>looks like</em> the last thing that worked, and hoping that proximity to the trend is a substitute for insight into the future.</p><p>This is the number that matters: if Thrive had <em>not</em> gotten onto Stripe’s cap table, they wouldn’t have lost $400 million. They would have lost the trajectory from $400 million to $25 billion — roughly <strong>$24.6 billion in future AUM</strong>, generating hundreds of millions per year in management fees. The cost of not being on that cap table is <em>many multiples</em> of the total capital they could have raised from LPs in any single fund.</p><p>This is the GP calculus in a single sentence: <strong>the opportunity cost of missing a consensus winner is larger than your entire fund.</strong> And because every GP now understands this, they are all rationally overpaying — not for conviction, but for optionality on the chance that any given company becomes the next consensus name.</p><h3>The Machine</h3><p>To understand why this particular moment is so frenzied, you have to zoom out far enough to see the machine. Markets operate in cycles — debt accumulates, wealth concentrates, internal tensions rise, and technological disruption reshuffles the deck. These forces don’t move independently. They interlock like gears. And in 2026, the gears are grinding.</p><p>Start with the wealth transfer. US dollar-denominated assets are lagging behind both gold (up roughly 60% in 2025) and non-US equities on a total-return basis. This is the kind of divergence that appears in the late stages of a reserve currency’s dominance cycle — when creditors begin to hedge, when the premium on “safety” erodes, when the narrative shifts from “America is the cleanest shirt” to “maybe we should own the laundromat.” Venture capital is the <strong>last stand of the dollar-denominated miracle</strong> — the one asset class that still promises 10x returns in American innovation. And the institutional system is flooding it with capital precisely because every other dollar-denominated bet is underperforming a bar of metal.</p><p>Now add the GDP math. AI-related capital expenditure is contributing roughly <strong>140 basis points to US GDP growth</strong> in 2026, even as the rest of the economy softens under the weight of persistent rates, consumer exhaustion, and trillion-dollar deficits. Strip out AI capex, and America looks like Europe. Leave it in, and you have a growth story worth trillions. Roughly one-third of US economic growth this year is attributable to artificial intelligence.</p><p>That isn’t diversification. That isn’t a broad-based technology revolution lifting all boats. That is a single point of failure dressed up as a paradigm shift. Every allocator in America is trying to buy exposure to the same 140 basis points. And they’re paying anything the market asks.</p><p>This is the real clearing price mechanism. One billion dollars is not a valuation. It is a <strong>throughput requirement</strong> — the minimum check size that allows institutional capital to flow through a venture fund and into the only slice of the economy that is growing. The pension fund in Sacramento that needs to deploy $150 million into alternatives this quarter doesn’t care whether your Series A is “expensive” by historical standards. They care that it exists, that it has an AI label on it, and that the fund is large enough to accept their wire.</p><h3>The General Contractor Mirage</h3><p>Show me the incentive and I’ll show you the outcome. The AI infrastructure companies are exhibit A.</p><p>CoreWeave is valued like a high-margin software company on secondary markets. The reality: as of September 2025, it carried a <strong>debt-to-equity ratio of approximately 2.66</strong> and annual <strong>interest expenses exceeding $1.2 billion</strong>. These companies are not building proprietary models. They are not creating network effects. They are not developing software that improves with scale and generates compounding returns over time. They are plugging GPUs into walls and renting them by the hour. They are general contractors in the data center business — capital-intensive, cyclically exposed, and entirely dependent on the continued willingness of hyperscalers to outsource compute rather than build it themselves.</p><p>VCs are pricing “plugging things into walls” at 40–50x revenue because they need the “AI Infrastructure” logo to satisfy LPs. And every GP knows it. They are paying the price not because they believe in the business, but because <strong>the alternative — missing the cycle entirely — is worse for their careers than overpaying</strong>.</p><p>The geopolitical backdrop gives everyone cover. The push for resilience over efficiency — the onshoring imperative, the chip export controls, the scramble for energy sovereignty — has created a permission structure to build redundant capacity at any cost. Every data center is a jobs program. Every GPU cluster is a national security asset. Every overpriced infrastructure company is justified by the vague but unfalsifiable claim that “we can’t afford to fall behind.” The result is a capex supercycle financed by venture dollars, valued by software multiples, and rationalized by defense metaphors.</p><p>When social proof, FOMO, authority bias, and incentive-caused bias all converge on the same trade at the same time, the results are reliably catastrophic. This is one of those times.</p><h3>The Return Math (For Those Who Still Care)</h3><p>For a fund to return 10x on a $1 billion seed investment, the company must achieve a <strong>$100 billion exit</strong>. In the entire history of venture capital, the number of companies that have reached $100 billion in enterprise value from venture-backed origins is approximately fifteen. Against a universe of roughly 150,000 venture-backed companies, that’s a hit rate of <strong>0.01%</strong>.</p><p>But here’s the thing the return math misses: the GP doesn’t need the investment to work. The GP needs the <em>logo</em> to work. A $1 billion entry price into a company that becomes a household name — even if the investment itself generates mediocre returns — secures the GP’s ability to raise the <em>next</em> fund. And 2% of a $10 billion next fund is $200 million per year, forever.</p><p>The entry price is not an investment. It is an insurance premium against irrelevance.</p><h3>The Sanity Check</h3><p>There is a test that every financial innovation must pass before it earns the right to be called anything other than a bubble. Explain the trade to a normal person — someone with common sense, a mortgage, and a healthy suspicion of anyone wearing a vest with no shirt underneath — and see if they look at you like you’ve lost your mind.</p><p>A billion dollars for a company that hasn’t even bought a coffee machine yet? I can’t get a plumber to show up on a Tuesday, and we’re giving $200 million to a kid who “thinks” he can build a model better than OpenAI? The whole thing! We’re all just standing in line at a club where the drinks are five thousand dollars and the music is just a recording of Sam Altman clearing his throat. And the bouncer? The bouncer is a pitch deck with a TAM slide that says “Every Human on Earth.” I wouldn’t let my dry cleaner run a business this way, and I don’t even like my dry cleaner!</p><p>The 2026 AI venture market fails the sanity check spectacularly. Not because the technology is fraudulent — it is genuinely transformative. Not because every company is worthless — some will define this century. But because the price of admission has completely decoupled from the probability of success, and the entire apparatus of institutional capital allocation is organized around ensuring that nobody says this out loud.