What did Y Combinator fund in 2025?

The 2025 numbers

The portfolio now runs past 5,600 companies since 2005, with a combined valuation above $600 billion and 82 unicorns produced, a 4% hit rate against roughly 2.5% for comparable venture-backed seed companies. Around 87% of YC companies are still operating, against something closer to 50% for typical startups at five years, and about 45% reach a Series A where the industry average sits near 33%.

The structural news of 2025 was the move from two batches to four.

Batches are growing while the acceptance rate holds, which means YC is reviewing more applications rather than relaxing the filter. The newer Spring and Fall batches are smaller and investors are still learning how much attention to give their demo days, so they carry a different competitive profile from Winter and Summer.

The AI takeover, and what it means if you aren't AI

88% of the S25 batch was AI-native, meaning the core product wouldn't exist without AI. Of that batch, 141 companies were AI-native, 9 were AI-adjacent, and 10 had no AI at all. More than 60% put AI in their one-line pitch.

The flavour shifted too. 2024 was the year of copilots that help a human work. 2025 was the year of agents that take the workflow away, with about half the Spring batch building agentic systems and coding agents showing the strongest revenue traction of any agent category. A notable sub-trend: companies building tooling for other companies' agents, which is to say evaluation frameworks, observability and testing infrastructure.

If you're not AI-native you're competing for something like 12% of the places. Biotech, hardware and parts of fintech still get in, so the door is open, and the bar is higher. The failure mode I see is founders bolting AI language onto an application they can't defend, which partners spot immediately. Owning the absence and arguing why your market doesn't need it beats trend-surfing.

Where the money went

• Developer tools and infrastructure, around 30%. The largest single category, with 20 software development companies in S25 alone.
• Fintech, 15%+. YC became the most active fintech investor globally in 2025, joining 151 deals, up 24.8% on the 121 it did in 2024.
• Healthcare and biotech. Consistent across all four batches, with the Fall cohort focused on AI as operational backbone. These companies raise larger seeds, $4.6M median against $3.1M overall.
• Defence and dual-use. Five companies in S25, a category that would have been unthinkable at YC five years ago.
• Voice AI. 16 companies in S25, building regulated enterprise interactions rather than chatbots.

The B2B skew is extreme, at roughly 80 to 85% of companies. What that pattern tells you is that YC backs markets it believes are about to change structurally, so when you apply you're making a claim about your market as much as your company. A founder who can't name the structural shift driving their sector is at a real disadvantage.

What the $3.1M median hides

Every accepted company gets $500,000 on the standard deal: $125,000 on a post-money SAFE for 7%, plus $375,000 on an uncapped SAFE with an MFN provision. What happens after demo day varies enormously.

The top of a batch raises $3 to 4M+ at $20 to 25M post-money, often oversubscribed, usually with $150K to $500K ARR already. The middle raises $2 to 3M at $18 to 22M over four to eight weeks. A large share of every batch takes much longer, and some never close a seed at all. Standouts from the 2025 batches included Harper, the AI insurance brokerage out of W25 that raised $47M, and BUBrowser at $17M.

The YC brand gets you a warm introduction to every investor in the market. It doesn't close the round. What closes the round is the same thing that closes any round: traction, a clear narrative, and a process someone is running properly.

Demo day is a launch window, not a finish line

This is where I see YC founders leave the most value on the table, and it's a process point rather than a YC point.

Demo day is a fixed date that every relevant investor already has in their calendar. That is a rare gift, because the hardest part of engineering momentum in a normal raise is getting investors to move on the same timeline. The founders who convert it best treat the weeks before as an activation window: they've already built relationships with target investors during the batch, they confirm in advance who is coming and who will take a meeting, and then they let the meetings land together in a compressed block afterwards. Density is what creates competitive pressure. The same set of meetings dripped across two months produces none of it.

The founders who struggle do the opposite. They treat demo day as the starting gun for relationship-building, then spend eight weeks trickling out intros while their batchmates' rounds close around them. If you're going through the programme, run your fundraise process in parallel with it. The mechanics of doing that are in how to build momentum in a funding round.

What this changes if you never apply

The data reshapes the market you're raising into, so it matters even at a distance. Every investor now sees 88% AI-native deal flow from the biggest accelerator in the world, which means your pitch has to address the AI question one way or the other. $3.1M has become a reference point for what a seed looks like. The consumer founder needs a sharper narrative to stand out in a market that has decided enterprise is where the money is. And the San Francisco concentration, up from 21% in 2021 to 73%, tells international founders something uncomfortable about where relationships get built.

Gian Seehra Ex-Octopus Ventures, part of deploying $200M as a VC. 3x founder, raised $16M. Has coached 120+ founders who have raised $250M+ collectively.