YC’s recent Techno-Industrialist Request for Startups, developed in collaboration with the Reindustrialize Summit and New American Industrial Alliance, signals a strategic recognition that “America needs better physical technology, and startups are the best way to build and scale it.”

But Y Combinator’s directory lists a relatively small number of hardware startups, reinforcing the perception that YC is still primarily a software accelerator.

But that’s changing.

Thanks to cheaper prototyping tools, AI-enabled design workflows, and a wave of infrastructure-driven demand, hardware is becoming as investable, and as scalable, as software.

At Lobster Capital, we believe YC is already the YC of hardware. The same model that built generational software companies — rigorous selection, relentless acceleration, and early-stage clarity — works just as well when applied to atoms, not just bits.

Breaking the “Hardware is Hard” Myth

Boom Supersonic entered YC in 2016, followed by Astranis in that same batch. The pattern continued through subsequent years with companies like Gecko Robotics achieving unicorn valuations and Cruise selling for ~$1 billion.

Hardware companies in YC’s portfolio demonstrate survival rates approaching 77%, compared to software’s 84%. In space-focused companies, survival rates actually exceed software benchmarks.

These figures challenge the fundamental assumption that hardware investing carries disproportionate execution risk compared to software. For YC startups specifically, the difference is in methodology.

Traditional hardware accelerators focus on product development timelines measured in years. YC forces hardware founders to compress proof points into three-month cycles, identifying the smallest viable demonstration of technical and commercial feasibility. This approach transforms hardware development from a capital-intensive, multi-year commitment into an iterative process resembling software development.

YC too consistently remains ahead of the curve of Venture land on each trend.

As the geopolitical realities have been shifting since 2022, so have YC’s acceptance of hardware companies.

China’s energy infrastructure is expanding from three to eight terawatts while the US moves from one to two. Supply chain vulnerabilities exposed during COVID-19 revealed dangerous dependencies on foreign manufacturing for everything from pharmaceutical components to semiconductor fabrication equipment, and the Ukraine conflict demonstrated how $500 drones can destroy $100 million military assets, forcing a reassessment of defense tech requirements.

The timing aligns with structural changes making hardware development more accessible. Industrial robot costs dropped from $46,000 in 2010 to $27,000 in 2017, with projections reaching $10,856 by 2025. 3D printing has commoditized prototyping. AI-enabled design tools accelerate development cycles. Off-the-shelf components reduce custom engineering requirements. These factors combine to lower the barrier between software-style iteration speeds and hardware development.

YC Is the YC of Hardware

When I started investing in YC, many investors saw hardware as slow, capital-intensive, and domain-specific. But a new generation of technical founders is rising up using software-style velocity to tackle physical problems.

YC’s core advantages — selection, speed, and storytelling — are just as applicable to atoms as to bits. If you can build a credible prototype, land a real LOI, and show you can do it faster and cheaper than incumbents, the YC playbook works.

Traditional hardware investors obsess over detailed roadmaps, supply chains, and multi-year timelines. We don’t. Our framework is simple:

• Can they prove the core technical insight in 3 months?

Not full product, just the physics or the mechanism.

• Can they get commercial validation via LOIs, not handshakes?

Binding commitments with delivery terms and revenue potential.

• Can they do it faster and cheaper than anyone else?

Speed and capital efficiency matter more than pedigree.

Boom Supersonic exemplifies the YC hardware methodology. During their Winter 2016 batch, founders built a technical prototype and secured a $100 million letter of intent from Virgin and Richard Branson. Eight years later, the company successfully flew supersonic test jets, validating both technical feasibility and commercial demand. The key insight involves breaking ambitious hardware visions into demonstrable proof points achievable within YC’s three-month timeframe.

Astranis followed a similar pattern in the same batch. The team built and launched a satellite during and immediately after YC, establishing immediate technical credibility. The company now operates profitably with multiple satellites in orbit and manufactures new units in a facility directly across from YC headquarters. This geographic proximity serves as a powerful symbol of hardware manufacturing returning to Silicon Valley.

Spaceium represents the extreme end of commercial validation within YC’s compressed timeline. The company concluded their batch with $1 billion in binding letters of intent. These commitments included clear deadlines and receivable structures, distinguishing them from typical non-binding expressions of interest that carry minimal enforcement mechanisms.

The pattern across successful YC hardware companies involves identifying the minimum viable demonstration of both technical capability and market demand. Rather than building complete products, founders focus on proving they can solve the most difficult technical challenge while simultaneously securing commitments from potential customers.

Inversion Semiconductor represents the most technically ambitious hardware company in Lobster Capital’s portfolio.

The startup aims to shrink particle accelerators by 1,000x to build high-power light sources using laser wakefield acceleration (LWFA), targeting semiconductor lithography applications currently monopolized by ASML.

The technical approach could produce light sources 33 times more powerful than ASML’s existing technology, potentially speeding chipmaking by 15x. ASML’s current roadmap targets 1 kW of EUV light output requiring 1 MW of input power. Inversion aims for 10 kW output using tabletop particle accelerators, enabling single light sources to power multiple lithography machines simultaneously.

Founders Rohan Karthik and Daniel Vega bring relevant technical backgrounds. Karthik previously worked in the systems group at Arm, specializing in automated chip design while Vega conducted research for CERN’s AWAKE project, applying machine learning techniques to plasma wakefield acceleration. The team established a laser laboratory in YC’s basement, developing initial prototypes while partnering with Lawrence Berkeley National Laboratory on the BELLA-LUX project. Companies including Tesla and Applied Materials have expressed interest in early developments.

The technical challenges remain substantial. LWFA requires petawatt-class lasers that are expensive and power-hungry. Unless Inversion collaborates with ASML or other lithography manufacturers, the company must develop complete scanner systems and create new industry ecosystems. However, the physics presents no fundamental barriers and market demand is guaranteed if technical specifications can be achieved. The primary risk centers on execution capability rather than market validation.

This case study demonstrates how YC’s methodology applies to the most technically complex hardware challenges. Rather than requiring complete lithography systems, Inversion Semiconductor has focused on proving laser wakefield acceleration can generate industrial-grade light sources. Commercial validation is slowly coming through partnerships with established companies rather than direct sales and the three-month timeline of YC compressed a traditionally decade-long development process into demonstrable technical milestones that enables subsequent funding rounds.

The Tailwinds Making Hardware Easier

Cost reduction across the hardware development stack has fundamentally altered startup economics.

• 3D printing has commoditized prototyping, reducing iteration costs by orders of magnitude compared to traditional manufacturing processes.
• Robotics components that previously required custom engineering are now available off-the-shelf
• AI-enabled design tools accelerate development cycles while cloud manufacturing platforms provide on-demand production capabilities without capital investment in fabrication infrastructure.

These technological improvements combine with favorable policy tailwinds. Defense procurement increasingly favors agile startups over traditional contractors. Energy infrastructure investment has become a national priority. Manufacturing reshoring policies create domestic demand for production capabilities. The result is an environment where hardware startups can achieve software-like iteration speeds while addressing large, government-backed markets.

The convergence of reduced development costs, government policy support, and urgent infrastructure needs has created conditions where hardware startups can achieve venture-scale returns while addressing strategic national priorities. YC’s recognition of this opportunity positions the accelerator to capture disproportionate value from the atoms-to-bits transformation reshaping startup investing.

The future feels inevitable.

We’ve written before about Y Combinator’s predictions around AI agents, and momentum continues building toward something we now accept as certain. This latest YC batch reflects that reality, with 90% of companies focused on AI and 67 specifically building AI agents.

We’ve also seen Scale AI’s $12.5 billion exit to Meta this month. For investors, while there’s tremendous activity in this space, you only need to pick one winner to materially impact fund direction. Venture capital operates on power laws, and Lobster Capital’s thesis centers on YC companies providing the best odds for these outcomes. YC has now backed 20% of all companies valued over $5 billion.

YC’s inaugural Spring 2025 batch just wrapped demo day with 137 companies presenting and a 0.8% acceptance rate. Yet again, the revenue maturity, execution speed, and investor competition we observed point to structural changes that every serious venture investor needs to understand.

Revenue Scale and Due Diligence Evolution

Once again, YC batches are growing revenue batch over batch.

Our analysis shows roughly 70% of companies reported meaningful revenue, with an average of approximately $66k across the entire cohort.

More significantly, we’re seeing a dozen or more companies exceeding $500k in revenue, a threshold that was previously exceptional.

This revenue acceleration creates new due diligence challenges. I’ve developed a three-part framework that goes beyond topline numbers.

1. Retention analysis becomes critical

When companies have only 1–3 months of sales history. We focus on early cohort behavior and usage patterns, though companies with 6–9 months of data provide more reliable signals. I recently invested in a company with nine months of sales data showing exceptional retention, which significantly de-risked the investment thesis.