</p><h3>The Spread</h3><p>So where does this leave the thoughtful allocator?</p><p>The opportunity is not to avoid AI. It is to recognize that the consensus entry points — the $1 billion seeds, the $50 billion secondaries, the infrastructure “utilities” priced like software monopolies — are the <strong>wrong door</strong> into the right building.</p><p>The spread between what the market prices and what it misses is the alpha. The market prices logos, narratives, and access. It systematically underprices <strong>operational mechanics</strong> — the boring, unglamorous work of building competitive advantage through distribution, through pricing power, through technology applied to legacy industries where the incumbents are asleep and private equity hasn’t yet arrived.</p><p>In periods of late-stage wealth concentration and monetary excess, capital flows toward the “obvious” bets with increasing desperation and decreasing discipline. The one-third of GDP growth driven by AI is real, but the pricing of access to that growth is not efficient — it is panicked. And panicked pricing, for those with the patience to look past the consensus, is where asymmetric returns have always been found.</p><p>The billion-dollar cover charge is real. But the best party is always the one nobody’s heard of yet.</p><p><em>Disclaimer: This post is for informational and discussion purposes only and does not constitute investment advice. All estimates and third-party data cited are drawn from publicly available sources (PitchBook, Crunchbase, Tracxn, company announcements) and author analysis.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0cdca14c096" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Raw Dogging Boredom]]></title>
            <link>https://medium.com/@mccoype_64598/raw-dogging-boredom-9f20626ca157?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/9f20626ca157</guid>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Sun, 08 Feb 2026 14:20:20 GMT</pubDate>
            <atom:updated>2026-02-08T14:20:20.880Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5Ou3dYW6YtTHPclZtfW4JA.png" /></figure><p>I spend about six hours a day on my phone. I know this because my phone tells me, every Sunday morning, like a passive-aggressive doctor reading lab results. Six hours. Sometimes more. I reach for it when I’m taking a piss. When the sandwich line at lunch is more than three people deep. When I’m waiting for an elevator. When I’m <em>in</em> the elevator. I have become a person who cannot tolerate twelve seconds of nothing.</p><p>And here’s the thing I noticed. My best ideas — the ones that actually matter, the ones that turn into real deals, real insights, real clarity — don’t come to me when I’m scrolling. They come when I’m working out. Or lying in bed right before sleep. Or just staring at the ceiling like a psychopath. They come when my brain has absolutely nothing to do.</p><p>I’m not special in the addiction part. The average American now spends over seven hours a day on screens and checks their phone up to 200 times. But I might be onto something with the boredom part. Because when I went looking for why my best thinking happens when I’m doing nothing, what I found changed how I spend my lunch break.</p><h3>The slot machine in your pocket</h3><p>I want to be clear about something: this is not entirely my fault. And it’s not entirely yours either.</p><p>The people who built the apps on your phone knew exactly what they were doing. They’ve said so. Out loud. On camera.</p><p>Sean Parker, Facebook’s founding president, admitted at an Axios event in 2017 that the core design question behind Facebook was how to consume as much of users’ time and conscious attention as possible. He described the like button as a dopamine delivery mechanism — a “social-validation feedback loop” that gives users a little neurochemical reward every time someone engages with their content. Then he said the part that should make your skin crawl: the people who created these tools — Parker, Zuckerberg, Systrom — understood the psychological manipulation consciously and proceeded anyway.</p><p>Weeks later, Chamath Palihapitiya, Facebook’s former VP of User Growth, told a Stanford audience that the dopamine-driven feedback loops they built were undermining civil discourse and cooperation. His bluntest line: the users’ behaviors were being programmed without their awareness or consent. He said he felt tremendous guilt, doesn’t use social media, and won’t let his kids near it.</p><p>These aren’t conspiracy theories. These are the architects of the system describing the blueprints.</p><p>The mechanics are specific. Infinite scroll — invented in 2006 by Aza Raskin — eliminates natural stopping points by exploiting “unit bias,” our instinct to finish a unit of something. Raskin later reflected that optimizing for ease of use doesn’t mean it’s best for the user or for humanity, and co-founded the Center for Humane Technology out of regret. Variable reward schedules — the same intermittent reinforcement that makes slot machines the most addictive form of gambling — drive the randomized feeds on Instagram, TikTok, and X. Snapchat streaks exploit social obligation. Red notification badges trigger urgency without revealing content, forcing you to tap.</p><p>I know all of this. I’ve read the research, watched the documentaries, nodded along to every podcast about it. And I still can’t stop.</p><p>Tristan Harris, the former Google design ethicist who wrote a 141-slide internal presentation warning that the company was hijacking users’ attention, described it as a race to the bottom of the brainstem. Nir Eyal codified the playbook in his book <em>Hooked</em> — Trigger, Action, Variable Reward, Investment — which became Silicon Valley’s product bible. Frances Haugen, the Facebook whistleblower, leaked internal research showing Facebook’s own teams found the platform worsened body image and mental health for significant numbers of teen users.</p><p>The numbers tell the rest of the story. Americans spend roughly 40% of their waking hours on screens. Teens receive a median of 237 notifications per day, with some getting over 500. Google, Meta, and Amazon alone generated hundreds of billions in advertising revenue last year — all of it purchased with your attention. Every second you spend staring at your phone is a second someone is getting paid for.</p><p>I am a customer of that machine. Six hours a day. And the thing I started to realize is that the cost isn’t just time. It’s something the neuroscientists have a name for. And it’s the thing I was accidentally shutting off every time I picked up my phone instead of letting myself be bored.</p><h3>Your brain has a creativity engine. You keep turning it off.</h3><p>Here’s where this gets interesting. And by interesting I mean: here’s where I got angry at myself.</p><p>In 2001, a neurologist named Marcus Raichle at Washington University noticed something strange in brain imaging data. Certain regions of the brain became <em>more</em> active when subjects stopped doing tasks. Not less. More. He called it the Default Mode Network — a constellation of brain areas that lights up during mind-wandering, daydreaming, self-reflection, and future planning.</p><p>This is not the brain idling. This is the brain doing its deepest work.</p><p>A 2024 study published in <em>Brain</em> provided causal evidence that the DMN is essential for creative thinking — researchers used electrical stimulation to temporarily suppress DMN regions, and subjects’ responses became measurably less original. Other neuroimaging work has shown that highly creative people exhibit stronger connectivity between the DMN and the brain’s executive control network — meaning the most creative minds aren’t just better daydreamers, they’re better at <em>catching</em> their daydreams and evaluating them.