<?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 Samuel Black on Medium]]></title>
        <description><![CDATA[Stories by Samuel Black on Medium]]></description>
        <link>https://medium.com/@bs.media?source=rss-29ba1f2c6fb0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*j2Has6iSLWp9Ew4X1T25CQ.png</url>
            <title>Stories by Samuel Black on Medium</title>
            <link>https://medium.com/@bs.media?source=rss-29ba1f2c6fb0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 07:11:00 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bs.media/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[“Like the Exodus from Egypt”: How Arkady Volozh Lost a $31bn Company and Built a $67bn One in Two…]]></title>
            <link>https://medium.com/@bs.media/like-the-exodus-from-egypt-how-arkady-volozh-lost-a-31bn-company-and-built-a-67bn-one-in-two-6a83a0a79718?source=rss-29ba1f2c6fb0------2</link>
            <guid isPermaLink="false">https://medium.com/p/6a83a0a79718</guid>
            <dc:creator><![CDATA[Samuel Black]]></dc:creator>
            <pubDate>Mon, 27 Jul 2026 13:12:42 GMT</pubDate>
            <atom:updated>2026-07-27T13:12:42.885Z</atom:updated>
            <content:encoded><![CDATA[<h3>“Like the Exodus from Egypt”: How Arkady Volozh Lost a $31bn Company and Built a $67bn One in Two Years</h3><h3>“Like the Exodus from Egypt”: How Arkady Volozh Lost a $31bn Company and Built a $67bn One in Two Years</h3><p>On 21 October 2024, trading resumed on Nasdaq in a security that had not moved for thirty-two months. The ticker had changed, the name had changed, the business had changed entirely. The only thing that had not changed was the founder.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*l4FlqzMvQPVqRQX7.jpg" /></figure><p>Nebius Group is today worth roughly $67bn, sells computing capacity to Microsoft and Meta under contracts totalling more than $46bn, and is growing faster than almost anything else listed in America.</p><p>Formally, this is not a story about a flotation. And that is the most interesting thing about it.</p><p><em>This is not an interview. Arkady Volozh did not answer the questions in this piece. All figures are taken from Nebius Group’s filings with the SEC and from market data; all of his words come from public appearances, shareholder letters and interviews, with the occasion noted. The awkward questions are the author’s own.</em></p><h3>Who is Arkady Volozh</h3><blockquote>Founder and chief executive of Nebius Group N.V. Born in 1964 in Guryev, now Atyrau, in Kazakhstan. Educated at the Gubkin Institute of Oil and Gas in Moscow. Founded CompTek in 1989 and, in 1997, Yandex — Russia’s largest search engine. Took the company to Nasdaq in 2011. Left all his posts in 2022 after the European Union imposed sanctions on him, and returned to run the Dutch parent company in 2024. Has lived in Israel since 2014 and holds Israeli citizenship; in February 2026, according to Bloomberg, he renounced his Russian citizenship.</blockquote><h3>Thirty-Two Months of Silence</h3><p>To understand the scale of what followed, the starting point has to be recorded precisely — and it was well below zero.</p><p>In November 2021, Yandex N.V. was worth more than $31bn on Nasdaq. In February 2022, after the war in Ukraine began, trading was suspended. Nasdaq announced its intention to delist the company along with other Russia-linked issuers; Yandex N.V. appealed and kept its listing — frozen, with no date attached.</p><p>In June 2022 Volozh was placed under EU sanctions and stepped down from every position he held. In August 2023 he issued a public statement declaring himself categorically opposed to Russia’s invasion of Ukraine. He later described hiring personal security for the first time in his life, and likened the experience of being sanctioned to imprisonment. In March 2024 the EU lifted the measures against him.</p><p>Negotiations over the split ran in parallel. In February 2024 the deal was announced: all of Yandex’s Russian assets would go to a consortium of Russian investors for 475bn roubles — about $5.2bn, roughly half the market valuation, the discount effectively compulsory for anyone leaving. The transaction closed in July 2024 and stands as the largest corporate exit from Russia since the invasion. The Dutch holding company, Yandex N.V., was renamed Nebius Group.</p><p><em>The arithmetic is worth stating plainly, because it explains everything about the founder’s subsequent behaviour. A company worth $31bn was broken up, and the larger part of it — the Russian business — went for something close to a fifth of its former valuation. Speaking to Bloomberg, Volozh compared what had happened to the Exodus from Egypt. The metaphor sounds grandiose right up until you look at the figure he walked away with.</em></p><h3>Nebius Group by the Numbers</h3><ul><li><strong>~$67bn</strong> market capitalisation as of 22 July 2026; a 52-week range running from <strong>$49</strong> to <strong>$299.86</strong> a share</li><li><strong>$399m</strong> in first-quarter 2026 revenue — up <strong>684 per cent</strong> year on year</li><li><strong>$1.92bn</strong> in annualised run-rate revenue at the end of the quarter; guidance for the end of 2026 is <strong>$7bn–$9bn</strong></li><li><strong>$529.8m</strong> in revenue for the whole of 2025, against <strong>$117.5m</strong> for 2024</li><li><strong>More than $46bn</strong> contracted with Microsoft and Meta over five years</li><li><strong>$4.8bn</strong> already collected in customer prepayments, of which $4.1bn is for services to be delivered in 2027 and 2028</li><li><strong>$20bn–$25bn</strong> of capital expenditure planned for 2026 alone</li><li><strong>3.5 GW</strong> of contracted power, with a target of at least <strong>4 GW</strong> by year end; more than <strong>75 per cent</strong> of it in owned rather than leased facilities</li><li><strong>9.3 per cent</strong> of Nebius disclosed as held by Nvidia in July 2026</li></ul><blockquote><em>Source: Nebius Group filings and market data, July 2026.</em></blockquote><h3>What Was Left in the Box</h3><p><strong>What did the company actually start with?</strong></p><p>With a collection of assets that no investment banker in 2022 would have described as the foundation of a hyperscaler.</p><p>After the separation, the Dutch holding company retained: a data centre in the Finnish town of Mäntsälä; the Nebius cloud division; Toloka, a data-labelling service; Avride, the autonomous-vehicle unit formerly known as Yandex Self-Driving Group; TripleTen, an edtech platform; around a thousand employees relocated out of Russia; and roughly $2.5bn in cash from the sale of the Russian business.