2. Batch dynamics can inflate early revenue.

I passed on a deal this cycle where 30 of 56 customers were batchmates, despite impressive $500k+ revenue. The founder couldn’t articulate a clear strategy for acquiring customers outside the YC ecosystem. When I asked about his plan to capture 30 or more customers from future batches, his answer was essentially “we’ll do the same thing we did now.” That approach works when you’re in the same batch and people know you, but becomes much harder when you’re old news to incoming companies.

3. Go-to-market sustainability matters more than raw revenue numbers.

The most important question I ask founders centers on customer acquisition while sleeping. Companies dependent on founder-led sales or batch networking without systematic acquisition processes face significant scaling challenges. Some founders think they can rely on YC deal discounts or networking events, but everyone tries those approaches.

AI Concentration and Competitive Dynamics

The above framework becomes an even more important signal when the market is crowded with similar solutions.

90% of this batch focused on AI, with 67 companies specifically building AI agents.

This creates an interesting dynamic where you might have three companies in the same sector within a single batch.

Remember when every pitch was “Uber for X”? Laundry, dog-walking, 2 a.m. ice cream delivery — pick a noun, add an app, raise a seed round. We got a handful of unicorns, a graveyard of zombie startups, and a masterclass in brutal unit economics. Today’s equivalent is “Cursor for X,” and the pattern feels familiar. AI-powered coding for marketing copy, AI-powered coding for legal docs, AI-powered coding for whatever vertical you can imagine.

The concentration might concern some investors, but the underlying dynamics can suggest opportunity rather than oversaturation. When evaluating similar companies, I focus on

• team strength
• iteration velocity, and
• early traction metrics.

Within any vertical stand out companies are shipping features weekly, responding to customer feedback in real-time, and demonstrating clear product-market fit signals.

If you have gene mutation every day through rapid iteration, you adapt much better than companies with yearly cycles. This pressure cooker environment means some teams move faster than others, and those teams typically end up with the most revenue by demo day.

Valuation Dynamics and Market Forces

But with rising revenue, so too has competition for deals.

One company in this batch was raising at $40M on $500k ARR, already expensive at 80x revenue.

When oversubscription pushed the valuation to $50M (100x revenue), I chose to pass despite securing allocation. While I’ve consistently advocated for investing in the best deals regardless of price, there are practical limits.

Some founders are leveraging incredible investor interest to reduce dilution. Others accept only uncapped MFN terms, essentially raising at unknown prices. I’m here to be reasonable for my investors, not buy at unknown prices. Though I acknowledge this approach might work if you catch the next $50 billion exit and get an uncapped seed that converts at Series A pricing.

But this dynamic reflects both market and institutional shifts.

On the wider market, anything AI has of course ballooned in valuation. This chart below shows that the avg. (not median, sadly) AI round is 47x revenue, in comparison to 7x of public SaaS companies.

Institutional shifts too are happening. Traditional venture funds that previously avoided YC due to ownership targets are restructuring their strategies. One fund manager told me she was “fed up” missing quality deals and was changing her fund structure specifically to invest in YC companies.

The best early stage investors in the valley have typically over indexed for growth and traction rather than price sensitivity. Consider Wiz, the cybersecurity company acquired by Google for $32 billion. Their seed round was valued at $67 million in 2020, and they exited for 200x+ returns in five years. The investors were repeat founders in a hot space with apparently the best tool available, growing quickly.

Of course, at other rounds this becomes increasingly important and for follow-on investments, I evaluate companies using frameworks that approach public market metrics.

NASDAQ trades at roughly 20x revenue on average, but the fastest-growing companies command significant premiums. Snowflake trades at 50x revenue due to its growth rate. The goal is identifying companies that can maintain high growth rates while scaling toward these multiples.

One portfolio company from my 2024 investments is approaching $10M ARR just one year post-YC. Their seed was expensive, but their Series A will likely price around $100–150M, making the strong seed round irrelevant. This illustrates the power law nature of venture returns.

If one out of five or one out of ten bets delivers a massive exit over $1 billion, the returns justify high valuations. Remember, one out of four YC unicorns becomes a decacorn.

You can invest in low-valuation seeds outside YC at $7 million valuations targeting 10% ownership, but your startup has one chance in a thousand of making it. YC startups with significant revenue have probably one shot out of 10, 20, or 30 of reaching venture scale.

Market Implications and Future Outlook

Multiple industry observers suggested this batch could produce 20 unicorns from 140 companies, a 14% unicorn rate that would double YC’s historical 6.5% average. This assessment reflects the quality observers saw in the startups and the broader trend of YC attracting the best and brightest founders globally.

The combination of AI-enabled execution speed, higher revenue entry points, and increased competition suggests we’re witnessing a step-function improvement in startup outcomes. Companies are entering YC with more developed products, clearer market signals, and faster execution capabilities. This requires updated evaluation frameworks that account for compressed timelines while maintaining focus on fundamental metrics.

YC remains the most reliable source of high-potential startups. Like buying real estate in Manhattan, you pay premium prices, but the long-term value creation justifies the investment for those who can access the best deals.

The Spring 2025 batch represents this new normal for YC.

2025 is the year of agents, and YC’s latest RFS proves it with an almost exclusive focus.

But for YC, is this obvious, or overly safe? Where are the non-consensus bets that made YC famous?

Or is this exactly what venture returns look like when the platform shift is big enough?

The Full Stack Opportunity

Amongst the ideas in the Request For Startups, one of the most interesting pieces is YC asking for full stack AI companies. Instead of selling AI tools to legal firms, just start your own AI-powered legal firm. Don’t be the shovel seller, be the owner.

I love this mentality because it’s the entrepreneur’s mindset. There are hundreds of billions in market cap up for grabs across industries that can be rebuilt AI-native. Traditional companies carry massive technical debt and organizational inertia that makes AI-first rebuilding economically prohibitive.

I invested in Harper doing exactly this with insurance brokerages. Insurance brokers typically spend 60–70% of their time on manual tasks — data entry, policy comparisons, client communications. Harper is automating all of this, hiring virtually no traditional staff.

While legacy brokerages are constrained by human overhead and manual processes, Harper can process the same volume with a fraction of the cost structure. They’re not competing on margins, like most firms, they have fundamentally different business models.

The same is true of other companies enacting this model like Throxy (lead generation, X25)

This creates venture-scale TAM expansion. Instead of capturing a percentage of what companies spend on AI tools, you’re capturing the entire economic value of the industry you’re replacing.

The Platform Risk Problem

But across many of the other RFS categories. Many feel like features, not companies.

AI personal assistants — OpenAI already has memory, context, and 160 million users. They just need API access to your email and calendar to become your chief of staff. The technical barriers are low — it’s mostly about integrations and user permissions, not AI breakthroughs.

Voice email management — I’ve been dreaming about this for years, but Apple could ship this as an iOS update tomorrow. They control the hardware, the OS, and already have Siri infrastructure. Why wouldn’t they just build voice email directly into Mail?

We’ve seen this movie before. Facebook copied Snapchat Stories and suddenly everyone was posting stories on Instagram instead of Snapchat. The platform always wins unless you build something they can’t or won’t replicate.

Where the Real Opportunity Is

The defensible plays are vertical. You need specialized models trained on proprietary data in industries where OpenAI won’t compete directly.

Sync (Lobster Capital portfolio) is crushing it with lip sync technology, and they’re trained exclusively on lip movement data, highly verticalised. Their models understand facial geometry, phoneme-to-viseme mapping, and temporal consistency in ways that general-purpose models simply can’t match. OpenAI’s GPT might be brilliant at conversation, but it can’t compete with a model that’s seen millions of hours of synchronized audio-visual speech data.

The key is picking verticals with three characteristics -

1. Specialized data requirements
2. Regulatory or privacy constraints that limit platform company entry
3. Workflows complex enough that general AI can’t immediately replicate them.

Healthcare imaging, financial compliance, legal discovery — these are markets where domain expertise and specialized datasets create real moats.

The New Bar

We all know AI is raising the bar — my benchmark now is $1M ARR, (signed, contracted revenue, not just pipeline!).

AI makes it incredibly easy to build impressive demos. Any decent founder can now ship something that looks magical in a 2-minute presentation. But proving real market demand with paying customers is harder than ever. When everyone can ship an AI product, you need much stronger signals of genuine product-market fit.

Several companies in the current YC batch have already hit this milestone, which is genuinely impressive.

But it’s also becoming table stakes.

The companies that stand out will be those hitting $1M+ with strong unit economics and defensible retention beyond just early adoption spikes.

The Consensus Bet

Maybe the obvious bets are fine this time. The AI platform shift is so massive that even consensus plays can generate venture returns.