</p><p>Here’s the kick in the teeth: the DMN and the brain’s task-positive network are anticorrelated. When one is on, the other is off. Every time I pick up my phone to scroll Instagram in a sandwich line, I activate the task-positive network and silence the DMN. Every notification. Every feed refresh. Every mindless swipe. I’m flipping the off switch on the exact part of my brain that generates my best thinking.</p><p>Mary Helen Immordino-Yang at USC put it starkly in her paper “Rest Is Not Idleness”: constantly demanding external attention from people may prevent them from advancing beyond processing events to constructing deeper understanding of what those events actually mean.</p><p>Every time you fill a boring moment with your phone, you are actively preventing your brain from doing the work that turns information into insight. You’re not resting. You’re sabotaging.</p><p>Now here’s why you get your best ideas in the shower, on a run, or right before you fall asleep. Those are moments when the brain shifts into alpha wave production — electrical patterns in the 8 to 13 Hz range associated with relaxed, internally focused cognition. Research by Caroline Di Bernardi Luft at Queen Mary University of London found that elevated alpha waves helped participants generate more unexpected and remote associations — the kind of non-obvious connections that define creative breakthroughs. Alpha waves suppress habitual thinking patterns, creating space for novel ideas to surface.</p><p>That moment in the shower when a solution suddenly appears? That’s not random. That’s your DMN running in alpha-wave mode, making connections your focused, phone-addled brain never could.</p><p>I think about this constantly now. The best pattern recognition I do in my job — seeing how a fragmented industry could consolidate, or how layering technology onto a legacy business could create a market leader — doesn’t happen at my desk staring at a financial model. It happens when I’m doing nothing. And for years, I was filling that nothing with Twitter.</p><h3>The pebble counter, the woman with no pen, and the guy who got the plague</h3><p>I am not the first person to figure this out. I am, in fact, about 2,400 years late.</p><p>Charles Darwin built a gravel path at his home in Kent called the Sandwalk — what he called his “thinking path.” He walked it multiple times a day, and he had a system: he’d stack flint pebbles at the entrance and kick one away with each lap. Some ideas were “three-flint problems.” His son Francis later wrote that Darwin found horseback riding too mentally stimulating for real thinking — the demands of handling the horse occupied enough of his mind to block deep concentration. The <em>monotony</em> of walking was the point.</p><p>Darwin didn’t have a phone to reach for. But he intuitively understood something that neuroscience wouldn’t confirm for another 150 years: the brain does its most creative work when the body is doing something boring and repetitive.</p><p>Then there’s J.K. Rowling. In 1990, she was sitting on a delayed train from Manchester to London when the idea for Harry Potter arrived in what she described as an intense rush of excitement. But here’s the detail that matters: she had no pen. She dug through her bag looking for anything to write with — not even an eyeliner — and came up empty. So she just sat there. For four hours. No pen. No distraction. Just her brain, a delayed train, and nothing to do but think. By the time she arrived in London, she had the architecture for a seven-book series that would become the bestselling franchise in literary history.</p><p>She later called the absence of a pen the best thing that could have happened to her.</p><p>Now imagine that same train ride in 2026. She’d have opened Instagram before the train left the station. Harry Potter would be a half-remembered daydream wiped out by a push notification from DoorDash.</p><p>And then there’s Isaac Newton. In 1665, the Great Plague forced Cambridge to close, and a 23-year-old Newton retreated to his family’s farmhouse for 18 months of enforced isolation. No colleagues. No library. No stimulation beyond a garden and his own thoughts. During that stretch he developed early calculus, discovered the nature of light, and formulated his laws of motion and gravity. He later wrote that during those months he was in the prime of his life for invention — more focused on mathematics and philosophy than at any point before or since. The apple tree still stands.</p><p>The most productive intellectual period in recorded history was born from enforced boredom and a plague.</p><p>The pattern is everywhere once you look for it. Nikola Tesla walked 8 to 10 miles a day and envisioned the complete design for the alternating current motor during a stroll through a Budapest park. Einstein loved sailing precisely because the wind sometimes died, leaving him becalmed with nothing to do but scribble in his notebook. Lin-Manuel Miranda picked up Chernow’s Hamilton biography on his first vacation from <em>In the Heights</em> and later said it was no coincidence that his best idea arrived when his brain finally got a moment’s rest. Bill Gates started taking biannual “Think Weeks” in isolated cabins — one produced the memo that launched Internet Explorer.</p><p>All of these people, across centuries, discovered the same thing: the breakthrough doesn’t come from more input. It comes from less.</p><h3>The science of doing nothing</h3><p>The research confirms all of this so cleanly it’s almost annoying.</p><p>Timothy Wilson at the University of Virginia ran a now-famous experiment in 2014. He asked several hundred people to sit alone in a room for 6 to 15 minutes with no phone, no book, nothing — just their thoughts. About half found it unpleasant. When tested at home, roughly a third cheated. Then he gave 55 subjects access to a button that would deliver an electric shock — a shock they had previously said they would pay money to avoid — and left them alone for 15 minutes. Two-thirds of the men and a quarter of the women shocked themselves rather than sit quietly with their own minds. One man hit the button 190 times.</p><p>People would rather <em>electrocute themselves</em> than be bored for fifteen minutes. And that was 2014 — before TikTok, before Reels, before the attention economy went fully nuclear.</p><p>But here’s the twist. Sandi Mann at the University of Central Lancashire took that discomfort and turned it into a weapon. She had participants spend 15 minutes doing the most boring thing imaginable — copying phone numbers from a directory — and then tested their creativity. The bored group significantly outperformed the control group on divergent thinking. A follow-up found that the most passive, boring condition — simply reading phone numbers aloud — produced the most creative participants of all. The mechanism was daydreaming: greater passivity created more of it, and daydreaming drove the creativity boost.</p><p>The <em>more bored</em> you are, the <em>more creative</em> you become. Not despite the boredom. Because of it.</p><p>Stanford confirmed the walking connection: Marily Oppezzo and Daniel Schwartz found that walking boosted creative output by an average of 60%, with 81% of participants improving on divergent thinking tests. It didn’t matter if they walked outside or on a treadmill facing a blank wall. The effect even carried over after sitting back down. And a UC Santa Barbara study found that people who did a simple, undemanding task during a break — maximizing mind-wandering — improved roughly 40% on creative problem-solving, beating rest, demanding tasks, and no break at all.