Y Combinator shifts its bet from software to hard tech, robots, chips and lunar manufacturing

Y Combinator, the accelerator long associated with "two founders and code in a garage," has sharply rewritten its public investment thesis. In its summer Request for Startups 2026 list, the fund signals that it expects the next wave of big companies to come not only from software, but from AI projects at the intersection of hardware, manufacturing and heavily regulated industries.

YC's new course

Y Combinator published its Summer 2026 Request for Startups in late April 2026, just days before the deadline. The document lists 15 types of companies that the accelerator's partners want to see in the new cohort. Eight of these areas cannot be built by a team of developers and a cloud account alone: they require capital expenditure, proprietary hardware or work with physical infrastructure. For YC this is a notable turn, because its image was long built around startups that could be assembled quickly, cheaply and almost entirely in software.

The way the list is written matters too. Each category is overseen by a specific partner, so it reads less like an abstract wish list and more like a set of concrete bets the fund is ready to make now. The spring 2026 list was almost half as long and mostly revolved around AI layers on top of familiar software. The summer version, by contrast, shows that YC believes the next big wave of companies will grow where AI meets robotics, manufacturing, logistics, defense and other complex industries that classic SaaS entered only weakly.

Where the money is going

Among the new priorities are areas that, not long ago, a classic accelerator would have considered too heavy in development cycle and risk. The list includes AI for agriculture that minimizes pesticides, defense systems against drone swarms, inference chips for space and even lunar manufacturing from molten regolith. Software for semiconductor supply chains gets separate emphasis: a single modern chip goes through about 1400 manufacturing steps, crosses more than ten countries and takes months to produce.

• AI for targeted treatment of weeds and pests
• Software-defined defense against drone swarms
• Space inference chips designed around mass, temperature and radiation
• Lunar manufacturing of materials from regolith
• Software for managing semiconductor supply chains

This does not look like a set of exotic ideas chosen for hype. Each category in the document rests on clear market logic: defense budgets are growing, demand for AI chips keeps accelerating, and reusable rockets are lowering the cost of putting hardware into orbit. Where venture capital once feared long cycles and thin margins, there are now examples of companies on the scale of SpaceX and Anduril. In effect, YC concedes that the thesis that hardware companies cannot deliver venture returns no longer works as a universal rule.

Software is changing too

YC is not abandoning software, however. It is just defining differently what kind of software the market now needs. Seven of the fifteen categories remain purely software, but this is no longer the old SaaS playbook with yet another dashboard for humans. The document singles out Software for Agents, Company Brain, Dynamic Software Interfaces and SaaS Challengers. The idea is that future products must work not only with people but with AI agents, through APIs, machine-readable documentation, a permissions layer, identity and payment infrastructure.

That drives a new logic for replacing old markets. YC looks squarely at ERP, industrial software, chip design tools and supply chain management systems: categories where large vendors move slowly and charge a lot for updates. In parallel, the fund points to personalized medicine and stablecoin infrastructure, where the barrier is no longer the algorithms themselves but regulation, liability and institutional trust. In other words, the model and the interface are ceasing to be the main competitive advantage; what matters more is the ability to embed in a real process and change the economics of an industry.

What this means

For founders the signal is very direct: the era when you could win on code-writing speed alone is ending. If even Y Combinator, the symbol of the software startup, is publicly shifting its focus toward robots, chips, defense, manufacturing and agentic infrastructure software, then the scarce resource is no longer access to a model but the ability to bring AI into the real world. Put simply, a garage is no longer enough: now you need not only to write code, but to build the whole system around it.