Every industry is up for grabs. Every workflow can be rebuilt. The barriers to creating billion-dollar companies have never been lower because so many legacy players are vulnerable to AI-native competition.

But the barriers to defending them have never been higher. The question isn’t whether AI will disrupt these industries — but which founders will move fast enough and build deep enough moats to survive the inevitable platform consolidation.

With Spring 2025 demo day coming up, let’s see what happens!

At Lobster Capital, we exclusively invest in Y Combinator startups, focusing on companies that have already demonstrated early traction. When market uncertainty strikes — like we’re currently experiencing with US tariffs and recent market volatility — many investors hesitate. But as someone who’s invested through multiple downturns, I’ve found these periods often create the most compelling opportunities, especially within the YC ecosystem.

Y Combinator — A Protected Ecosystem During Market Turbulence

When investors get spooked by market volatility, they often pull back from startup investments entirely. We saw this pattern during the 2008–2010 financial crisis, the COVID pandemic, and early 2023. At Lobster Capital, I’ve maintained a different approach — continuing to invest in YC companies even when others retreat.

Why? Y Combinator represents a uniquely protected ecosystem within the startup world. I often compare it to visiting the Four Seasons during economic downturns.

While the broader economy might be struggling, these elite environments continue operating largely unaffected. YC companies consistently raise funding even when the broader market slows because they represent the highest concentration of quality in Silicon Valley, which itself contains some of the world’s best startups.

This protected ecosystem status is precisely why our investment thesis at Lobster Capital remains relatively unchanged despite market fluctuations. We’ve found that taking a longer-term view when investing in YC companies pays significant dividends.

The Long-Term View: Why Macro Conditions Shouldn’t Dictate YC Investment Decisions

When investing in YC startups, I maintain focus on the 5–10 year horizon. Current market conditions — whether we’re experiencing a downturn or a bull market — matter far less than finding exceptional founders building high-potential companies. This long-term perspective silences the noise of short-term market fluctuations.

The stock market and startup ecosystem operate on different dynamics. While public markets might fluctuate wildly based on tariff announcements or interest rate changes, the best YC companies remain focused on building. This creates a fascinating paradox: periods of broader market uncertainty often yield better entry valuations for investors willing to maintain conviction.

I’ve seen this firsthand with investments during COVID in 2020 and the early 2023 downturn. While others retreated, we found companies with exceptional metrics at more reasonable valuations. By the time these investments reach maturity several years later, today’s market conditions will have cycled through multiple phases and become largely irrelevant to outcomes.

This extended time horizon also means we’re witnessing something quite interesting in the startup world lately — particularly with YC companies. The conventional wisdom about exit timelines is being challenged in ways few predicted.

YC Startups and the Accelerating Exit Timeline

A few months ago, I stated we’ve hit peak illiquidity in the startup world.

After watching exit timelines stretch from 8 years to 10–12 years, I believed we were due for a significant reversal.

Recent data is slowly proving this prediction correct, with several 2025 Q1 companies experiencing remarkably quick acquisitions:

• StarJar (YC 2024) and Rehook (YC 2022) recently acquired as part of major M&A deals
• OpenAI reportedly acquiring Windsurf (founded 2023)
• Voyage AI (founded 2023): Acquired at 19 employees for $220M ($11.6M per employee).
• Helio (founded 2022): Web3 payments startup with 20 employees acquired for $175M ($8.8M per employee).
• DealerClub (founded 2024).

What’s driving this acceleration for the broader ecosystem? Three forces are at work:

1. AI acceleration: AI is compressing everything — development cycles, go-to-market timelines, and customer acquisition. What once took years now happens in months, and YC companies are at the forefront of this transformation.
2. Market resistance to extended illiquidity: The 10–12 year timelines we’ve seen recently don’t serve anyone well. I predict we’re heading back toward average exit timelines of six years or even less, which includes these 2–3 year acquisitions we’re starting to witness.
3. Political climate change: The current administration appears more favorable toward M&A activity than previous years, unlocking deals that might have faced regulatory challenges before.

These shortened timelines represent a welcome development for both founders and investors in the YC ecosystem — creating more dynamism and potential for returns. But despite these accelerating timelines, my approach to evaluating YC founders remains focused on fundamentals that transcend exit timing.

What I Look For in YC Founders: The Personal Connection

Earlier in my investing career, I noticed founders were almost offended by asking about exit strategies — and rightfully so. How can anyone truly know what will happen 5–10 years from now, especially in the rapidly evolving AI landscape?

Instead, I focus on whether founders have a genuine connection to the problem they’re solving. In our most recent YC investments, this pattern emerged clearly:

• One founder had spent over a decade as a designer before creating tools for that community
• Another came from a multi-generational family of insurance brokers
• A third built hospital intake automation after nine years of personal experience with medical administration inefficiencies

This connection creates the resilience needed to persevere through inevitable challenges. It doesn’t require founders to be lifelong experts — just authentically engaged with their users and problem space

We see this in YC’s greatest success stories: Brian Chesky still occasionally handles Airbnb customer support emails, and Tony Xu of DoorDash continues making deliveries to understand the customer experience.

Rather than fixating on exit strategies, I ask: “Are you genuinely committed to creating value?” YC founders who obsess over solving meaningful problems eventually discover that exits take care of themselves — even unexpectedly quick ones, as we’re now seeing with companies like Windsurf.

But in an AI-accelerated landscape, founder connection to problems is also helping to build robust distribution moats that can create lasting defensibility.

YC Pivots and Building Distribution Moats in the AI Era

The YC mantra of “make something people want” sometimes requires founders to pivot — both before getting into YC and during the program. Some of the most successful YC companies like Brex (originally a VR startup) completely transformed their focus during the accelerator.

As AI continues reshaping the startup landscape, I’m particularly attentive to how YC founders are building defensible moats. With Vibe coding and other AI tools making it easier than ever to clone products, technical differentiation alone rarely provides sufficient protection. This is why when evaluating YC companies for Lobster Capital, I focus intensely on:

1. Customer acquisition systems: I specifically ask founders, “How do you acquire customers while you’re sleeping?” Companies with systematic, repeatable growth engines will outperform regardless of market conditions.
2. Community and content strategy: Every founder should leverage content to build distribution. Whether through YouTube, blogs, or social media, content creation represents one of the most asymmetric opportunities available to early-stage companies.
3. User experience: Why are companies like Cursor and Windsurf succeeding when thousands of similar AI tools exist? Their superior interfaces and user experiences create lasting differentiation.

The YC companies that will thrive through market uncertainty are those building these durable advantages — combining authentic founder connections to problems with systematic growth engines that create compounding value regardless of market conditions.

At Lobster Capital, that’s exactly where we’re focusing our investments.

As we pass the quarter of the 21st century, new ideas are taking hold about how to build a game-changing company. Y Combinator’s Winter 2025 batch has shattered revenue records, with some companies achieving $50 million in profits at Demo Day while others secured $100 million valuations. The rise of AI development tools has fundamentally changed startup velocity. In previous years, having $20–30K MRR was exceptional, now a quarter of companies exceed $100K,” raising the bar for investors seeking the top 2%.

This acceleration comes from AI-powered development enabling founders to build MVPs in days rather than months, with “vibe coding” generating over 90% of codebases. Technical debt remains a concern, but the most successful companies balance rapid AI development with strategic expertise.

After examining Y Combinator’s latest Request for Startups focusing on American dynamism, vertical agents, compliance, and AI infrastructure, it’s worth looking at how the cohorts reflect those priorities and how they’re rewriting the startup playbook in real time.

Vertical agents vs Analog AI roll up

YC is openly bullish on the potential of AI agents to transform the long tail of specialised industry niches, in the same way vertical SaaS did. What it can’t agree on is what the best model is to do so. The first approach borrows from vertical SaaS, though its pricing economics are radically different. Vertical Agents promises not to build industry-specific foundation models, but use general models in a way that is customised for those industries.

The challenge for YC is that most of the cohort founders are technical experts who are industry agnostic, so getting that customization is what matters. Given the intellectual might of both the cohort founders and the partners, the question remains: will the strategy that worked in the SaaS era — getting genius kids obsessed with a certain niche — work in today’s AI landscape?

One such startup is Paratus Health, which uses AI agents to replace nurse intake processes.

Founded by Tannen, who experienced firsthand the critical gaps in medical intake after his family struggled with undiagnosed rare disease. He (and co-founder Pablo) built an AI-powered intake platform that collects comprehensive symptom data before appointments.

The system follows doctor-approved clinical protocols, cross-checks medical literature, and generates structured reports with symptom history, differential diagnosis, recommended tests, and pre-visit SOAP notes for EHR integration. In an industry struggling with staffing shortages and rising costs, there’s significant potential for adoption, though penetrating healthcare’s complex procurement systems will require strategic partnerships with established providers and EHR platforms.