</p><p>Mihaly Csikszentmihalyi, the psychologist who defined flow state, saw boredom not as flow’s enemy but as its fuel. Boredom signals that your skills exceed your current challenge — it creates internal pressure to seek a harder problem, which is the doorway to flow. You can’t reach the deep-focus state everyone romanticizes if you never let yourself get bored enough to crave it.</p><p>The detox research has caught up too. A 2025 Harvard study found that a one-week social media detox reduced anxiety by 16% and depression symptoms by nearly 25%. A Georgetown study found that two weeks of reduced phone usage produced well-being improvements comparable to cognitive-behavioral therapy — larger than typical antidepressant effects. Participants slept 20 extra minutes per night.</p><p>And here’s the one that should haunt you: Adrian Ward at UT Austin found that the mere presence of a smartphone on a desk — face-down, turned off — reduced cognitive capacity. Your brain allocates resources to suppress the urge to check it even when you’re not using it.</p><p>Your phone is costing you IQ points just by sitting there.</p><h3>My lunch break</h3><p>So I started doing something about it.</p><p>A few months ago, I began what I now — half-jokingly, half-seriously — call raw dogging boredom. The routine is simple. During the week, when I can manage it, I leave the office to grab lunch on foot. No phone. No AirPods. No screens of any kind. I walk downstairs, head down the block, stand in line, get my food, walk back. In Manhattan this takes 30 to 45 minutes. It’s not a meditation retreat. It’s a sandwich run. But I do it completely unplugged.</p><p>The first few times were brutal. I won’t pretend otherwise. My hand kept reaching for my pocket like a phantom limb. Standing in line with nothing to look at felt almost physically uncomfortable — itchy. I noticed buildings I’d walked past a hundred times. I people-watched like it was 1997. I felt restless. Then I felt stupid for feeling restless about something as mundane as buying lunch without entertainment.</p><p>And then, about fifteen minutes in, something shifted.</p><p>My brain started working on things I hadn’t asked it to. A deal I’d been circling for weeks — I’d been so focused on the customer base that I’d missed the real value sitting in the service contracts. A conversation I needed to have with a partner — the right framing materialized out of nowhere. Connections between sectors I’d never paired at my desk with seventeen browser tabs open. It was like my brain had been trying to talk to me for months and I’d been drowning it out with podcasts and push notifications.</p><p>This is what the neuroscientists call the DMN firing up. This is alpha waves doing their thing. This is what Darwin felt on the Sandwalk and Rowling felt on that train. I’m not comparing my lunch walks to the theory of evolution or Harry Potter. But the mechanism is the same. The brain, given nothing to react to, starts to <em>create</em>.</p><p>I now protect this time. Not every day — I’m not a monk. But when I can, I go get lunch on foot with no phone. I let my mind go wherever it wants on the walk there and back. Sometimes I think about deals. Sometimes I think about my daughter — what kind of world she’s going to inherit from a generation that can’t sit still for six minutes. Sometimes I think about absolutely nothing and that’s fine too.</p><p>The irony is that in an industry built on information — where I spend all day consuming research, data, decks, and deal memos — the most valuable thing I’ve found is the deliberate absence of information.</p><p>Pascal wrote in the 1660s that all of humanity’s problems stem from our inability to sit quietly in a room alone. That was 360 years ago. Before electricity. Before radio. Before television. Before the internet. Before a trillion-dollar industry was constructed around the singular goal of making sure you never, ever sit quietly in a room alone.</p><p>The people who built the attention economy are brilliant. They’ve deployed the most sophisticated behavioral science ever assembled to ensure that every second of potential boredom is converted into engagement. And they’ve largely won. Most people can’t sit with their own thoughts for six minutes. Some would rather get shocked.</p><p>But here’s what they can’t engineer away: the Default Mode Network doesn’t care about their business model. It’s been running the same software for hundreds of thousands of years. It activates when stimulation drops. It makes connections you can’t make consciously. And it only works when you give it room.</p><p>You don’t need a cabin in the Pacific Northwest. You don’t need a delayed train or a plague. You need thirty minutes and the willingness to be bored.</p><p>Leave your phone on your desk. Walk out the door. Let your brain itch. Let it wander. Let it do the thing it was built to do before someone figured out how to monetize the alternative.</p><p>Raw dog the boredom. See where your mind goes when nobody’s paying to point it somewhere else.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9f20626ca157" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Precision Illusion]]></title>
            <link>https://medium.com/@mccoype_64598/the-precision-illusion-6034f113d170?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/6034f113d170</guid>
            <category><![CDATA[investing]]></category>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Mon, 19 Jan 2026 19:39:05 GMT</pubDate>
            <atom:updated>2026-01-19T19:39:05.477Z</atom:updated>
            <content:encoded><![CDATA[<p>The Precision Illusion</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TfSzYQza9-cWwmuFwecRjg.png" /></figure><p><strong>Why smart money keeps mistaking the blueprint for the building.</strong></p><p>In 2011, Marc Andreessen wrote that “software is eating the world.” It was a brilliant observation. But nobody mentions the side effect: for the last fifteen years, a lot of people have been getting rich by pretending to chew.</p><p>Steve Jobs, a dictator with good taste, had a way of making a simple sentence feel like a trapdoor. He famously said that once you realize everything around you was made by people no smarter than you, you can change it. You can mold it.</p><p>It’s a liberating idea. The world isn’t fixed; it’s a set, and you can rearrange the furniture.</p><p>But here’s the unsettling part. The part Jobs didn’t say out loud because it’s much less flattering:</p><p>Most people in professional investing are not trying to build a new world.</p><p>They are trying to get a room in the hotel lobby of someone else’s world before the rates go up.</p><p>We have divided the economy into two groups: <strong>World-Makers</strong> and <strong>Financial Tourists.</strong></p><p>The World-Makers are the locals. They speak the language, they know which roads wash out in the rain, and they build the infrastructure.</p><p>The Financial Tourists just visit. They take photos (management fees), buy souvenirs (carry), and rely on a guidebook (the spreadsheet) that was written by someone who has never been there.</p><p>And right now, the Tourists are charging 2-and-20 to tell you what the weather is like, without ever stepping outside.