Y Combinator held its 47th Demo Day on Thursday 10 September in San Francisco. The Summer 2026 batch - 234 publicly listed startups, sorted into 12 buckets by product and customer - presented to an invite-only room of roughly 1,500 investors and media. The batch is the clearest picture yet of where the accelerator that funded Airbnb, Stripe and Coinbase thinks the next decade is.
The day
The format has not changed: a minute or two per company, a directory investors have already combed through, and a second day for alumni. What changed is what walked on stage. Alongside the usual agent and infrastructure companies were a floating nuclear-powered data-center startup (Atomarine), semiconductor and optical-switching plays for AI clusters (Lamb Labs, Dipole Labs), an AI-guided defense company (Isengard Industries) and Nori, whose public YC page lists a $1,688 general-purpose robot that folds clothes and restocks shelves. Software was still there - Levocred AI says about $1B of receivables already run through its credit-fund platform - but it was no longer the whole show.
The shift
Two batches ago the story was "everything is an AI wrapper". This batch is about the physical bottlenecks AI created - energy, cooling, chips, robots. It is a harder, slower kind of company: real capital expenditure and years before profit, the opposite of the SaaS curve YC is famous for. The signal for the rest of the market is that the easy generative-software ideas are now assumed, not funded.
When the accelerator that taught the world to ship in a weekend starts funding floating reactors, the weekend-app category is not dead - it is just no longer where the scarce information is.
The deal, still
The terms have not moved: $500,000 for every company - $125,000 for 7% on a post-money SAFE plus $375,000 on an uncapped MFN SAFE - and the batch is still the fastest way to compress a year of fundraising into three months. YC now runs four batches a year, so a "no" is a four-month wait, not a twelve-month one. The full profile, terms and application link are on FindSeed.
What solo founders should take
Three things. First, YC has said openly that a large share of recent batches ship code that is mostly AI-generated; using the tools is table stakes, not a differentiator. Second, the software companies that still got in are selling into a specific customer with a specific number attached - receivables, pipeline, revenue - not a category. Third, the hard-tech turn opens a gap: the boring, profitable, one-person software business is exactly what the room was not looking at this week, and that is usually where the next cycle's best bootstrapped companies come from. Our list of 30 programs that back solo founders before revenue covers the alternatives if the YC odds look long.
Read the newsletter that top founders & VCs keep in their inbox.