This is the traditional approach, though some in the industry have argued that in the AI revolution you don’t want to be another software vendor drowning in the buzzword-based noise. It is purportedly better to be a cashflow-generating brick and mortar business that is fully AI enabled. General Catalyst have backed quite a few of what are essentially AI roll ups, YC are now doing the same at an earlier stage.

Gail is an AI-powered immigration law firm that is starting with the controversial H-1B Visas to create a law firm that is tech first. In many ways, this is similar to the way Deel or Scale AI ended up, where the means to cashflow comes from stripping services to minimal human inputs with AI. This is a great strategy for areas where the process is formulaic and factory-like, though it will be interesting to see how AI-enabled vs AI software compare to each other in 3 years time.

Another similar play is Operand, which promises to replace management consultants, starting in Retail & E-commerce. Again they are essentially selling AI-enabled consulting which marks a break from the traditional software self-serve purism of cohorts of the past. Finally, Harper is reinventing a commercial insurance brokerage with a tech-enabled model. Rocketable is even a direct holding company for software roll ups, showing the metamorphosis from VC to PE. Is saving 40% costs with AI better than selling annual subscriptions for 3000$?

Infrastructure

YC’s Request for Startups showed us that it’s going deeper and deeper by the batch, with a focus on AI infrastructure. The RFS called for a startup to take on NVIDIA, and while they don’t have that, they’ve got the next best thing even further down the value chain. Inversion Semiconductor is taking on the ASML Lithography machine monopoly claiming it can build 15X faster fabrication machines.

The Dutch firm’s monopoly has long been unchallenged. It has the knowhow and infrastructure to build machines with such precision it was seen as infeasible to compete. Lithography uses light to pattern circuit features on silicon. Inversion wants to scale transistors to their physical limits by shrinking particle accelerators 1000x, to create a high power light source.

Another ambitious play takes aim at the cloud hyperscalers. TensorPool stylises as the Vercel for GPUs. The founders were frustrated by the cloud experience of Google and AWS for training Machine Learning models. They handle GPU orchestration and execution at half the cost of the hyperscalers. Vercel was a key player in providing the infrastructure behind the Next.Js framework for front-end, so if TensorPool can do the same for backend, they might be able to take on the hyperscalers.

Fintech as the breeding ground for agents

YC has always loved fintech whether its Brex (W17) or Stripe (S09). With the advent of AI, the financial services industry is ripe for disruption. YC’s heavy fintech representation represents the influence of new partner Tom Blomfield. With three main focuses. 1) Replacing the junior analyst 2) reducing compliance and accounting headcount 3) Stablecoins.

The former is a tough nut to crack. Trata, Fira, and Finbar are both looking to do this for Hedge Funds. It’s tough to provide value here beyond the core models without proprietary datasets, so it’s paramount that technical founders build out the best system for searching through large volumes of unstructured data, in the same way that Hebbia has done for the PE market. Blomfield himself called for risk and compliance startups in the RFS — and he has got his wish with Cardamon focusing on financial compliance. In an adjacent space, Tergle, Tally, and Bluebook are entering the AI audit & accounting race, alongside established companies such as DataSnipper.

After Stripe’s acquisition of Stablecoin infrastructure, Bridge, YC is betting big on emerging markets as the driver of adoption amidst Trump-led US volatility impacting currencies. Infinite, a processor, and Blindpay, an API are two cohort members building the rails for widespread take up.

Winning in Crowded Markets in the Application Layer

As ever with YC, there are ideas that are concentrated and those more esoteric.

Perhaps the prize is larger for the former, selected as a function of YC’s own portfolio construction. YC is backing AI sales, customer service and recruitment agents. It has even backed an AI-cofounder named Woz! That shouldn’t gloss over the fact that these are big ideas launched in the perfect incubator to win their large TAMs.

Beyond the general purpose moonshots, there are great technical products being launched in YCs bread and butter: Dev tools. And following the American Dynamism theme, a focus on space (Reditus Space, Instinct, Orbital Operations) and on Government contracting (Candor, Archon.) We could go on and on, from autonomous patrol boats to pothole-filling robots.

We’re seeing a fundamental shift in YC’s attitude towards sectors it had previously left unexplored, with a renewed focus on deeper products, to accompany the swathes of AI copilots and agents.

All in all, I’m excited to get into some of these deals after seeing first hand the energy coming from the campus this time around.

There are some founders here building special products and Lobster Capital will be there to back them.

YC has an uncanny ability to see the future before anyone else.

So when YC releases their ‘Request for Startups’ — a signal of where they believe the biggest opportunities lie ahead — the startup world listens.

The latest RFS just dropped. And in a world where AI progress is measured in weeks, not years, it might be their most important one yet.

The rate of change in the world is compounding. For hundreds of years in the middle ages, time stood still. Innovation lagged. Peasants tilled the fields for their feudal lords, who ate the bread it produced. This status quo persevered until the Renaissance, where reason, science and engineering spawned a Cambrian explosion of new ideas.

If you’d asked a Victorian financier who issued a loan for an early cotton mill to predict what he’ll be funding in 10 years, he’d likely be bewildered if you said it would be a coal-powered mill.

As venture investors today, we face a unique challenge: a 10–14 year horizon to exit our most successful companies, yet technology is advancing at breakneck speed. What seemed cutting-edge just two months ago gets surpassed by the latest breakthroughs, whether it’s DeepSeek’s capabilities or the rise of vertical AI agents.

At Lobster Capital, our core mission is to identify these Black Swan opportunities — the outliers that reshape entire industries. This is precisely why YC’s shift to four cohorts per year and their Request for Startups are so significant.

Given YC’s proven ability to shape the future — and our privileged access to invest in their top founders — analyzing their chosen themes will reveal opportunities that, years from now, we’ll recognize as prescient bets on tomorrow’s giants.

A Secure AI App Store

Garry Tan

“We want a new kind of AI App Store and OS layer that sits on your computer”

This one from Garry is straight out of the David and Goliath playbook.

This is a play akin to what created the Apple and Google duopoly through the App store and Playstore in the mobile era. Their only threat is developers routing payments out of their purview, and an Antitrust case with Epic Games. Garry sees one marketplace with user memory from sprawling AI apps as being the epicentre of our new world.

Yet people have said, ‘Won’t big tech build this” about every idea since the dot-com bubble.

Google and Apple are behemoth corporates — they face an innovator’s dilemma. That’s why Apple’s AI efforts have been embarrassing, while Google, with their endless Free Cash flow and distribution moat through copious workspace apps, have disappointed.

Enter Garry’s giant killer.

Many might even say that this is the obvious place that Open AI, Anthropic or X AI end up. On top of providing models through their APIs, they want to become the platform layer for specialised AI apps for narrower use cases. The problem with that, is that most application layer companies use different models for different purposes. That means that someone outside of the foundation model providers is poised to build the aggregator — think the G2 / Capterra of AI, or the App Store reimagined for the AI age.

Marketplaces are a hard model to scale with delicate 2-sided network effects. Keeping the memory from various apps in one centralised space may be something either AI apps or LLM providers object to.

That said, if there is ever a place to execute an idea this ambitious, it is YC… their largest exits have been from platforms and marketplaces (Reddit, Airbnb, Doordash, Instacart).

Datacenters

Diana Hu and Dalton Caldwell

“We can envision what the future will look like: software will handle every aspect of planning and constructing new datacenters or warehouses, from site selection and construction to setup and ongoing management. These applications will be “lights out,” with robots operating autonomously 24/7 without human intervention.”

YC’s turn towards deeptech came much earlier than many believe.

Boom Flight (YC16) has just flown the first hypersonic flight since the Concord. Even earlier Sam Altman selected Helion in 2014, taking a novel approach to Nuclear Fusion. That’s why Diana and Dalton called for the pressing need to reimagine how we build the capacity to power the AI revolution.

McKinsey estimate that 70% of the future demand for data centers is for new AI-ready ones.

If anything, this is a part-proptech / part robotics play.

With E-commerce returning to a linear trend after the pandemic, the case for where to put the data centers that power our new societies is not just about where capital can arbitrage cheap land. Instead, navigating through the patchwork of planning / zoning regulations, while finding the optimal areas for grid efficiency and cooling.

The case for a specialised project management system for construction of such complex structures remains strong in an industry that has resisted innovation. On the deep-end of this play, is the grand vision for autonomous robots in both construction and maintenance. At the current levels of humanoid robots, that may seem far fetched. As investors with a long time horizon in the most ambitious founders from the best accelerator, ‘far fetched’ is not in YC’s vernacular.

Compliance and Audit

Tom Blomfield

“Traditional compliance tasks require reading dense regulations, cross-checking internal policy and procedure documents, manually sampling front-line work done and producing repetitive reports. These manual, time-consuming workflows beg for automation.”

