The Shift Nobody’s Talking About Directly
When I pulled the data on Y Combinator’s Winter 2025 cohort, the first thing that jumped out wasn’t what got the most press coverage. Yes, there were the usual consumer apps. Yes, there were the obligatory fintech plays. But the real story lives in the concentration: roughly 40 percent of the 170-plus companies in the batch are building AI infrastructure or developer tooling. That’s not a trend. That’s a tectonic shift in how capital allocates when it’s actually thinking about returns rather than narrative.
Compare that to 2021 and 2022. Back then, Y Combinator was shipping batches loaded with consumer applications built on top of other platforms. Venture capital was playing the growth-at-all-costs game, and accelerators followed. The median pre-money valuation for YC companies at Demo Day was sitting comfortably above $30 million. Founders could raise at that number because investors believed in the greater-fool thesis: someone else would pay more tomorrow.
That era is dead. The companies in W25 are settling for median pre-money valuations around $20 million at Demo Day. That’s not a modest correction. That’s a 30 to 35 percent markdown from the peak. And here’s what matters: the founders accepting these terms aren’t desperate or weak. They’re rational operators who understand that a real valuation beats an inflated one that makes the next round impossible.
Follow the Money Into Infrastructure and Defense
The fastest-growing category in this batch isn’t what I expected to lead. It’s defense tech and hard tech. Not because Y Combinator suddenly got patriotic, and not because some partner made an impassioned speech at batch orientation. It’s because venture investors finally accepted what a few contrarians have been saying for years: the returns in infrastructure are real, compounding, and defensible in ways that consumer apps simply aren’t.
Defense tech specifically represents a thesis that seemed fringe five years ago. Peter Thiel spent a decade warning that American technological advantage was eroding and that private capital needed to build for national security. Most of Silicon Valley ignored him or nodded politely before going back to their seed decks. Now it’s mainstream Y Combinator curriculum. Founders are building tools for military applications, supply chain security, and critical infrastructure hardening. And investors are writing checks.
This matters operationally because the sales cycles are different. Defense tech has customer concentration risk but also customer concentration stability. A startup closes one $5 million deal with the Department of Defense, and suddenly the business looks completely different. It’s not sexy. It’s not the Instagram story that gets shared at Sand Hill Road cocktail parties. It’s exactly the kind of boring, high-margin, defensible business that actually returns capital.
Y Combinator’s Deal Terms Changed the Math
In 2024, Y Combinator quietly updated its standard investment terms. The organization moved from its historic $125,000 for 7 percent equity to $500,000 for that same 7 percent. On the surface, this looks like Y Combinator just got more generous. Actually, it’s the opposite.
That $375,000 difference fundamentally changes the cohort composition and the expectations baked into the accelerator economics. You can’t raise $500,000 rounds if you’re planning to build the next Instagram. You need a real problem with real customers willing to pay immediately. The math forces founders toward B2B infrastructure, developer tooling, and enterprise applications from day one. Consumer businesses require a different fundraising approach now, and many strong consumer founders are simply opting out of Y Combinator because the match is worse.
This is actually smart design. Y Combinator is self-selecting for founders who have product-market fit insights or clear go-to-market strategies before batch starts. The accelerator is no longer a blank canvas for raw ideas. It’s a turbocharger for businesses that already know their direction.
The Valuation Multiples Tell the Real Story
If you want to understand where venture capital actually believes the returns are, stop looking at pitch decks. Look at the multiples. According to PitchBook 2026 Venture Monitor, seed-stage AI infrastructure companies are trading at approximately 18x revenue multiples. That’s not a typo. It’s eighteen times annual recurring revenue for unproven startups.
Now compare that to generalist SaaS businesses, valued at roughly 6x revenue at the seed stage. You’re looking at a three-fold valuation premium for AI infrastructure specifically. That gap doesn’t exist because investors are irrational. It exists because infrastructure businesses have better unit economics, faster customer acquisition, and lower customer acquisition costs. The multiple reflects reality.
But here’s the operator’s caveat: those multiples also encode massive expectations. If you’re raising at 18x revenue, you’re committing to growth trajectories that don’t tolerate mistakes. You need to execute on your go-to-market thesis from day one. You can’t learn slowly. The valuation premium is a performance premium, full stop.
What This Means for Founders and Operators Right Now
The W25 batch is telling us something clear: venture capital has stopped betting on network effects and brand moats for early-stage companies. It’s betting on technical defensibility, immediate revenue traction, and businesses that serve other businesses building infrastructure. Y Combinator W25 Demo Day coverage confirmed this pattern company after company.
If you’re an operator thinking about starting a company or raising capital right now, the opportunity lies in the gaps. Consumer applications are out of favor at Y Combinator, which means you might actually be able to fundraise for them at reasonable valuations from other sources. Developer tools and infrastructure are crowded and well-funded, which means competition is real and your technical edge needs to be genuine. Defense tech is hot, but it requires long sales cycles and regulatory knowledge that most founders don’t have.
The most interesting position right now isn’t following the herd into AI infrastructure. It’s identifying where capital has gotten so concentrated that it’s creating opportunities in the categories investors are supposedly ignoring. That’s usually where the best returns hide.
What patterns are you seeing in your own market? I’m genuinely curious whether this shift feels real in your industry or whether it’s still mostly hype. Drop a note if you want to dig into any of this further.