</p><p>No brand, no audience, no market, no customers. But a Finnish data centre, money, and engineers who had already built a cloud from nothing once before.</p><p><em>It is worth calling this what it was: not a strategy but a residue. The strategy came afterwards, and consisted of selecting one item from that residue and deciding it was now the company. In 2022 the Finnish data centre was peripheral infrastructure attached to a search business. By 2024 it was the thing the world had begun paying almost anything for.</em></p><h3>Not a Flotation but a Return</h3><p><strong>Why does this not count as an IPO — and why does that matter?</strong></p><p>Because the listing never lapsed.</p><p>Nasdaq suspended trading but, unlike the other Russian issuers, did not delist the company. The legal entity — incorporated in Amsterdam in 2007, listed since 2011 — carried on existing, complete with its reporting history, its auditors, its corporate governance and, crucially, its existing base of Western shareholders.</p><p>So when trading resumed on 21 October 2024, what actually happened was this: a company with a new business, a new name and, in substance, the annual revenue of a middling Series B arrived on the public market without going through a single step of a flotation. No roadshow, no underwriters, no lock-up, no two years of preparing accounts to SEC standards.</p><blockquote><strong>A start-up with $120m of run-rate revenue inherited the public status of a company that had once been worth $31bn</strong></blockquote><p>What that bought in practice: Nebius could raise capital through share placements, convertible bonds and secured debt immediately — precisely as money began pouring into AI infrastructure, rather than two or three years later. In a business where everything turns on how fast you can buy capacity, those two or three years are worth more than any amount of money.</p><p><em>This is the most consistently underrated part of the story. It is usually told as an entrepreneur starting from scratch and building again. He did not start from scratch. He started from a highly unusual combination: the assets of an early-stage start-up housed inside the corporate shell of a mature public company. Nobody assembles that deliberately. It only occurs as the by-product of a catastrophe.</em></p><h3>The One Thing You Cannot Buy Quickly</h3><p><strong>What is the bet?</strong></p><p>Electricity.</p><p>The industry’s public explanation runs as follows: the scarce resource in AI is graphics processors. At the level of operational reality that stopped being true somewhere around 2025. Chips can be ordered, queued for and received. What cannot be obtained quickly is everything else: a site, connected megawatts, permits, transformers and cooling.</p><p>Nebius built its strategy around power from the outset — the Finnish data centre was, above all, a site with electricity rather than a warehouse full of hardware. Kansas City followed, then the United Kingdom, then Israel, and finally a 1.2 GW facility in Pennsylvania announced in May 2026.</p><p>By the first quarter of 2026 the company had contracted 3.5 GW of power, running ahead of its own full-year plan, and raised the target to four. More than three-quarters of that sits in owned facilities rather than capacity leased from third-party data centre operators.</p><p>The distinction between owned and leased is not an accounting one. A tenant pays the operator a margin and lives by the operator’s build schedule. An owner pays once, in capital expenditure, and keeps the margin — at the price of having to find the capital.</p><blockquote><strong>A company with connected megawatts is selling a scarcity. A company with only chips is selling a commodity</strong></blockquote><h3>Selling What Does Not Yet Exist</h3><p><strong>How does a build-out on this scale get financed?</strong></p><p>By customers, in advance.</p><p>The model Nebius has arrived at works like this. A multi-year contract is signed with a large customer for capacity that will come online in twelve to eighteen months. Capital is then raised against that contract. The facility is then built with that capital.</p><p>In September 2025 a contract with Microsoft worth up to $19.4bn over five years was signed. In March 2026 came a deal with Meta worth up to $27bn, comprising $12bn of dedicated compute on Nvidia’s Vera Rubin platform and $15bn of committed capacity purchases. Between those two customers alone, the contracted book passed $46bn.</p><p>The customers pay up front. By the end of the first quarter of 2026 the company had collected $4.8bn in prepayments, of which $4.1bn relates to services to be delivered in 2027 and 2028. On the quarterly call, chief financial officer Dado Alonso summarised the state of demand briefly: capacity had sold out once again, with demand continuing to exceed available supply.</p><p>Volozh, on the same call, explained why the company had raised its 2026 capital expenditure plan from $16bn–$20bn to $20bn–$25bn: everything they build, they sell, and it is still very early days.</p><p><em>A structure of this kind — prepayment for capacity that does not physically exist — is only possible in one circumstance: when there are more buyers than sellers and everybody knows it. It is also the principal risk. The contracts run five years forward; the capital against them is being raised today.</em></p><h3>The Financing Stack</h3><p><strong>Where does $25bn come from?</strong></p><p>From a sequence of progressively more sophisticated instruments — and this is a part of the strategy that rarely gets noticed.</p><p>The order ran as follows. December 2024: a $700m private placement with Accel, Nvidia and Orbis participating. Mid-2025: convertible bonds. Autumn 2025: a large equity raise. March 2026: a direct $2bn investment from Nvidia, carrying priority in the queue for next-generation chips. July 2026: the company’s first secured credit facility, at roughly $775m.</p><p>That last item matters more than its size suggests. Moving from equity funding to secured debt means lenders have agreed to treat Nebius’s data centres as collateral with a predictable cash flow attached. For an infrastructure business that is a change of weight class: equity is expensive and dilutes the founder, secured debt is cheap and does not.</p><p>On 20 July 2026 Nvidia disclosed a 9.3 per cent holding in Nebius. The shares rose about 19 per cent that day.</p><h3>What He Actually Took From Yandex</h3><p><strong>Which part of the old company proved transferable?</strong></p><p>Not the technology and not the brand. The experience of one particular class of problem.