</p><h3>Scene 1: Funding Market Fit vs. Product Market Fit</h3><p>Imagine you’re at a venture firm in late 2023. The office has polished concrete floors and an espresso machine that requires a PhD to operate. There are just very earnest people in Patagonia Better Sweater vests, specifically the “Stonewash” grey ones, staring at a demo where a chatbot writes an email.</p><p>The email sounds like it came from a medium-talented Associate who has slept four hours and is quietly resentful.</p><p>OpenAI has done something unmistakable: it has made a new world. The best investors, the ones who actually have intuition, saw this coming years ago. They understood the shift in physics.</p><p>But the room is full of Tourists. They aren’t looking for <strong>Product Market Fit</strong> (building something users love). They are looking for <strong>Funding Market Fit</strong> (building something that raises the next round).</p><p>The reaction is a stampede. It’s the scene in <em>Curb Your Enthusiasm</em> where everyone runs for the buffet, but instead of shrimp, they’re grabbing domain names.</p><p>Pitch decks start to sound like Mad Libs:</p><p>“We are AI for ____.”</p><p>Legal. Real estate. Dermatology. “AI for dental front desks.”</p><p>By the end of 2024, funding for AI reached roughly $79.2B. But look at the split. A massive chunk isn’t going to the people building the foundation. It’s going to the people building “UI layers.”</p><p>We’ve seen this movie before. The pattern is identical to fintech circa 2019. Stripe solved distribution. Everyone else copied the category. Billions went to “Stripe for X” companies that had beautiful APIs but no natural distribution mechanism.</p><p>Right now, OpenAI has solved a distribution problem. The chat interface is the distribution. But the tourists aren’t asking “what distribution problem does this AI company solve?” They’re asking “is this AI?”</p><p>They are buying novelty belt buckles and trying to sell them to the miners, forgetting that the miners are the only ones who can afford pants.</p><h3>Scene 2: The Missionary vs. The Mercenary (The Stripe Fallacy)</h3><p>Stripe is the ultimate World-Maker story. But almost everyone tells the story wrong, and that error has incinerated billions of dollars.</p><p>The Tourist version of the story is: <em>“Stripe made payments easy via API. Therefore, if I make an API for [insert complex backend process], I will be worth $90 billion.”</em></p><p>This logic gave us a deluge of “Stripe for Medical Billing,” “Stripe for Fraud Detection,” and “Stripe for Identity Verification.”</p><p>The thesis was seductive: Take a messy piece of financial plumbing, wrap it in clean documentation, and wait for the unicorn valuation.</p><p>But the World-Maker version of the story is different: <em>“Stripe solved distribution.”</em></p><p>Before Stripe, selling payments was a door-to-door combat sport. You needed a sales guy to buy a steak dinner for a CFO.</p><p>The Collison brothers were Missionaries. They realized that if you made payments an API, you didn’t need a sales team. You needed developers.</p><p>Developers are the ultimate viral coefficient. One developer puts Stripe into an app, and suddenly that app can take money. This was a <strong>One-Two Punch</strong>: The API (product) + The Developer (distribution).</p><p>The Mercenaries missed the second punch. They poured billions into “Infrastructure-as-a-Service” plays that had lovely APIs but no natural distribution mechanism.</p><p>A developer doesn’t wake up at 2 a.m. desperate to integrate a medical billing API. That requires a sales team. It requires grinding. It requires enterprise contracts.</p><p>The Tourists confused a technical feature (API) with a business model. They built beautiful toll booths on roads where nobody was driving.</p><h3>Scene 3: The Platform Purgatory</h3><p>If Venture Capital is “vibes,” Private Equity is supposed to be “rigor.” But recently, PE has been playing a dangerous game of musical chairs where they removed all the chairs and hoped nobody would notice.</p><p>Around 2018, the industry discovered the “Rollup.”</p><p>The thesis: Buy 50 small dentists/HVAC repairmen/car washes, mash them together, call it a “Platform,” and sell it for double the multiple.</p><p>It was the perfect “safety in numbers” trade. It felt conservative. Everyone did it.</p><p>But they forgot to ask the only question that matters: <em>Who are you going to sell this to?</em></p><p>A true “Strategic Buyer,” a real operating company like P&amp;G or Honeywell, doesn’t want your Frankenstein monster of 50 different billing systems and a bunch of dentists who hate you.</p><p>So, who buys it?</p><p>Another PE fund.</p><p>But wait. Now interest rates are up. The other PE funds can’t afford to pay 18x EBITDA for your dental chain.</p><p>So what happens? The liquidity dries up.</p><p>We are now seeing a flood of “Continuation Vehicles.”</p><p>This is a polite financial term for: “I can’t sell this thing, so I’m going to move it from my left pocket to my right pocket and charge you a fee for the privilege.”</p><p>Billions of dollars of capital are currently stuck in these zombie platforms. They were built for a spreadsheet, not for a buyer.</p><h3>Scene 4: The Spreadsheet vs. The Rain</h3><p>If you want a single emblem for the Precision Illusion, look at the car wash.</p><p>To a spreadsheet, a car wash is beautiful. It’s a recurring subscription model. It’s a SaaS company that washes Jeeps.</p><p>But Zips Car Wash, one of the largest chains in America, didn’t file for bankruptcy because they forgot how to wash cars. They went bust because of the Growth Trap.</p><p>Here is what happens when a Financial Tourist tries to play Operator:</p><p>The Private Equity model demands growth. You can’t just own a nice car wash; you need to double EBITDA. But organic growth in car washes is low. There are only so many cars in a town.</p><p>So, to feed the model, Zips had to acquire.</p><p>They bought the prime locations first. Those were easy.</p><p>But the model was hungry. So they bought the “B” locations.</p><p>Then everyone else got into the trade. Prices skyrocketed.</p><p>So Zips had to keep buying, but now they were forced into “C” locations. Places with worse demographics and, crucially, worse weather.</p><p><strong>They literally bought rain.</strong></p><p>You can’t model the weather in Excel, but you can definitely feel it in your debt covenants. When it rains in a “C” location, volume drops to zero. But the debt payments (which were floating rate, because they assumed money would stay free forever) kept coming.</p><p>They didn’t ignore the details because they were stupid. They ignored the details because the <strong>Financial Engineering</strong> required them to. The need for “Platform Scale” forced them to buy bad assets just to keep the growth chart pointing up.</p><h3>What Private Equity Actually Was</h3><p>Here’s the thing: Private Equity didn’t used to be a game of financial hot potato.</p><p>The original model, the one that actually generated the returns the industry still advertises, was different. You bought a broken company cheap. You fixed it. You grew it 20%+ per year. Then you either took it public or sold it to a strategic acquirer who actually wanted the specific thing you built.</p><p>Clayton, Dubilier &amp; Rice buying Kinko’s in 1996, fixing the technology infrastructure and franchise model, then selling to FedEx. That’s a strategic exit. FedEx wanted the physical footprint and customer base.</p><p>KKR buying Sealy Mattress, consolidating a fragmented industry (sound familiar?), but actually fixing the supply chain and retail relationships, then selling to Tempur-Pedic. Strategic wanted the brand and distribution.