Tom (Monzo and GoCardless) knows a few things about disrupting regulated industries.

Compliance is a large cost centre for financial services, industrial & energy companies, in particular. Getting it right is mission critical to avoid costly fines and reputational damage. While Trump may have initiated the deregulation agenda, particularly in financial services, layers of regulation will stay with us for years to come. Tom is taking aim at the burdensome KYC/AML processes, all the way to internal audits in industrial and food production.

There is a deep irony here. Those who know about the most arduous compliance challenges for agents to solve are probably pretty risk averse people. After the global financial crash, many front office workers went from trading to compliance. The versatility of the technical founders that YC selects will put them (and us) in prime position to find the people who can build enterprise-grade systems that turn bureaucracy into a competitive advantage.

After all, LLMs excel at synthesising complex ‘legalistic’ information at scale in a way that humans find mentally draining. There are thousands of compliance and regtechs waiting to be built… they just need to find the customer pain points first.

DocuSign 2.0

Michael Seibel

“We want founders to rethink how documents should be created and distributed for signature in a world powered by AI.”

Docusign ($17B market cap) was a remarkably obvious idea that was executed to perfection, all the way to becoming a public company.

It is a typical ‘Web 2’ idea, digitising paperwork and the ink it is signed on, cryptographically. Michael has controversially decided that it needs an upgrade for the AI era.

Existing tools are rigid and expensive. Michael rightly points out that the process of creating and editing documents is often a collaborative or adversarial one, done on Word or Google Docs. Other than those tools’ limited co-pilot features, there is little in the user experience that makes the process smooth for redacting and explaining terms.

The interoperability of this idea with various tools would be key. While there are many players in this space working in specific domains such as Private Equity or Securities Law, there may be a chance to create a general ledger for documentation that dethrones King Docusign.

Browser & Computer Automation

Jared Friedman

“Every website and every app now effectively has an API. It means that any workflow that people can do on a computer can be automated.”

Browser use agents expand the horizon for the automation of process-driven knowledge work. We’ve seen Open AI’s Operator and Anthropic’s computer use show the green shoots of what can be done at a generalist level. We’ve even seen competitors such as Balderton-backed Convergence, beating their benchmarks for browser use. After all, the web is just a load of HTTP and JavaScript, waiting to be navigated.

The question here is twofold. Who can build the best general computer-use agent that becomes a sticky product for B2C and B2B workers. And who can use the underlying technology in the most creative ways for specific verticals. Jared, in his typical cerebral style, doesn’t stipulate which one of those he is looking for. That might be because he knows the possibilities are endless when you give the brain a hand.

Of course the early versions may look at booking a flight and other trivialities. The real test is how reliably you can get browser use to perform mission-critical tasks autonomously. If Marc Andreesen invented the browser, then there must be someone out there who will master navigating it. You’d be wise to put your money on them being in YC.

AI Personal Staff for Everyone

David Lieb

“Software is a proven way to bring to everyone what only the rich could afford before.”

Her — Spike Jonze’s 2013 film — depicts a world where everyone has a personal voice assistant. It’s well documented that this is the future that Sam Altman wants, but how close are we to it?

The low hanging fruit of a personal assistant, booking meetings and sending emails is already fiercely competitive. How far can this go? As David says PA’s, private bankers, personal trainers and chauffeurs were once the privilege of the wealthy. To democratise them requires a mix of domain-specific models that bring the personalisation that you’d get from a bespoke service into the AI realm.

With the core LLMs paying sectoral experts to create new training data now they’ve exhausted the corpus of the internet, it will be tough for application layer companies to build such systems without proprietary data.

That said, there is a unique opportunity for AI personal assistants to become non-mercenary systems. A financial advisor is not always acting in your interest because it’s in their interest to push products that generate fees; even a doctor or nutritionist, will have products to ‘sell.’ This creates a gap for the unbiased personal assistant that acts in your best interests, with the feeling of a 1-to-1, human-led service.

Devtools for AI Agents

Nicolas Dessaigne

“We are now witnessing the emergence of the next wave: AI agents that go beyond assisting humans and can autonomously make decisions.”

Agent builders are the most competitive arena for startups right now. Much like the no-code revolution enabled non-technical people to ship CRUD apps, the agent builder space promises to have an even greater impact.

The orchestration layer enabled by powerful reasoning models are the picks and shovels of the AI goldrush. There are early breakouts such as N8N, Relevance AI & Crew AI, while other incumbents such as Hubspot and Salesforce are pushing agent builders natively on their platform.

The winners in this space will be the ones who crack the scale question that eluded no-code apps for so long. Equally, much of the tooling available on these agents are similar… connect to a Vector Store, search the internet, scrape a webpage for research. This has pushed lots of use cases to be sales-focused, which is an arena the core LLMs will probably win themselves. Nicolas points out, the chance to create better tooling, and a killer, intuitive agent builder that scales will likely win a $10B prize.

The Future of Software Engineering

Pete Koomen

“Humans won’t write much code directly. Instead, they’ll manage teams of agents that build software for them.”

Code is dead…except it isn’t, it’s just different now. It is no longer a game of more inputs to create more code.

Agents enable 10X developers to scale with lean startups just as Paul Graham ideal espoused. We’re already seeing leaner dev teams like Bolt, Lovable and Cursor scale to 10s of millions of dollars of revenue in months. Whether those products are novelties is a different question, but the future of development is almost certainly leaner teams managing teams of stratified agents doing the stuff that makes coding laborious.

This is a rewriting of the labour force. The division of labour will be amongst agents performing specialised functions, with agents managing them. This will free generalist developers to become more efficient, as long as there are tools that enable them to do so.

AI Commercial Open Source Software (AICOSS)

Dalton Caldwell

“It’s not uncommon for the organization that releases the open source code to not want to focus on providing commercial support. For instance, Google and Facebook have open sourced a number of tools but aren’t always in the business of providing commercial support for businesses to use those tools — which has yielded an opportunity for startups.”

“Open source is always 2 years behind closed source.” That held true until DeepSeek swept through the West.

Two people you should never bet against are Zuckerberg and Elon, both proponents of the open source approach. The problem is that large corporations and open source are a bit of an oxymoron. You may open source one of your models, but if the developers are left to their own devices to work out how to deploy them, then that is mere lip service.

Fortunately this creates a gap for startups to exploit. Legacy firms may want to run models on their own on-site bare metal, but are left in the dark on how to do so. Bridging the documentation gap will be a key area for startups to capitalise on.

AI Coding Agents for Hardware-Optimized Code

Diana Hu

“AI hardware is still constrained by software. We’d love to see more founders work on AI-generated kernels that make more hardware alternatives work for AI.”

Taking on Nvidia is no mean feat. The Chip monopoly is upheld not by hardware supremacy, but by software supremacy. Continuing YC’s push into deeper terrain, there may well be people out there who can crack the Nvidia secret sauce. Whether that entails launching a competitive hardware company or supplying rivals like AMD with the means to compete.

Such an upheaval would have dramatic geopolitical and market impacts. We know by now that China (DeepSeek) likely got advanced Nvidia Chips through Singapore, rather than using the older ones as stipulated by US sanctions. That didn’t stop Nvidia stock from sinking.

Perhaps there’s a YC startup who can take on Jensen Huang’s empire?

B2A: Software Where the Customers Will All Be Agents

Dalton Caldwell

“With the advent of AI and agents, it seems like it’s a good idea to build software and services where agents as customers are actively supported and documented rather than an edge case.”

It may sound dystopian or utopian. The reality is merely an evolution of the current state of play in internetland. The web has already been home to crawlers and scrapers since Google indexed the world’s data. SEO was internet experiences built for bots, even if the use of apps and websites remained human-centric.

With the proliferation of agents that could shift. Agents will have different preferences and ways of operating when browser-use comes to the fore. User Research and UX will no longer be about human psychology. Rather, it will be how agents interact with code. How they navigate to achieve their goals; how they make decisions.

This will require a fundamental rethink of front-end and design principles, something that YC has helped shape.

Vertical AI Agents

Jared Friedman

“The value prop of B2B SaaS was to make human workers incrementally more efficient. The value prop of vertical AI agents is to automate the work entirely. Vertical AI agents that reach human-level performance grow extremely quickly. It’s entirely possible this opportunity is big enough to mint another 100 unicorns.”

Jared’s thesis here is simple. In the SaaS era, the ‘obvious ideas’ were won by the incumbents. There were competitors to Google Docs, but Google managed to vanquish them.

The long tail of vertically-specialised variants managed to create venture-level outcomes. If the markets were big enough, great. If not, they used it as a base to expand horizontally. Rippling (YC W17) and Gusto (YC W12) started out as just payroll providers, expanding to become HR superapps.