</p><p>Yandex was one of very few companies anywhere to have built a complete search infrastructure — its own data centres, its own distributed stack, its own cloud — in a market where Google stood opposite it with an incomparable budget. That is not a story about inventing an algorithm. It is a story about twenty years of building and running physical computing infrastructure under a permanent shortage of capital.</p><p>That competence turned out to be exactly the scarce one in 2024. Companies that can train models are now numerous. Companies with a team that has spent thirty years building data centres and knows how to do it cheaply are not.</p><p><em>This also answers a question that comes up often: why Nebius’s margins are improving faster than comparable operators’. Gross margin rose from 52 per cent in 2024 to nearly 69 per cent by the end of 2025, and the cloud division’s adjusted EBITDA margin reached 45 per cent in the first quarter of 2026, against 24 per cent the quarter before. That is not a pricing effect. It is the effect of infrastructure being designed by people who have done it before.</em></p><h3>Where It Is Thin</h3><p><strong>Which of the risks here are real rather than decorative?</strong></p><p><strong>Capital expenditure against revenue.</strong> The company plans to spend $20bn–$25bn in 2026 against expected revenue of roughly $3.2bn–$3.4bn. The gap is bridged with debt and equity raised against contracts that only begin to perform in 2027. Any slippage in the build schedule hits both sides of that equation at once.</p><p><strong>Customer concentration.</strong> The overwhelming majority of the contracted book belongs to two names, Microsoft and Meta. Both are simultaneously building their own infrastructure. Nebius is, for them, a way of acquiring capacity faster than they can construct it; the moment their own construction catches up with demand, the supplier’s position changes.</p><p><strong>Nvidia’s double role.</strong> Nvidia is at once the critical supplier, a shareholder with 9.3 per cent, and — through its investments elsewhere in the sector — a participant in flows that run in both directions. For Nebius this means access to scarce chips outside the queue. It also means dependence on a counterparty free to reorder its priorities.</p><p><strong>Governance.</strong> A dual-class share structure gives the founder voting control substantially in excess of his economic interest; Bloomberg puts the Volozh family’s holding at around 13 per cent of the company. Nebius formally qualifies as a controlled company under Nasdaq rules, with reduced board-independence requirements. In its annual report filed in April 2026, the company disclosed material weaknesses in internal controls over property, plant and equipment accounting and over revenue recognition at the TripleTen subsidiary, with remediation targeted for the end of 2026.</p><p><strong>Volatility.</strong> The share price has traded between $49 and nearly $300 over the past twelve months. That is not a market pricing a business. It is a market pricing a hypothesis about a business, and it changes its mind by tens of per cent within weeks.</p><p><em>It should be said honestly that none of these risks is an argument against the company. All of them are a description of what an investor is buying. Nebius is a bet that demand for computation keeps outrunning the industry’s ability to build capacity, and that the company delivers what it has contracted on time. Both halves are confirmed by every quarter so far. Both remain hypotheses.</em></p><h3>By Way of an Epilogue</h3><p>In the thirty-two months between February 2022 and October 2024, Arkady Volozh lost the company he had spent twenty-seven years building, was sanctioned and then unsanctioned, sold the Russian business at half its market value, lost his Russian citizenship, and emerged with a Finnish data centre, a thousand engineers and $2.5bn.</p><p>What he did next is rarer than building a unicorn from nothing: he took the shell of a destroyed company and filled it with a different business, quickly enough that the market never quite registered it was looking at a start-up and priced it as an infrastructure operator instead.</p><p>Of the Exodus, the source text records that the journey took forty years and that not everyone who set out arrived. This one took two, and on the figures it is going rather better than the original. Nobody, of course, is promising it will stay that way — and in this industry, it seems, nobody is supposed to promise anything.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6a83a0a79718" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[“Thirty Days from Going Out of Business”: How the Man Selling Shovels Became Worth More Than the…]]></title>
            <link>https://medium.com/@bs.media/thirty-days-from-going-out-of-business-how-the-man-selling-shovels-became-worth-more-than-the-200efe75dd9f?source=rss-29ba1f2c6fb0------2</link>
            <guid isPermaLink="false">https://medium.com/p/200efe75dd9f</guid>
            <dc:creator><![CDATA[Samuel Black]]></dc:creator>
            <pubDate>Thu, 23 Jul 2026 07:12:54 GMT</pubDate>
            <atom:updated>2026-07-23T07:12:54.088Z</atom:updated>
            <content:encoded><![CDATA[<h3>“Thirty Days from Going Out of Business”: How the Man Selling Shovels Became Worth More Than the Gold Rush</h3><p>In every gold rush, the reliable money is in shovels. Sam Brannan understood this in 1848: before announcing to San Francisco that gold had been found at Sutter’s Mill, he quietly bought every pick, pan and shovel in the territory. He became California’s first millionaire without ever kneeling in a river.</p><p>Nvidia is the most extreme version of that trade in commercial history — a company that sells the equipment for a rush and is now worth roughly five trillion dollars, more than the combined market value of most of the companies doing the actual digging.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*DBZSp0ycKpad1YcrbLb5tw.jpeg" /></figure><p>It is a tidy story. It is also, in three specific places, wrong — and those three places are the whole point.</p><p><em>This is not an interview. Jensen Huang did not answer the questions in this piece. All figures come from Nvidia’s filings and market data; all of his words are drawn from public keynotes and appearances, with the occasion noted. The awkward questions are the author’s own, and they are answered with things that can be checked.