</p><p>These weren’t “multiple arbitrage” plays. They were operational transformations. The PE firm added real value. Better systems, better management, better strategy. The exit multiple was higher because the business was better, not because the debt markets were frothier.</p><p>But somewhere around 2015, the script flipped.</p><p>Interest rates went to zero. Capital flooded the market. Fund sizes ballooned. And suddenly you couldn’t deploy $2 billion by buying one broken company and spending five years fixing it. You needed volume. You needed “platforms.” You needed to buy fifty things at once and pray someone else would buy the pile.</p><p>The strategy shifted from “Buy broken, fix it, sell to strategic” to “Buy decent, scale it fast, flip to another fund.”</p><p>And that’s how you end up with Zips. They didn’t buy broken car washes and fix them. They bought perfectly fine car washes and broke them by forcing growth where growth didn’t make sense.</p><p>The debt markets loved it. The spreadsheet loved it. The rain did not.</p><h3>The Actual Alpha</h3><p>So, is everyone a fraud? No.</p><p>The irony is that while the Financial Tourists are fighting over scraps in the hotel lobby, the World-Makers are quietly generating the massive returns the industry promises but rarely delivers.</p><p>Real Alpha comes from doing the thing the spreadsheet can’t quantify: <strong>Contact with Reality.</strong></p><p><strong>In Venture</strong>, the World-Makers are the ones who ignore “funding market fit” and bet on physics. They aren’t funding the 50th AI copywriter. They’re funding the companies building the energy infrastructure that AI data centers will eventually require. They’re funding the companies solving actual distribution problems, not just slapping AI onto an existing workflow.</p><p>In Private Equity, the World-Makers are doing what PE was originally designed to do:</p><p>They buy a $50M industrial distributor that’s losing money because nobody has looked at the pricing model in fifteen years. They spend six months building a margin optimization system. Gross margins go from 12% to 22%. Revenue grows 25% per year because now the sales team can actually quote competitively. Three years later, they sell to Grainger or a larger distributor who wants the customer relationships and the pricing system you built.</p><p>They buy complexity (which is cheap) and sell simplicity (which is expensive).</p><p>When they sell that business, they aren’t looking for a “Sponsor-to-Sponsor” trade. They sell to a Strategic Buyer like a Honeywell, a Danaher, or a Microsoft.</p><p>Why? Because a Strategic Buyer pays a premium for something they actually need. And you built something they actually need because you weren’t trying to build a financial instrument. You were trying to build a functional machine.</p><h3>The Theory of the Crime</h3><p>So why does this keep happening? Why do smart people, people with decades of experience, impeccable credentials, access to infinite information, keep making the same mistake?</p><p>René Girard gave us the answer in the 1960s: Mimetic Desire.</p><p>We don’t want things independently. We want what other people want.</p><p>In investing, mimetic desire explains everything.</p><p>Nobody independently evaluated “Stripe for Medical Billing” and concluded it was the future. They saw Stripe valued at $95 billion and thought: That category is hot. I want exposure to that category.</p><p>The desire wasn’t for the specific company. It was for what everyone else wanted: fintech exposure. The category became the object of desire, not the unique insight.</p><p>This is why all the pitch decks looked identical. It wasn’t coordinated. It was mimetic. Everyone was copying each other because everyone was looking at each other instead of the actual problem.</p><p>Girard’s insight is that this kind of desire creates a closed loop. We stop looking at the world and start looking at each other. In investing, we’ve built an entire asset class that increasingly only knows how to buy and sell from itself.</p><p>According to Bain &amp; Company, sponsor-to-sponsor transactions now make up the majority of PE exits, while sales to strategic buyers, the ones that actually validate you built something useful, have been steadily declining.</p><p>We are passing the same companies back and forth, adding a layer of debt each time, like a game of Hot Potato where the potato keeps getting more expensive and nobody wants to admit it’s starting to smell.</p><p>But the cracks are showing.</p><p>Junior investors, the ones actually doing the Excel modeling, are realizing the dream of “Carry” is a mirage. The exit isn’t coming. The liquidity is trapped in a Continuation Vehicle.</p><p>So they are leaving. They are starting “Search Funds.” They are trying to buy one real business and actually run it. They are trying to become World-Makers because they realized the Tourist game is rigged.</p><p>Steve Jobs was right. The world is made by people no smarter than you.</p><p>But he didn’t mention that the people with the most capital, the ones whose job is to find and fund the world-makers, spend most of their time as tourists in worlds other people built.</p><p>Not because they’re stupid. But because the system rewards showing up to the hotel lobby more than building the hotel. The 2-and-20 fee structure. The sponsor-to-sponsor exit market. The LP expectations. They all select for tourists over builders.</p><p>The real returns, the ones that actually compound and the ones you actually keep, don’t come from the spreadsheet. They come from knowing which roads wash out in the rain.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6034f113d170" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Portrait on the Wall]]></title>
            <link>https://medium.com/@mccoype_64598/the-portrait-on-the-wall-4f0df4e4f3bd?source=rss-9db8f4f15a12------2</link>
            <guid isPermaLink="false">https://medium.com/p/4f0df4e4f3bd</guid>
            <category><![CDATA[culture]]></category>
            <category><![CDATA[history]]></category>
            <dc:creator><![CDATA[Peter McCoy]]></dc:creator>
            <pubDate>Mon, 12 Jan 2026 05:06:09 GMT</pubDate>
            <atom:updated>2026-01-12T05:06:09.518Z</atom:updated>
            <content:encoded><![CDATA[<h4>How Andrew Jackson Explains Donald Trump and Why America Keeps Casting This Role</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*sKXtkXX68K5dHHuju5JWeA@2x.jpeg" /></figure><p>Donald Trump’s first decorating decision as president wasn’t the couch or the drapes or the lighting, which is unfortunate because the Oval Office lighting makes everyone look like they’re about to be questioned by airport security.</p><p>It was Andrew Jackson.</p><p>This is not a neutral choice. This is not Lincoln, the national comfort blanket, or Washington, the safe brand. Andrew Jackson is the guy you choose when you want the room to say something aggressive. Jackson is the portrait you hang when the message is: I am not here to be approved of. I am here to fight the people who think this place belongs to them.</p><p>Trump didn’t stop there. He went to Jackson’s home, The Hermitage, and gave a speech that felt less like a historical commemoration and more like a knowing wink. He talked about Jackson defying arrogant elites and then added, essentially, does this sound familiar?</p><p>The audience laughed. Trump laughed. Jackson, in oil and canvas, did not blink.</p><p>That laugh matters.</p><p>Because it suggests something most Americans are uncomfortable admitting: Trump is not an anomaly. He is not a glitch. He is not a once-in-a-lifetime disruption dropped from the sky.</p><p>He is a type.