In the agentic era this could play out similarly. The core LLMs will capture the more general use cases, while the long tail is taken by vertical agents. The key difference is that instead of facing legacy analog incumbents, these vertical agents will face the last generation of SaaS leaders, who are scrambling to put AI into their established product.

That’s a different dynamic, though you can easily see AI-first startups surmounting the last generation of CRUD apps. And you’d bet that those agentic startups will come from YC.

Startup Founders with Systems Programming Expertise

Dalton Caldwell

“Being able to understand and innovate on the entire software stack can be an unfair advantage to a startup founder. We would love to fund founders with similar expertise and focus on low-level systems that they can bring to bear on whatever startup idea they choose.”

You cannot create scarcity when there is abundance.

The last generation of founders came up with the tailwind of cloud computing and low interest rates. They never experienced the scarcity of hosting a website in the dotcom in a garage. The bar moved from, “does this work” to “does this look shiny.” While we no longer have the pre-cloud constraints, there is less capital available for startups.

That means we must return to first principles of doing more with less. The purported lesson from DeepSeek was that necessity is the mother of invention. Whether you believe their cost figures or not, YC is looking for a particular mould of founder that understands the entire systems, not the sum of their modular components.

That is as much of a mindset as a skillset.

Inference AI Infrastructure in the World of Test-Time Compute

Diana Hu

“With Deepseek R1 and OpenAI o1 and o3, there is a new scaling trend that suggests we’ll need far more compute at inference time when AI apps actually use these models.”

The one indisputable truth from DeepSeek is that inference is as important as pretraining. The scaling law only works until you run out of stuff to ingest.

DeepSeek’s great innovation was distillation at the point of use, making a more efficient model. Enabling models to reason and respond to scenarios not seen in the training data will be key as systems scale up. The unit economics of an agent business is not the same as a SaaS business, even as API prices fall by the day. Making the infrastructure more efficient for batch processing when there’s thousands, if not millions of users will be key.

The new call for startups marks a shift from the previous, with a focus on American dynamism and reindustrialisation of America — shifting to boosting the AI infrastructure for an agent-led world.

YC is always 12 months ahead of the curve, and we will likely see the fruition of these grand ideas coming to a seed round near you.

And as always, Lobster Capital will be in that round, rain or shine.

“Fundraising is the beginning of a starting line of a new race. Winning the race is what matters, not what brand of fuel you put in the tank,” says Y Combinator CEO Garry Tan.

This perspective is becoming increasingly relevant as we witness a fundamental shift in how startups are built and funded.

The venture capital industry has long operated on a well-worn playbook: raise seed funding to build, Series A to grow, Series B to scale, and keep going until exit or IPO.

But something’s changing. In YC’s Fall 2024 batch, we saw two-person teams hitting $4 million in revenue without a single employee. A year ago, that would have seemed impossible.

When investors like Chamath Palihapitiya called that seed investing would become the only stage that matters in the AI era, it sounded like Silicon Valley hyperbole at the time. Now? Not so much.

The math is changing. AI tools are slashing development time in half. Distribution channels like TikTok can create overnight success stories. Cloud infrastructure makes scaling cheaper than ever. For the first time, the limiting factor isn’t capital but execution.

The New Economics of Building Startups

Google engineers using AI tools cut their development time by 21%. Microsoft’s GitHub Copilot users completed tasks 56% faster.

Not only is that the slowest it will be — it creates a completely different way of building products.

One YC company from the Fall 2024 batch: two founders, no employees, $4 million in annual recurring revenue (contracts projected out) after just 3 months.

YC seems to have spotted this shift early. In 2022, they increased their standard deal to $500,000.

But now, instead of using that money to hire faster, companies are staying leaner and stretching their runway further. Some aren’t even raising more money after Demo Day.

Why?

• Development: AI coding tools can replace entire engineering teams for early-stage products
• Distribution: You don’t need a marketing team when AI can procedurally generate 100s of ad variants
• Infrastructure: Cloud services that used to cost $100,000 monthly now run for a fraction of that

But… While it’s easier than ever to build, the bar is also higher.

Seed-stage startups today are expected to show more revenue and more polished products than before. You’re not raising on a pitch deck anymore — you’re raising on metrics.

This creates two clear paths: either raise a small seed and race to profitability, or raise big and use that capital advantage for aggressive market capture. There’s less room in the middle ground.

The market reflects this bifurcation. Median seed valuations hit €5.4 million in 2024, up 15.7% in just one year.

Deal sizes now typically land between $1–4M, with 7% quarterly growth in Q3 2024 alone.

Better tools mean you can build bigger companies with less money. But that same efficiency means investors are willing to write bigger checks at the seed stage — median valuations are up 15.7% in 2024 alone.

The YC Effect

YC calls this the golden era of building — and they might be right. No industry seems safe from disruption as new computing paradigms emerge.

The traditional three-month pressure cooker of YC has taken on new meaning in the AI era. When development time is cut in half by AI tools, those three months become exponentially more valuable.

Companies are both shipping products faster and using that extra time to crack distribution and find customers. The intensity of the program becomes a multiplier.

With companies finding traction sooner, for seed investors, value accrues earlier. Look at YC alum Zapier — they raised once after Demo Day and never looked back. Now they’re doing over $100M in revenue. Baubap (YC S21), where I’m a proud investor, followed the same playbook, reaching $100M+ revenue in a single round.

FWIW, Google raised just one traditional VC round, while others like Klaviyo, Trade Desk, and Veev all followed similar paths. What’s new instead is how accessible this path may become.

For seed investors, this can mean significantly less dilution. When companies reach profitability or substantial revenue on seed funding alone, early investors maintain their ownership through exit. This opens up new possibilities for returns — whether through secondary markets or more traditional exit paths.

But there’s a catch. The AI gold rush has users trying every new tool that launches. Sure, we’re seeing record-breaking early revenue numbers, but we will have to see how well these tools can retain customers, even once the novelty wears off.

But the market’s confidence in this new era shows in the numbers.

YC valuations doubled between 2020 and 2022, hitting $25–30M at Demo Day, using YC’s $500k standard deal.

Instead of rushing to raise more, many founders are keeping more equity and focusing on profitability. The question turns from “How much can we raise?” to “How much do we really need?”.

This is a big shift.

2025 is also the first year of YC’s shift to four batches.

Smaller, more frequent batches mean more focused attention from partners, better timing for companies ready to raise, and faster feedback loops on what’s working. It’s YC adapting to a world where companies can be built and scaled faster than ever.

For YC, this could be a defining moment.

The accelerator that has been preaching “make something people want” for nearly two decades now has tools that let founders do that faster than ever.

In this environment, investors can’t compete on capital alone — it’s about relationships and true value-add.

Firms like Lobster Capital are thriving by focusing on deep relationships within the YC ecosystem. When capital becomes less of a commodity, being a trusted partner becomes the differentiator.

We’re excited about the future.

Y Combinator’s first-ever Fall batch just wrapped up and after I watched founders pitch from the stage and investors literally sprint to speak with them afterward, I’m breaking down some thoughts.

With 93 companies selected from over 10,000 applicants this cohort is YC’s first step toward their new quarterly batch structure launching in 2025.

The shift from two to four annual batches addresses the accelerating pace of technological advancement alongside several key modifications:

• Reducing batch sizes from ~250 to ~100 companies
• Expanding group partner engagement with each startup
• Reviving the signature in-person Demo Day format

At Demo Day this smaller cohort size also enabled deeper investor engagement, and there was talk that this was one of the most impressive batches yet.

• First-ever batch to achieve 3X average revenue growth during the program
• 87% of companies building with AI/ML capabilities
• 74 B2B companies
• 24 developer tools companies

But from my direct observations what makes this cohort unique is what it suggests about the future of startup acceleration.

New Batch Performance

Meeting a YC alum from the 2005 batch (when there were only ~8 companies) highlighted how far the pendulum had swung. While that extremely small size might have been too limited, this new 93-company format is a return to balance from 2020–22 batches of up to 400.

1. Partner Engagement: Six YC partners (Dalton Caldwell, Diana Hu, Pete Koomen, Nicolas Dessaigne, David Lieb, and Garry Tan) worked directly with the companies, providing focused mentorship impossible in larger batches.
2. Investor Attention: With three pitch sessions and 45-minute breaks, investors could engage with each company. At previous demo days, I often found myself “playing catch up,” with hundreds more companies to evaluate.
3. Community Building: Founders formed stronger connections within their smaller cohort. This dynamic has proven powerful. Historical examples like Supabase and PostHog, now used by 80–90% of new YC companies, demonstrate how the YC ecosystem can provide initial traction before broader market expansion.

The batch composition heavily reflected the expertise of participating partners. For example:

• Diana Hu, Dalton Caldwell, and Jared Friedman’s presence correlated with strong developer tool representation (24 companies)
• The absence of Surbhi Sarna (biotech focus) was reflected in fewer biotech startups

This partner-batch alignment enables more specialized mentorship and targeted growth strategies.