</em></p><h3>Who is Jensen Huang</h3><blockquote>Founder and chief executive of Nvidia. Born in Tainan, Taiwan, in 1963; sent by his parents to the United States at nine and educated at a boarding school in Kentucky. BSc in electrical engineering from Oregon State, MSc from Stanford. Worked as an engineer at LSI Logic and AMD. Founded Nvidia in 1993 at the age of thirty and has never left the chief executive’s chair since — an almost unique tenure for a technology company of this size.</blockquote><h3>Making Shovels Before Anyone Found Gold</h3><p>The company was founded in April 1993 in a Denny’s diner in San Jose — the same chain where Huang had bussed tables and washed dishes as a teenager. Three engineers, one technical proposition: the central processor is bad at three-dimensional graphics, so build a separate chip that does only that, and does it extremely fast.</p><p>There was no market for such a chip. There was a hypothesis that the personal computer would eventually become a machine for entertainment rather than spreadsheets. The bet was on the hypothesis.</p><p>The first product, the NV1, shipped in 1995 and failed outright. Not for want of engineering: NV1 built images out of curved surfaces, which was elegant. Microsoft shipped Direct3D, which built everything out of triangles, which was not elegant but was a standard. The industry followed the standard. Nvidia had spent everything building a tool that spoke a language nobody bothered to learn.</p><p>Headcount was cut from around a hundred to roughly forty. There was money for a few months. The next chip, the RIVA 128, was developed essentially blind — simulated in software, with no budget for test silicon and exactly one attempt at getting it right. It shipped in 1997, sold a million units in four months, and saved the company.</p><p><em>Out of that near-death came the line Huang reportedly opened internal meetings with for years.</em></p><blockquote><strong>“Our company is thirty days from going out of business”</strong></blockquote><p>It has been factually untrue for about two decades. It works anyway, as a permanently illuminated warning light: an organisation that believes it is a month from death does not spend the month arguing about slide decks.</p><h3>Nvidia by the Numbers</h3><ul><li><strong>33 years</strong> on the market; headquarters in Santa Clara, California</li><li><strong>~$5.0–5.1 trillion</strong> market capitalisation as of 22 July 2026 — the most valuable company in the world, roughly a trillion clear of the runner-up</li><li><strong>$215.9 billion</strong> in revenue for fiscal 2026, up 65% year on year</li><li><strong>$81.6 billion</strong> in a single quarter (Q1 FY2027), of which <strong>$75.2 billion</strong> came from data centres — up 92% year on year</li><li><strong>~75%</strong> gross margin — the margin profile of a software company, not a hardware manufacturer</li><li><strong>$1 trillion</strong> in cumulative orders for the Blackwell and Rubin platforms through 2027, on the company’s own projection</li><li><strong>~40,000</strong> employees, effectively all of whom, by Huang’s account, worked on the Vera Rubin platform</li><li><strong>$119 billion</strong> in supply commitments — the price of buying manufacturing capacity years before the demand contracts exist</li><li><strong>$0</strong> in data-centre Hopper shipments to China last quarter, against $4.6 billion a year earlier</li></ul><blockquote><em>Source: Nvidia filings and market data, July 2026.</em></blockquote><h3>First, Convince People a Shovel Is a Thing</h3><p>Selling equipment into a rush requires the equipment to have a name. In October 1999 Nvidia shipped the GeForce 256 and attached to it a term that had not previously existed in the market: GPU, the graphics processing unit.</p><p>Technically this was a chip with hardware transform and lighting. Strategically it was the announcement of a product category in which Nvidia was automatically ranked first, on the grounds that it was the only occupant.</p><p>By 2000 the competition was thinning: the assets of 3dfx, the great rival of the 1990s, were simply bought. By the middle of the decade, discrete graphics was a two-horse race.</p><p><em>Nvidia has run this move at roughly ten-year intervals ever since: “GPU” in 1999, “accelerated computing” in 2007, “AI factory” in 2024. Each time the company declines to offer a better product on an existing shelf and instead builds a new shelf with its own name printed on the front. Category creation is the cheapest form of market share ever invented, and almost nobody attempts it.</em></p><h3>Ten Years of Forging With No Ore in Sight</h3><p>Then comes the decision that breaks the shovel-seller analogy for the first time. Brannan bought his inventory a few days before the rush. Nvidia started building for a rush a decade before anyone had confirmation there was gold in the ground.</p><p>In 2006 and 2007 the company did something that, read off a quarterly income statement, looks like deliberate self-harm. It redesigned its architecture so that the chips could run arbitrary computation rather than only graphics, and released a free software layer, CUDA, so that people could write such code. The silicon got more expensive: transistors spent on programmability improved not one frame of one game. Gamers were paying for a feature they did not use.</p><p>There was no market for that feature. There was a proposition that one would appear — in physics, chemistry, seismic imaging, financial modelling, anywhere a calculation decomposes into thousands of identical independent pieces.</p><p>Nvidia funded this for roughly ten years. Analysts asked, with some regularity, when the company would stop financing a division with no revenue attached. The company carried on sending engineers to universities, donating cards to laboratories, and writing libraries for disciplines whose total addressable market ran to millions rather than billions.</p><blockquote><strong>The moat around Nvidia took ten years and billions of dollars to dig — and it is not made of silicon</strong></blockquote><p><em>This is the part usually skipped in accounts of the AI boom. Nvidia’s competitors are not primarily behind on chips. They are behind on those ten years. Building a comparable accelerator is hard but achievable; several companies have done it. Reproducing fifteen years of accumulated libraries, frameworks, university courses and several million developers who know how to write CUDA and do not know how to write anything else is a problem of an entirely different order. The shovel is copyable. The fact that every miner in the territory learned to dig with that particular handle is not.</em></p><h3>The Assay Report</h3><p>Confirmation arrived in 2012, from a direction nobody at Nvidia had planned for.