</p><p>And Andrew Jackson is the instruction manual.</p><p>This is not a comparison essay. It’s a diagnosis.</p><p>Once you see that, American history stops looking like a straight line of progress interrupted by the occasional monster. It starts looking like a cycle, one that produces the same character, with the same habits, under the same conditions, over and over again.</p><p>The details change. The role doesn’t.</p><p>⸻</p><p><strong>The First Version of the Movie</strong></p><p>Andrew Jackson did not grow up wanting to be president. He grew up wanting not to be humiliated.</p><p>That distinction turns out to matter more than almost anything else.</p><p>Jackson’s childhood reads like a factory for grievance. Orphaned at fourteen. Scarred literally and psychologically by the Revolutionary War. Captured by the British as a teenager, ordered to clean an officer’s boots, refused, and slashed across the face with a sword for it. That scar stayed with him for life, the kind of mark that, once you have it, explains your whole life backward.</p><p>He learned early that the world divides cleanly into two groups: people who give orders and people who take them. And that the only real sin is being forced into the second category.</p><p>Jackson grows up, becomes a frontier lawyer, a land speculator, a general, a plantation owner. Which is already revealing, because by the time he’s calling himself the champion of the common man, he owns enslaved people and more land than most of his supporters will see in their lifetimes.</p><p>But American populism has always been comfortable with this contradiction. In fact, it relies on it. The leader doesn’t have to be the people. He has to fight the people their enemies hate.</p><p>This pattern, building coalitions around shared enemies rather than shared interests, will become the signature of this political style. It explains why these movements can hold together groups that, on paper, have very little in common.</p><p>Jackson’s true political birth happens in 1824.</p><p>He runs for president. He wins the most popular votes. He wins the most electoral votes. And then he loses.</p><p>No one gets a majority in the Electoral College, so the House of Representatives decides the election. John Quincy Adams becomes president. Henry Clay, who helped Adams win, becomes Secretary of State.</p><p>Jackson doesn’t just think this is unfair. He thinks it’s rigged.</p><p>This is the moment everything snaps into focus.</p><p>Because “I lost” is embarrassing.</p><p>“They stole it” is clarifying.</p><p>It turns a candidate into a crusader. It turns followers into a movement. It tells everyone involved that they are not merely disappointed. They are enlightened. They see how the system actually works. They are no longer naive.</p><p>Jackson spends the next four years doing something new in American politics. He runs a permanent campaign of rage, not persuasion, not negotiation, but rage. He runs not against Adams so much as against the system that allowed Adams to exist.</p><p>When he wins in 1828, the country experiences something it has not fully prepared for.</p><p>The people show up.</p><p>At Jackson’s inauguration, crowds flood the White House so aggressively that staff set out tubs of whiskey punch on the lawn to lure them back outside before the furniture is destroyed. This is often told as a funny anecdote about rowdy frontiersmen, but it’s actually a governing theory in action.</p><p>If the elite won’t let you in politely, come in loudly.</p><p>This is Jacksonian politics in its purest form: the country doesn’t belong to the well-mannered. It belongs to whoever is willing to take it.</p><p>⸻</p><p><strong>Wait, I’ve Seen This Movie</strong></p><p>Donald Trump’s origin story is obviously different. No sword scars. No frontier wars. But the psychological architecture rhymes.</p><p>Trump grows up in Queens, not Manhattan. This matters more than people think. Manhattan is not just a borough. It’s a caste system. Trump spends his early career trying to be accepted by people who make it very clear they do not consider him one of them.</p><p>He kept lists. Actual lists of people who snubbed him, who wouldn’t let him into their clubs, who refused to take his calls. He called reporters in the 1980s and 90s posing as his own publicist, a spokesman named “John Barron” or “John Miller,” to plant stories about how successful and desirable he was, how deals were breaking his way, how women were pursuing him.</p><p>This is not how relaxed power behaves.</p><p>It’s the behavior of someone for whom the original wound never quite closes, someone constructing a performance of invulnerability in real time.</p><p>He compensates the way many people do: louder, bigger, shinier, more aggressive. He builds a public persona out of dominance. He creates a version of himself that cannot be ignored, because being ignored is the original injury.</p><p>Trump is not an ideologue. He’s a performer. But he’s performing something very specific: a man who cannot be controlled.</p><p>That performance becomes political in 2015 because the country is ready for it.</p><p>Trust in government has collapsed to historic lows. By the mid-2010s, fewer than one in five Americans said they trusted the federal government to do the right thing most of the time. Wealth inequality has become visible in a way it hadn’t been for generations. The financial crisis happened, the people responsible were never punished, and institutions began to feel distant, self-protective, and strangely uninterested in explaining themselves to people who don’t speak their language.</p><p>And then Trump shows up and says the quiet part out loud.</p><p>Not eloquently. Not carefully. But recognizably.</p><p>He doesn’t promise to fix the system. He promises to fight the people running it.</p><p>This is the key misunderstanding. Trump’s appeal is not policy. It is dominance as representation.</p><p>He doesn’t say,<em> I will govern better than them.</em></p><p>He says, <em>I will humiliate them the way you wish you could.</em></p><p>Once you see that, a lot of otherwise baffling behavior stops seeming random.</p><p>⸻</p><p><strong>The Violence Question (or, Why the Cane Matters)</strong></p><p>Jackson’s violence is literal. Duels. Brawls. Executions ordered during wartime. Bullets lodged in his body for decades from altercations he refused to walk away from.</p><p>The most famous moment comes late. An assassin tries to kill him outside the Capitol. Two pistols misfire, both of them, which is statistically absurd. Jackson, sixty-seven years old, charges the man and beats him with a cane until others pull him away.</p><p>This is absurd. It’s also diagnostic.</p><p>Because the violence isn’t about cruelty. It’s about proof.</p><p>It answers a question followers always ask, even if they don’t realize they’re asking it: <em>Will he actually fight?</em></p><p>Jackson’s answer is physical.</p><p>Trump’s answer is rhetorical.</p><p>When Trump says something transgressive, something polite society agrees you are not supposed to say, the content is less important than the outcome. The outrage happens. The headlines explode. Pundits declare this is finally the end. And then nothing happens.</p><p>He survives.</p><p>Each survival is a misfired pistol.</p><p>Each scandal that fails to destroy him reinforces the idea that he is untouchable. And untouchability is the product being sold.</p><p>The audience isn’t cheering violence.</p><p>They’re cheering invulnerability.</p><p>⸻</p><p><strong>How Outsiders Talk Like Insiders</strong></p><p>Jackson is often remembered as impulsive. Trump too.</p><p>This misses something important.