This may correlate with exceptional performance:

• First batch in YC history to achieve 3X average revenue growth during the program
• Multiple companies reaching $100k+ ARR within weeks
• Several companies showing 50–130% month-over-month growth

The Paul Graham Benchmark Realized

Paul Graham’s 7–10% weekly growth target has long served as both north star and stumbling block for YC companies.

When Graham first proposed this growth rate, it required significant capital deployment, aggressive hiring, and often unsustainable burn rates. The F24 batch demonstrates a new paradigm: companies achieving this growth through capital efficiency rather than brute force scaling.

Implications for the Future of Company Building

The compression of startup development cycles is fundamentally changing early-stage investing. When a developer tool in F24 reaches $102K ARR within three weeks of launch, it challenges core assumptions about company building.

Traditional paths from seed to Series A are compressing — not from capital constraints, but from reaching key metrics faster. Zapier (S12) proved this model by reaching hundreds of millions in ARR after only a seed round. This is becoming more and more common, and Lobster portfolio companies like SoloSuit (W21) and another of my investments, Baubap (S21), are achieving highly capital-efficient trajectories.

Chamath Palihapitiya’s recent prediction on All-In that seed investing might become the dominant form of venture capital is playing out in real time. When companies can reach $1M+ ARR with small teams and minimal capital, the traditional venture scaling playbook needs revision. This benefits seed investors through reduced dilution, but introduces new challenges.

The tools enabling rapid growth also lower barriers to entry, leading to faster market saturation. Speed to defensibility, rather than just speed to market, becomes the key differentiator. For investors, this means developing new evaluation frameworks that account for both accelerated growth potential and evolving competitive dynamics.

Beyond the 87%: A Deeper Look at AI in YC F24

As someone who wrote “Why I’m Not Investing in AI” a few years ago, I’ve watched this landscape evolve from hype to a given.

Three Categories of AI Companies

From evaluating hundreds of AI startups, here is a framework for categorization:

1. Infrastructure Players: Companies building foundational AI technology. These are rare and require deep technical differentiation.

• CTGT: Building an entirely new AI stack as an efficient and safe alternative to LLM ecosystem
• Circlemind: Open-source RAG using knowledge graphs and PageRank ($55K MRR)
• Synth: Helping AI agents learn from data
• Lune AI: Building better GPT and Claude specifically for coding

2. AI-Native Solutions: Startups like Andoria ($93K ARR in YC) that couldn’t exist without recent AI advancements. Their entire business model depends on AI capabilities.

• Andoria: AI customer agent for in-app walkthroughs ($93K ARR)
• Foundry: “Retool for AI Agents” ($102K ARR in 6 weeks)
• OpenFunnel: Personalized agents for AI-first GTM teams ($8K MRR in first 3 weeks)
• Origami Agents: AI agents finding perfect customers ($37.5K MRR in 40 days)

3. AI-Enhanced Operations: The largest category — companies using AI to execute traditional business models more efficiently. Sandra AI’s growth to $6.2K MRR in two weeks exemplifies this approach.

• Sandra AI: AI receptionist for car dealerships
• Vly.ai: Building full-stack software without developers
• Ember Copilot: Medical scribe reducing claim denials
• Crustdata: Real-time B2B data via APIs, enabling more AI agents (🦞portfolio with super impressive growth)

Investment Strategy in AI

Future Proofing: I passed on a company doing $10M in ARR in AI music creation because the space could be fundamentally different in 2–3 years. Sam Altman predicts AGI might be 3 years away — any AI investment needs to make sense even in that scenario.

Sustainable Advantages: Pure AI feature plays rarely justify investment. The companies we backed combine AI capabilities with sustainable competitive advantages in their target markets.

Market Transformation: The most compelling opportunities use AI to transform existing industries rather than creating new AI-first categories. Several of our F24 investments follow this pattern — using AI to enhance operations while building moats in established markets.

The next wave of AI companies will likely focus less on AI as a feature and more on fundamental market transformation. YC’s recent request for startups reflects this, emphasizing:

• GovTech opportunities as government modernization accelerates
• Tools for human-AI collaboration rather than pure automation
• Solutions for traditional industries being transformed by AI

The reality is that every modern startup must use AI — as fundamental as having a website. The differentiator is how strategically companies deploy AI to create sustainable competitive advantages.

Investment Strategy in the New Landscape

At Lobster Capital, our F24 investment approach reflects a stark contrast in early-stage investing philosophies. While one investor at demo day proudly announced backing 45 companies, so far we made three carefully selected investments.

This concentration strategy has proven effective — our portfolio consistently demonstrates that selective, thesis-driven investing outperforms spray-and-pray approaches in this new environment.

The companies we backed share two characteristics: exponential early revenue growth and business models that would remain compelling even in an AGI world. One of my YC investments has reached $100M+ in revenue and profitability after a single seed round.

The metrics that matter have shifted. When evaluating F24 companies, we look for:

• Revenue velocity rather than just growth
• Team leverage (revenue per employee)
• Market positioning that strengthens with AI rather than being disrupted by it

Our optimism about YC’s quarterly batch structure stems from seeing these patterns emerge more frequently. The compression of company building timelines means each batch will likely produce more mature, revenue-generating companies than its predecessors.

Looking ahead to Winter ’25, we’re particularly excited about:

• Companies transforming traditional industries with AI-enabled efficiency
• Space tech opportunities (historically strong in YC but underrepresented in F24)
• GovTech modernization plays
• Web3 infrastructure, particularly in stablecoins

F24 suggests we’re well into an era where capital efficiency dominates growth at all costs, and where seed rounds might increasingly be the only institutional capital many successful companies need. For Lobster Capital, this validates our focused investment strategy and reinforces our conviction that the best returns come from identifying and backing transformative companies early, even in — perhaps especially in — smaller batches.

The startup world loves a good narrative. As someone who’s built Lobster Capital around investing exclusively in YC companies, I’ve heard them all: “YC has gotten too big,” “the valuations are crazy,” “it’s just an AI factory now.”

These stories persist despite Y Combinator’s continued ability to produce world-changing companies. The numbers tell a different story — YC’s portfolio of nearly 5,000 companies is now worth north of $600 billion, possibly approaching an unprecedented $900 billion.

After nearly a decade of investing in YC companies, I’ve developed a front-row view of what’s really happening inside the world’s most influential accelerator.

Myth 1: “YC Valuations Are Way Too High”

I hear this constantly. It’s the first objection most investors raise, and understandably so. When a YC company asks for a higher valuation while raising less money than their non-YC peers, it can seem disconnected from market dynamics.

But let’s look at the data.

The current median seed deal in the broader market sits at $3.1 million with a $12 million pre-money valuation. While YC companies often command premium valuations, they also typically sell less equity (around 10% compared to the standard 20% for non-YC seed rounds).

Take the return metrics for investors who backed at least three YC startups per batch between 2018–2020:

• Top performers delivered 16x their initial investment
• Top quartile returned 8x
• Even the median performance hit 5x
• Bottom quartile still managed 3.3x

For context, in traditional venture capital, only the top 25% of funds achieve returns above 3x. YC companies’ bottom quartile performance beats what most VC funds consider a win.

Beyond early-stage outperformance, YC companies that go public do so at an average market cap of $15 billion — three times higher than their peers. If in the end your portfolio includes companies like Airbnb, these “high” entry valuations start to look like bargains.

Think of my experience as a real estate investor in Paris. Everyone told me not to invest there because it was ‘too expensive’ and to look at smaller cities instead. But Paris, being a uniquely concentrated market, saw steady price appreciation while those ‘cheaper’ alternatives struggled — even during COVID. Sometimes it’s better to buy something that’s expensive but has proven staying power than to chase bargains that don’t appreciate.

The same applies to YC companies — they may be more expensive, but they are filtered and price in YC’s proven ability to boost startup success rates. With 39% of YC companies reaching Series A and 18% hitting $100 million valuations, investors are betting on a system built to create category-defining companies.

Myth 2: “YC Has Become a B2B Factory”

While B2B represents 45% of YC’s current portfolio, this overlooks YC’s continued success across diverse sectors.

The numbers speak to YC’s breadth. 81 hardware startups have gone through YC, building everything from satellites to rockets to electric planes. Deep tech companies from the Summer 2024 batch showcase this diversity — Spaceium secured $86.1 million in commercial contracts for spacecraft refueling stations, while NetworkOcean develops underwater data centers for sustainable GPU operations.

YC’s track record in consumer markets remains strong, with international successes like Rappi and Meesho demonstrating the program’s ability to scale consumer businesses globally. The crucial factor remains traction and market demand, not sector classification.