</p><p>Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton trained a neural network called AlexNet, which won the ImageNet image-recognition competition by a humiliating margin. They trained it on two consumer GeForce GTX 580 cards bought off a shelf.</p><p>Two lines converged at that moment: a scientific hypothesis about deep learning that had been considered a dead end for forty years, and a commercial hypothesis about programmable GPUs that had been considered an expensive indulgence for ten. It turned out that neural networks were precisely the class of problem Nvidia had spent a decade building for, without knowing anything about neural networks.</p><p>After that the company moved quickly. In 2016 Huang personally hand-delivered the first DGX-1 supercomputer to a then-obscure non-profit laboratory called OpenAI. In 2017 the chips gained tensor cores, blocks designed specifically for the matrix operations of deep learning. By November 2022, when ChatGPT shipped and the world abruptly decided it wanted computation, Nvidia had been ready for five years — product, infrastructure, supply chain and all.</p><p><em>This is usually narrated as luck. That is imprecise. The luck was that the revolution happened in deep learning rather than in computational chemistry. What was not luck was that whichever revolution arrived, Nvidia would be the only company with a finished platform waiting for it.</em></p><h3>From Selling Shovels to Selling the Mine</h3><p>Here the analogy breaks a second time — and this is the strategic move of the past five years.</p><p>The logic runs like this. While you sell a component, you are a supplier to whoever assembles the system, and your margin is capped by their willingness to pay. The moment the bottleneck stops being the chip and becomes the links between chips, the power delivery and the cooling, it becomes more profitable to design the whole system and sell the whole system.</p><p>Hence the 2019 acquisition of Mellanox, the Israeli InfiniBand networking company, for $6.9 billion. The deal was widely described at the time as puzzling. In the most recent quarter, networking brought in roughly $11 billion — more than Nvidia’s total revenue in any year before 2020.</p><p>Then came its own central processor (Grace, now Vera), its own rack, its own liquid cooling, its own software stack. Nvidia today does not sell an accelerator. It sells what it calls an AI factory: an assembled, configured, working computational plant that the customer connects to the electricity grid.</p><p>The Vera Rubin platform, unveiled at GTC in March 2026 and in full production by June, comprises seven new chips built on TSMC’s 3nm process, and delivers, by the company’s account, an order-of-magnitude reduction in cost per token against the previous generation. At the Taipei keynote Huang summarised the transition in his own compressed idiom: useful AI, he said, has arrived — and AI now makes money.</p><blockquote><strong>The equipment supplier who owns the mine, the rail line and the assay office is no longer an equipment supplier</strong></blockquote><h3>Shipping Faster Than the Shovel Can Be Copied</h3><p>The current strategy is, more than anything, a matter of tempo. Nvidia has moved to a new architecture every year instead of the industry-standard two: Hopper, Blackwell, Rubin, then Rubin Ultra. Each generation is not merely a new GPU but a new processor, a new network, a new rack and a new software stack.</p><p>The speed is chosen deliberately and functions as a weapon. A competitor catching up with the previous generation ships its product against the next one. The pursuer stays a generation behind not because it runs slowly but because the target keeps accelerating.</p><p>The price of that tempo is written into the balance sheet as commitments. At the end of the last quarter, Nvidia’s obligations for manufacturing capacity and components stood at $119 billion. That is a wager: the company buys years of TSMC and memory-maker capacity in advance, against demand that has not yet turned into signed contracts. If the demand does not arrive, the obligations remain.</p><p>So far it arrives. Revenue for the quarter ending April 2026 came in at $81.6 billion, with guidance of $91 billion for the next. Cumulative orders for Blackwell and Rubin through 2027 are projected by the company at a trillion dollars — double the figure it gave a year earlier.</p><h3>Fifty Direct Reports and No One-on-Ones</h3><p>The management model deserves more attention than it usually gets, because it is genuinely unorthodox.</p><p>Huang does not run a conventional pyramid: the number of people reporting directly to him has sat around fifty for years. He has said publicly that he holds no one-on-one meetings and writes no individual development plans — strategic information is delivered to everyone at once, in the room. The reasoning is austere: if something matters to only one person, it is not strategy, it is operations.</p><p>The company has no formal strategy document either. Huang has described a system in which employees at any level send him short notes on the five things they are working on, and he reads something on the order of a hundred a day — a direct feed from the floor that bypasses the managerial filters entirely.</p><p>Failure is handled distinctively. Huang is known for almost never firing people over failed projects. NV1 failed. Tegra never took the smartphone market and was rebuilt into an automotive platform. The $40 billion acquisition of Arm collapsed under regulatory pressure in 2022. The teams stayed. Speaking at Stanford, he put his position in terms that promptly went around the world: greatness does not come without pain, and he sincerely wished his audience plenty of it.</p><p><em>It is worth recording honestly that this model works for Huang and may not be portable. A flat structure with fifty direct reports requires a chief executive who holds the entire technical stack in his head and can argue any of the fifty to a standstill on the merits. Such a person is not appointed; he either exists or he does not. Nvidia carries one of the sharpest unanswered succession questions in global business, and there is no answer to it today.</em></p><h3>Where the Analogy Breaks for the Third Time</h3><p>This is the section the anniversary pieces leave out.</p><p><strong>Customer concentration.</strong> The largest buyer category — the hyperscalers — accounts for somewhat over half of data-centre revenue. That is a handful of companies, every one of which is simultaneously developing its own accelerators specifically to reduce dependence on Nvidia. Their capital expenditure is both Nvidia’s revenue and its single point of failure.