</p><p>Jacksonian politics looks chaotic from the outside and disciplined from the inside.</p><p>Jackson builds a newspaper network. Loyal editors. Repeated phrases. Simple villains. Relentless messaging. Politics stops being genteel correspondence between elites and becomes mass communication designed to synchronize an emotional reality.</p><p>The most influential outlet was the United States Telegraph, run by Duff Green, a Jackson loyalist who understood that you don’t win arguments by being right. You win by being louder and more consistent than everyone else. Green printed the same accusations, corrupt bargain, aristocratic conspiracy, stolen election, in slightly different forms, day after day, until they hardened into common sense for Jackson’s supporters.</p><p>Trump does the same thing with different tools.</p><p>Twitter is not random. It’s not stream-of-consciousness policy. It’s narrative control. It defines what matters before institutions can react. It creates a shared language, shared enemies, shared victories that exist independent of whether anything actually changed.</p><p>People criticize Trump by saying he lies. But lying assumes persuasion is the goal.</p><p>The goal is alignment.</p><p>Jackson’s newspapers were not trying to convert Adams supporters. They were trying to synchronize Jackson supporters around a shared story about who the real enemy was.</p><p>Trump’s feed does the same thing.</p><p>This is why fact-checking so often fails. You’re answering a question no one asked.</p><p>⸻</p><p><strong>When Institutions Turn Into Villains</strong></p><p>Jackson’s great enemy is the Bank of the United States. Not because most people understand central banking, but because they understand the idea of a powerful unelected man controlling money.</p><p>Nicholas Biddle, the bank’s president, becomes a character. Not a policy. A villain.</p><p>Biddle was, by all accounts, competent. Educated. Thoughtful. Exactly the kind of credentialed expert you’d want running a complex financial institution, which is precisely why Jackson despised him. Biddle represented everything Jackson had learned to distrust: East Coast breeding, institutional authority, and the belief that some decisions were too important to be left to voters.</p><p>Jackson turned the bank recharter into a referendum on who should control the country’s future. He vetoed it in 1832 and used the veto message as a campaign document. It wasn’t a policy paper. It was a declaration of war.</p><p>This is the pattern.</p><p>Institutions become characters. Characters become enemies. Enemies become the entire story.</p><p>The Supreme Court becomes John Marshall.</p><p>The Bank becomes Nicholas Biddle.</p><p>“The media” becomes a single hostile force.</p><p>“The deep state” becomes a personality.</p><p>Once this happens, compromise becomes betrayal and restraint becomes weakness.</p><p>This isn’t because people are stupid. It’s because abstraction feels dishonest when your life feels unstable. A complex system is hard to fight. A villain with a name and a face is not.</p><p>Jackson and Trump don’t create this instinct. They recognize it.</p><p>⸻</p><p><strong>The Coalition That Shouldn’t Work (But Always Does)</strong></p><p>Jackson’s coalition makes no sense on paper. Frontier farmers and plantation elites. Populist rhetoric and slaveholder economics.</p><p>Trump’s coalition doesn’t make sense either. Working-class voters and billionaires. Evangelicals and a thrice-married casino owner. Libertarians and tariff enthusiasts.</p><p>But these movements are not built on shared interests.</p><p>They’re built on shared enemies.</p><p>Each group believes it is getting something different: protection, disruption, revenge, entertainment, dignity. As long as the fight stays focused outward, the contradictions don’t matter.</p><p>They surface only after the leader exits the stage.</p><p>By then, the role is already being recast.</p><p>⸻</p><p><strong>The Pattern</strong></p><p>Once you stop looking for uniqueness and start looking for structure, the rhythm becomes obvious.</p><p>Economic stress. Not collapse, but a widespread sense that the rules keep changing and only insiders know what they are. That the system is working for someone, but not you.</p><p>Institutional distrust. The people in charge seem more interested in protecting each other than solving problems. They speak a language designed to obscure rather than clarify. They fail, and then explain why the failure was actually success, or why the metric was wrong, or why you just don’t understand how complex everything is.</p><p>Media disruption. New technology that lets someone speak past the gatekeepers directly to audiences in a way that feels more authentic than filtered. Jackson had newspapers loyal to him, not to “journalism.” FDR had radio. Trump had Twitter. The medium changes but the effect is the same: the ability to define reality before institutions can respond.</p><p>Geographic resentment. The sense that one part of the country has contempt for how another part lives. That the people making the rules have never been where you are and don’t care to understand it.</p><p>A leader who offers dominance instead of reassurance. Who doesn’t promise to fix the system but to humiliate the people running it. Who proves, over and over, that he can survive what would destroy a normal politician.</p><p>The details change.</p><p>The role doesn’t.</p><p>This isn’t unique to Jackson and Trump. Huey Long built a national following in the 1930s promising to make “every man a king” while running Louisiana like a personal fiefdom, using radio to bypass newspapers that wouldn’t cover him favorably. He was, by all accounts, corrupt, but he also built roads, bridges, and schools in parts of the state that had been ignored for generations. His coalition didn’t care about the corruption because he was fighting enemies they hated: Standard Oil, the utilities, the newspapers, the old money families who had controlled the state forever.</p><p>George Wallace stood in schoolhouse doors and won five states in 1968 by telling working-class voters that both parties had abandoned them for intellectual elites who cared more about abstract principles than their lives. His politics were poisoned by racism, but his reading of voter alienation was precise.</p><p>Ross Perot spent $60 million of his own money in 1992 and won 19% of the vote, the best third-party showing in 80 years, by arguing the entire political system was a rigged game run by people who’d never signed the front of a paycheck. He had charts. He had specifics. But what he really had was a message: the people running things don’t know what they’re doing, and I’m not one of them.</p><p>Each one was dismissed as a fluke, an aberration, proof that something had gone wrong.</p><p>But they’re not aberrations.</p><p>They’re auditions.</p><p>The role stays open because the conditions keep recurring.</p><p>⸻</p><p><strong>The Portrait, Revisited</strong></p><p>Go back to the Oval Office.</p><p>Trump stands there, pointing at Jackson, talking about how much Jackson hated the elites, how the system tried to stop him and failed.</p><p>This isn’t nostalgia.</p><p>It’s recognition.</p><p>Jackson isn’t there to soften the room.</p><p>He’s there as a warning label.</p><p>Not because the character was good or bad.</p><p>But because the character is repeatable.</p><p>If the conditions remain, the character returns.</p><p>America keeps casting this role.</p><p>And someone will always audition.​​​​​​​​​​​​​​​​</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4f0df4e4f3bd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>