At Lobster Capital, we focus on the fastest-growing companies from each batch regardless of sector. The key is finding companies making products people want, whether those people are enterprises, developers, or consumers.

YC adapts to market opportunities while maintaining sector diversity. When B2B SaaS was ascendant, they backed those companies. As fintech gained momentum, they supported that wave. Now with AI transforming every industry, YC companies span both enterprise and consumer applications of this technology.

But the reality is also that B2B startups are much more likely to become Top YC Companies.… because that’s where the money is!

YC knows this and factors it into admissions decisions, which may help explain the sector’s perceived dominance.

Myth 3: “YC Has Lost Its Edge”

Critics often point to Airbnb and Stripe as ancient history, suggesting YC’s best days are behind it.

Once again, data paints a different picture.

Recent success stories abound. Jeeves (one of my investments) went from YC seed funding in 2020 to a $2.4 billion valuation. PostHog scaled to become infrastructure for 65% of new YC companies. Multiple companies from 2022–2023 batches secured $70M+ Series As. The list of private unicorns keeps growing: Razorpay, Deel, Rippling, Brex, Faire, Gusto, Scale, Rapyd, Papa, Sendbird — each representing billions in value creation.

Market critics misunderstand the timeline of success. The average startup takes 6.5 years to exit. With the IPO window largely closed since 2022, many scaling companies from 2015–2020 batches are still in their growth phase. YC companies demonstrate stronger survival rates through three key advantages:

• Superior capital access
• Extended runways
• Network support

The move to smaller, quarterly batches in 2025 strengthens YC’s model further. Each group partner works with 15–20 companies, maintaining the high-touch mentorship that defined YC’s early days. New group partners like David Lieb from Google Photos and Tom Blomfield of Monzo bring fresh perspectives while preserving YC’s core philosophy.

In fact, as YC grows in size, so does its real strength — network effects. Every successful company strengthens the ecosystem, whether through hiring, investing, or mentoring the next batch. This compounds over time and today’s YC founders can tap into thousands of alumni across every sector and stage of company building.

At Lobster Capital, we see this network advantage firsthand. The best YC companies often have access to customers, talent, and capital before they even leave the program. This early momentum, combined with YC’s proven playbook, creates a foundation for exponential growth.

Myth 4: “YC Should Be Geography Agnostic”

The push for remote-first, geography-agnostic programs sounds compelling in theory. But YC’s return to San Francisco-based batches reflects a fundamental truth about building global technology companies.

If you want to build a top tech company, move to The Valley.

The data does show YC’s global impact: 40% of YC companies come from outside the U.S. — India (205 companies), UK (146), and Canada (145). Cities like Bengaluru (114 startups) and Toronto (77) have also emerged as powerful nodes in the network.

But San Francisco remains magnetic for a reason. The city functions as a global talent aggregator, drawing ambitious founders from every corner of the world. Take recent YC founders — born in Egypt, Haiti, France, and dozens of other countries — who chose San Francisco to build their companies.

The advantages are structural:

• Access to the world’s deepest pool of venture capital
• A unified market of 360 million English speakers
• Single regulatory framework in the US unlike Europe
• Concentrated technical talent

YC’s in-person program accelerates these benefits. While remote batches during COVID proved possible, we all know the serendipitous connections, peer learning, and network building happen most effectively face-to-face. Bengaluru might have excellent engineers and Paris might have amazing AI talent, but San Francisco combines all these elements in one ecosystem.

The American market’s size and unity create unique scaling opportunities. European startups often wrestle with 14 languages and 28 regulatory frameworks. In contrast, U.S.-based companies can focus on rapid growth within a massive, homogeneous market before expanding globally.

This is not written to diminish other ecosystems. The numbers just show where founders can best maximize their odds of success.

When YC companies raise more capital and achieve higher valuations than their peers, location plays a crucial role.

YC in 2024 is not the same program that funded Airbnb, Reddit, or Stripe. Yes.

But… The fundamentals remain: unmatched deal flow, proven playbook, powerful network and the shift to quarterly batches represents focus, not compromise.

For founders and investors alike, the message is clear: judge YC on data, not discourse. A 6% unicorn rate, 39% Series A conversion, and a portfolio worth nearly $900 billion tell the real story.

The next Airbnb won’t look like Airbnb. But it’s probably being built in YC right now.

Y Combinator’s Summer 2024 (S24) batch just wrapped up, showcasing 250 startups that completed 3 months of the world’s most storied accelerator.

After witnessing their Demo Day presentations in San Francisco, I’m more convinced than ever of the power of the Y Combinator ecosystem.

In this post, I’ll share my insights on the standout trends and how Lobster Capital approaches investments in this landscape.

Traction is King

At Lobster Capital, we’ve developed a proprietary selection method that focuses on startups demonstrating early, significant traction. Our criteria are stringent:

- The startup must show exponential growth in dollars, not just users

- This growth should be sustained over several months

- The traction must be repeatable and defensible

While YC companies are early-stage by definition, we place a premium on those that have already demonstrated market validation through revenue.

This approach has led us to fight for allocation in some top companies. In this cohort we invested in a company that achieved the highest revenue of the entire batch of 253 startups.

We also evaluate how founders have acquired customers and their intentions for growing their user base. I want founders to have a way to get a new customer while they’re sleeping — a scalable, automated approach to growth.

What to Do with 200 “AI Startups”?

An unprecedented 75% of startups in this YC batch were working on AI-related products. While this is a given — soon using “AI” will be no different from having an app, or using the cloud — I remain cautious about startups leading with AI as a value proposition.

However, we’re not shying away from AI investments. We’ve backed several promising AI startups from this batch, including one that’s serving thousands of customers, from individuals and SMEs to teams at Google and ByteDance.

But what we always come back to is a focus on solving tangible, real-world problems. “Make something people want,” is undefeated, and where AI is just a value-add doesn’t seem like it will be enough.

There’s a noticeable split in the S24 batch:

1. Tools for AI or developers using AI

2. Using AI to transform traditional industries

We lean closer to seeing long-term value in the latter. Startups like Elayne (estate planning and settlements), Baseline AI (automating clinical trial data), and Passage (helping companies get goods through customs) are other solid companies taking this approach.

Thriving Deep Tech

When people think of YC, they often picture software startups and consumer apps. But YC has through the years been consistent in showcasing groundbreaking deep tech startups, this batch was no different.

What’s impressive is how YC pushes even deep tech startups to generate revenue within the short span of the program. This approach challenges the notion that hardware and science-based startups need years before showing market traction.

Spaceium, for example, are building fully automated space stations for spacecraft refueling and repair — a moonshot idea by any standard. Yet, within YC’s three-month program, they secured $86.1 million in binding commercial contracts and an additional $230 million in the pipeline.

Others like, NetworkOcean is building data centers underwater to cut power usage, and run GPUs more sustainably, and has GPUs available now.

Theseus develops unjammable drones. They’ve sold units to multiple customers, have six-figure LOIs from large drone manufacturers, scheduled integrations, and eight-figure funding requests from the DoD for FY25.

Lastly, Ionworks — who provide battery-simulation-as-a-service to accelerate battery design cycles. They have traction with a multi-billion dollar silicon anode company, two fortune 100 tech companies, and a public automotive OEM.

YC’s “make something people want” philosophy applies even to the most complex and futuristic technologies. In general, companies in the defense/space spaces are more VC-investable because something like a $60 million Air Force contract can offset much of your early-stage R&D costs — if you can clear the bureaucratic hurdles to winning a government contract — but the breadth of S24 Deep Tech has been impressive.

Looking Ahead

With this S24 batch, and the upcoming Fall 24 batch, some other areas I’m interested in are:

1. Battery Tech — batteries are projected to represent 4% of global GDP by 2030, the potential for innovation here is enormous, even beyond companies like Ionworks mentioned above.
2. Workflow Automation: Tools solving unique workflow problems, largely by integrating AI, I like to see these go as niche as possible — e.g. Abel Police are automating police paperwork.
3. Real Estate Tech: Companies like Modern Realty addressing the real estate market, with or without AI. This sector might still be underappreciated given its size.

2024 is of course the last year with the traditional YC format. In 2025 we will move to quarterly batches of about 100 companies each.

For Lobster this will spread our deal flow more evenly throughout the year, allowing for deeper evaluations and more focused interactions with founders.

What’s particularly impressive about YC is its ability to attract diverse talent. Nearly 50% of the founders in YC S24 have worked at some of the world’s most renowned companies, with a quarter coming from the big five — Google, Amazon, Meta, Microsoft, and Apple.

As ever beauty of YC is its ability to surface diverse, often unexpected innovations.

With each batch, the quality of startups only improves. As the startup landscape becomes increasingly competitive, the bar is higher, but the path to revenue is more accessible than ever for those who innovate and execute swiftly.