</p><p><strong>China.</strong> Data-centre Hopper shipments to China last quarter were zero, against $4.6 billion a year earlier. The company no longer includes Chinese demand in its outlook at all. A market that was central has been struck out by decisions Nvidia neither made nor can influence.</p><p><strong>Money going in circles.</strong> A portion of demand comes from companies Nvidia itself has invested in. How much of the revenue is organic demand and how much is its own capital returning through the till is a substantive dispute among analysts, and no precise answer exists from outside. Note what this does to the metaphor: the classic shovel seller stays solvent precisely because he does not own claims. Nvidia owns claims.</p><p><strong>Cyclicality.</strong> Nvidia has already lived through a demand collapse — in 2018, when the crypto-mining boom ended and the channel was left full of unwanted cards. The difference between that episode and this one is the scale of the commitments: then it was billions, now it is a hundred and nineteen of them.</p><p><em>None of this is a forecast of decline. It is a description of the structure. A five-trillion-dollar company rests on the proposition that a small number of the world’s largest buyers will keep increasing capital expenditure every year. So far each quarter confirms it. It remains a proposition.</em></p><h3>By Way of an Epilogue</h3><p>Sam Brannan, the man who cornered the shovel market before the rush he himself announced, was California’s first millionaire and died a great deal poorer than he had been. Selling equipment into a boom is the safest position in the boom, right up to the moment the boom is over — because the shopkeeper’s fortune is not independent of the rush. It is entirely made of it.</p><p>Nvidia’s answer to that, thirty-three years in, has been consistent enough to look like a doctrine: build the equipment for the rush before the rush, and once it starts, stop being a shopkeeper. Sell the tool, then the system, then the plant, then the standard everybody has learned. Each step further from the shovel and closer to the ground under it.</p><p>Sometimes the rush never comes; that was mobile processors. Sometimes it arrives a decade late; that was CUDA. A company that can afford to wait ten years beats a company that has to report in ninety days — on the sole condition that it survives the ten years.</p><p>“Thirty days from going out of business,” Huang told his people, for years. On the evidence, those thirty days have now been running for thirty-three years. That appears to be the whole trick.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=200efe75dd9f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[“We Don’t Promise Anyone Anything”: How a Cypriot Family Office Grew to $50 Million Under…]]></title>
            <link>https://medium.com/@bs.media/we-dont-promise-anyone-anything-how-a-cypriot-family-office-grew-to-50-million-under-dd8083a7f514?source=rss-29ba1f2c6fb0------2</link>
            <guid isPermaLink="false">https://medium.com/p/dd8083a7f514</guid>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[family-office]]></category>
            <category><![CDATA[investment]]></category>
            <dc:creator><![CDATA[Samuel Black]]></dc:creator>
            <pubDate>Wed, 22 Jul 2026 15:12:32 GMT</pubDate>
            <atom:updated>2026-07-23T13:13:46.198Z</atom:updated>
            <content:encoded><![CDATA[<h3>“We Don’t Promise Anyone Anything”: How a Cypriot Family Office Grew to $50 Million Under Management</h3><p>Five years ago Alexey Murashev set up a company with a single aim — to manage his own family’s money rather than hand it to outside managers. Today AMCH has more than $50 million in assets under management and over fifty staff, and roughly ten times as many people want to become clients as the firm is willing to take on. In this interview, AMCH’s founder talks about the first money he managed, the deal that burned to the ground, and why his company refuses, on principle, to make promises.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Gp0jHBLs4UPsqX9kMKjmWg.png" /></figure><p>The occasion for this conversation is a number. By the standards of the global wealth-management industry, $50 million is modest. By the standards of a company that began as a family office for one family’s money, and grew in years when investment firms with Russian-speaking roots were closing more often than they opened, it is a number worth talking about.</p><p><a href="https://amch.ltd/">AMCH</a> is a private company: it is under no obligation to disclose figures or to talk about itself, which makes it all the more interesting that it agreed to this conversation without preconditions. All the figures in this piece come from the company itself; where they cannot be verified independently, that is stated plainly in the text. One thing we settled at the outset: no question would be off limits.</p><h3>Who is Alexey Murashev</h3><blockquote><em>Founder and CEO of the investment company AMCH. Master’s in Finance (London Business School); MBA and DBA (Montreux, Switzerland). He founded a family office to manage his own family’s capital, which over five years grew into a boutique investment company headquartered in Cyprus, specialising in venture and pre-IPO deals.</em></blockquote><h3>Family Money</h3><p><strong>Let’s start at the beginning. At what point did you realise you wanted to manage capital yourself?</strong></p><p>The decision to set up a company never arrives in a single day. I’ll be honest: I had no intention whatsoever of building the company that exists today. Everything it has become grew out of an idea that succeeded within a very narrow circle — my own family.</p><p>The original aim was simple: to be able to manage my own capital without bringing in outside experts. For years I had watched closely what everyone else was doing — investment houses, funds, private managers: what ideas they were offering, what results they were producing. And, frankly, none of it satisfied me.</p><p><strong>For example?</strong></p><p>Take Tiger Global. An excellent investment fund, invests in venture. But from what I have observed, the average return their client sees year after year is somewhere around twenty per cent. Twenty per cent — with venture risk attached. The game always has to be worth the candle: taking on serious risk only makes sense when the potential outcome covers that risk several times over.