YC Demo Day Records Are Shattered… Again.
·
6 min read
·
Mar 24, 2025
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.