Y Combinator’s View of Startups in 2026: Hard Tech, AI and Solo Founders

The Lightcone participants discuss changes they say Y Combinator is seeing among founders in 2026. Their figures describe YC’s own accepted companies and should not be assumed to represent all startups.

Diana says the share of hard-tech companies in the batches analyzed rose from 8% to 20%. She also reports that median monthly revenue by the end of a batch rose from roughly $8,000 to $20,000; the comparison is the team’s internal analysis.

The group describes growth across robotics, industrial manufacturing, defense, semiconductors and power infrastructure. These categories may overlap, so the individual shares are not a breakdown to add together.

Jared says one in six founders in the current summer batch has a PhD. The panel connects the change to more research-heavy areas, where specialized expertise can be important.

The optimistic case for hard tech, they argue, is that code generation and stronger models can help small teams tackle software and research work that previously required more staff. They do not claim physical supply chains have disappeared.

The panel cites space companies and defense technology as sources of founder interest. They discuss startups building satellites, power systems, aircraft and counter-drone equipment, including companies they have funded.

They also see opportunities in dual-use supply chains and manufacturing. One example is a metal supplier serving newer defense companies that need materials delivered on a faster schedule.

AI compute has a physical side: data centers need construction, energy, batteries, chips and networking. The panel says rising demand creates room for companies throughout that infrastructure stack.

They describe startups exploring specialized processors and optical interconnects. These are company examples and hypotheses about performance; the discussion does not independently benchmark the products.

The group expects robotics to benefit from advances in AI and sees activity in both vertical applications and supporting infrastructure. They call a general breakthrough a possibility, while recognizing that it has not yet arrived.

Historically, capital-intensive companies could be difficult to fund when investors preferred software subscriptions. The participants say they are now seeing more investor interest in hard tech, though sentiment can change.

The panel does not consider software obsolete. A system of record can remain valuable if agents need it to access information and perform work; developers may increasingly design interfaces for both agents and people.

Returning to YC revenue figures, they argue that some AI products are more valuable because they perform a larger part of a workflow instead of merely recording it. They acknowledge that enthusiasm for AI can also influence spending.

A recruiting example illustrates the distinction: software can find candidates and assist with outreach and scheduling, while human recruiters may focus on judgment and relationships. The panel presents this as an example, not a universal outcome.

The speakers say some YC companies reached seven-figure revenue within a batch, much faster than they had typically observed before. They attribute this to a combination of real demand and faster product development.

They identify another fast-growing category: companies selling data and reinforcement-learning environments to model labs. They report that more than a dozen YC companies funded in the preceding two years exceeded $10 million in annual revenue in this area.

Specialized environments can combine domain expertise, software and evaluation for tasks such as finance. The panel also mentions demand for physical-world data, including first-person and robot-task recordings.

They speculate that businesses with proprietary data may increasingly adapt or train models for their own workflows. The value depends on the data, task and costs; it is not guaranteed by fine-tuning alone.

In robotics, one participant hypothesizes that domain-specific models and fine-tuning may be necessary because physical tasks have different timing and control requirements. They cite companies adapting a general robotics model to narrower applications.

Diana says solo founders rose from about 5% to 18–19% of accepted companies in the comparison discussed. The panel thinks better building tools can help one person get started, while still valuing complementary co-founders.

They discuss founders who started alone and later added partners. A solo start can change when equity and responsibilities are negotiated, but does not remove the need for sales, technical work and customer understanding.

The panel sees a renewed role for experienced founders in their late thirties, forties and beyond. Knowing a market, recognizing difficult problems and deciding what to build can become more valuable as prototyping gets faster.

They compare directing multiple agents with some aspects of managing engineering teams, such as setting goals and reviewing work. The analogy is partial: agents have different reliability and accountability constraints.

Their practical advice is to experiment with current tools and revisit problems that were previously difficult. The group is optimistic about continued progress, but acknowledges uncertainty about how long the pace will last.