</p><p>Then I did two things. First, I analysed which areas could deliver the greatest return for me as an investor. The answer was venture investing. Second, I looked at which of the companies working in that area I would be prepared to entrust my family’s money to. And you know what? I found nothing. Nobody.</p><p>And after that — well, as the saying goes: if you want something done properly, do it yourself.</p><blockquote><strong><em>“If you want something done properly, do it yourself”</em></strong></blockquote><p><strong>Whose money was the first under management?</strong></p><p>The family’s. The company’s original purpose had nothing to do with attracting clients or managing client capital. My father is a wealthy man: he has spent more than thirty-six years in senior roles at PwC, one of the Big Four. Over those years he built up serious capital — and capital cannot sit idle. We settled on a strategy: to create our own family office. Venture became its principal focus. Everything else — the company as it looks today — grew out of that.</p><p><em>A family office is a structure a wealthy family sets up to manage its own capital instead of handing it to an outside manager. The central argument for the format is simple: the manager and the owner of the capital are the same people, so their interests align by definition. AMCH travelled the road in reverse: not a management company that worked its way up to wealthy families, but one wealthy family whose in-house instrument acquired outside clients.</em></p><h3>AMCH by the Numbers</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VP8MjkovSOxrB9KjjYVM3Q.jpeg" /></figure><ul><li>Around <strong>5 years</strong> on the market; parent company headquartered in Limassol, Cyprus</li><li>More than <strong>$50 million</strong> in assets under management</li><li><strong>$19 million</strong> in shareholders’ capital, up from $5 million in four years</li><li>Around <strong>$30 million</strong> in client assets</li><li><strong>~600</strong> active clients out of <strong>~5,000</strong> applications received</li><li>More than <strong>50</strong> staff and a presence in three regions; Kazakhstan is next on the list</li></ul><blockquote>Figures supplied by the company.</blockquote><h3>From Dubai to Limassol</h3><p><strong>Why Cyprus rather than another jurisdiction?</strong></p><p>Cyprus wasn’t the first choice. The company was born in the United Arab Emirates. From the outside, Dubai is a beautiful, glittering, very “investor-facing” shop window. But once you are up to your neck in the business processes inside it, you discover that it is an expensive and awkward jurisdiction — and that there is a wide gap between the façade and the actual operational work.</p><p>We spent more than two years working in the Emirates — long enough to learn that in practice rather than in theory. And a strategic decision was taken: to move the parent company to European Cyprus, with a view to scaling the business further. Cyprus today is one of the most favourable jurisdictions for both Russian and Western capital. Look at the serious players: every significant name with Russian roots has a Cypriot arm. We had no wish to be the exception.</p><h3>Three Degrees</h3><p><em>Alexey’s CV carries three degrees from two countries. In an industry where a fair number of successful investors take pride in having dropped out of university, that is close to provocative.</em></p><p><strong>What part of your education has genuinely been useful in the job, and what turned out to be a decorative line on the CV?</strong></p><p>The first degree was London Business School, a master’s in finance. That is what I use every single day: it allows me to read and understand the accounts of the companies we invest in. Venture investing is built largely on fundamental analysis, and the ability to value a business correctly is probably the whole basis of our success.</p><p>After that came Switzerland, the town of Montreux — an MBA. That knowledge is what allows me to run a company with more than fifty employees operating across three regions — and we are now planning a fourth, Kazakhstan. The third degree is a DBA, a doctorate in business administration, also in Switzerland.</p><p><strong>And what did the doctorate give you?</strong></p><p>Honestly? I still don’t know exactly what it has been useful for. (Laughs.) But taken together, all of it gives you an overall vision — a sense of where you ought to be steering the business. Managers handle operations; that is what they are there for. The most important quality in a leader is being able to see where the business should be sailing.</p><h3>The First One Burned</h3><p><em>Alexey talks about education easily — it is his favourite ground. The next question is intended to knock him off the well-worn track. It doesn’t work straight away: first he rewrites the question itself.</em></p><p><strong>Which deal taught you the most — and was it a profitable story or a losing one?</strong></p><p>In my view that is the wrong question. The right one would be: “Tell me about your most memorable deal.” So that is the one I’ll tell you about.</p><p>It was YallaMarket, a Dubai startup doing fifteen-minute delivery in the Emirates. One of the very first deals we invested in. And here is the interesting part: that deal burned to the ground. We lost everything in it.</p><p><strong>Everything?</strong></p><p>Everything we put in. And for me it was a signal. Everything up to that point had been theory on paper. Now you have landed in the mincer, and here life does not forgive mistakes. School was over; real money and real numbers had begun. Time to take off the rose-tinted spectacles and learn to invest for real.</p><p>The first pancake always comes out lumpy, as we say. But it was a good lesson. It showed me that investing is not for the faint-hearted, that in investing there are no accidents and no miracles. After that deal I pulled myself together completely, and ever since I have tried not to repeat the mistakes made in that investment.</p><blockquote><strong><em>“In investing there are no accidents and no miracles”</em></strong></blockquote><p><em>In the whole conversation this is the only episode in which Alexey describes an outright failure — with no caveats and no softening. And it is probably here that you believe him most.</em></p><h3>An App Instead of a PDF</h3><p><em>The traditional wealth-management industry has a strange relationship with technology: millions under management, and a quarterly report emailed to the client as a PDF.</em></p><p><strong>Most managers still get by with a quarterly PDF. Why did you put money into a mobile app?</strong></p><p>In 2021, when I was doing my analysis, there were not many companies offering access to venture — and they all shared one big problem: everything they did existed in a sort of “paper” format. Yet what matters enormously to every client is having all their assets to hand, in their phone. Any broker gives you an app for exchange-traded assets: two taps and you can see your portfolio. But here — off-exchange positions, pre-IPO, venture deals — and nobody had so much as a basic <a href="https://amcapital.app/">mobile app</a>.</p><p>So I thought: I would like to see all my own deals in a convenient mobile app. Why, I wondered, doesn’t one exist? Well then, I’ll build one, and I’ll be first. The most important part is that we were not building it for other clients. We built it for ourselves, so that my family would find it easy to use. And four years on, that bet is precisely what gave us an enormous boost when it came to scaling.</p><p><strong>What does a client see in the app today?</strong></p><p>Their entire portfolio within the fund: off-exchange positions, pre-IPO and venture deals — presented the way you are used to seeing an equity portfolio in any brokerage app.</p><p><em>The app carries the same name as the company — AMCH — and has recently been through a comprehensive redesign.</em></p><h3>Six Hundred Out of Five Thousand</h3><p><strong>Let’s come to the number we met to discuss. Is $50 million under management a lot or a little for your segment?</strong></p><p>It depends which way you look at it. Set against American funds with ten-year histories, it is a drop in the ocean. But if you take a company that started life as a family office — with a family office’s objectives — and has existed for a little over five years, then $50 million is already a fairly substantial figure.</p><p>And it matters how those fifty million became fifty million. The company’s shareholders’ capital, which today stands at $19 million, grew from $5 million — in four years. The rest, around $30 million, is client assets. And let us not forget: the company grew in the most difficult of times, when many players in the investment market were shutting up shop and leaving Russia. We found a way to come into being — and, most importantly, we survived and grew to our present figures. I am proud of that number.</p><p><strong>Then here is the awkward question from my list: what return do you not promise clients?</strong></p><p>At first I assumed that was a typo and you meant to ask what return we do promise. I’ll answer both versions at once: we promise nothing. To anyone.</p><p>Let me give you an example. Imagine you have found an athlete who has won the same competition ten years running. Lined up beside him are nine newcomers — no experience, no track record. You have to place a bet. Who do you back? The whole difficulty lies in finding that particular athlete — not among ten, but among millions. That is what we specialise in. But note this: even once we have found him and put our own money behind him, we do not promise that he will win an eleventh time. Guarantees do not exist here — and anyone offering you one should put you on your guard. That is why we promise nothing.</p><p><em>“We promise nothing”, coming from a manager, sounds paradoxical only at first hearing. In a regulated industry, promising a return is the shortest route to trouble with the regulator and the surest sign that the person in front of you is not a manager at all. Refusing to make promises is less a philosophical position than a hygienic one; the only question is what the firm offers instead. Alexey’s answer: its own money in every deal.</em></p><p><strong>Where are you objectively weaker than the big players?</strong></p><p>Honestly? I’m not sure I know. We are not chasing client acquisition, and we are not competing with other investment houses on assets. We position ourselves as a boutique — for a narrow circle of people who genuinely need us. Take Freedom Finance: a first-rate investment house, strong products, an enormous number of clients. But they are not a competitor of ours — they are mass market. Their business model is built on the commissions clients pay. Our model is different: the company’s principal income comes from investing its own, shareholders’ money. We earn on the quality of our analysis and our ideas — and we give clients the chance to join deals we have already put our own money into. In the kind of business we do, we have no competitors.</p><p><strong>It sounds as though you don’t need clients at all.</strong></p><p>In a sense that is true: we don’t fundamentally need client money — the company earns from its own investments. But I built the company to help people invest alongside us. So that people don’t blindly hand money over for discretionary management, but see how we do it — and join deals in which our own money is already committed.</p><blockquote><strong><em>“We don’t fundamentally need client money — the company earns from its own investments”</em></strong></blockquote><p><em>I never do manage to extract a direct answer on weaknesses: twice Alexey rebuilds the question into an account of how AMCH differs from everyone else. I record that for the sake of accuracy — and put the last awkward question on my list.</em></p><p><strong>How many clients left you last year, and why?</strong></p><p>You do ask provocative questions. Honestly — I don’t know. We don’t keep statistics on departing clients, because as a rule the clients who come to us don’t leave; they have no reason to.</p><p>I can give you the figure from the other end. At present we have around five thousand applications from people who want to become our clients. Around six hundred of them have become active clients. So it isn’t simply that the client wants to work with us — we also decide whether the client suits us. We are extremely careful about choosing a partner, and a client is exactly that to us: a partner.</p><p><em>For the record: the question was about attrition and the answer was about intake, and the figures of a private company cannot be verified from outside. But the proportion itself — six hundred accepted out of five thousand hopefuls — says a great deal about the model. AMCH sells not a service but access. And access, as everyone knows, sells best when it is not being sold.</em></p><h3>By Way of an Epilogue</h3><p>Next on the company’s agenda is a fourth region, Kazakhstan. “School was over; real life had begun,” Alexey said of his first deal, the one that burned to the ground. Judging by the numbers, that real life is going his way so far. Nobody here will promise it will stay that way — and that, it seems, is the whole trick.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dd8083a7f514" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>