At a glance
WHAT IT’S REALLY ABOUT
YC sees AI driving startups from software bits into physical atoms
- YC reports a sharp rise in hard tech startups (8% → 20%), led by robotics, re-industrialization, defense, and compute/power infrastructure companies.
- AI (especially code generation and agentic development) is reducing the software labor constraints that historically made hard tech slow and expensive to scale.
- Exploding demand for AI compute is creating opportunities across physical infrastructure—data centers, energy/power, interconnects, and alternative chip architectures.
- Software is evolving toward agent-driven “harnesses” that complete end-to-end tasks, which YC links to faster revenue growth and higher willingness to pay.
- YC sees founder demographics shifting: more solo founders (5% → ~19%) and a resurgence of experienced founders who can “manage agents” and choose what to build.
IDEAS WORTH REMEMBERING
5 ideasHard tech is back: YC’s batch composition is rapidly shifting from “bits” to “atoms.”
YC reports hard tech rising from 8% to 20% of accepted companies, with notable jumps in robotics, industrial manufacturing, defense, semiconductors/photonics, and power. The episode frames this as a structural shift driven by AI, geopolitics, and compute/energy constraints—not a temporary fad.
AI is making hard tech easier by compressing the cost and time of the software layer.
They argue code generation and agentic coding reduce the “software engineering bottleneck” for hardware-heavy startups (e.g., Anduril-style full-stack systems). Smaller teams can now ship sophisticated software layers faster, changing the economics of building planes, chips, robots, factories, and defense systems.
AI compute is turning into a physical infrastructure and supply-chain problem.
Demand for GPUs and data centers is so strong that older GPUs (e.g., A100s) are described as appreciating—an inversion of typical hardware depreciation. This pushes startups into physical build-outs: data center construction, power, batteries, interconnect/switching, and alternative silicon architectures.
Robotics is approaching a breakout, but the bottlenecks are data, real-time control, and deployment.
The hosts describe a looming “ChatGPT moment” for robotics, fueled by progress in models and benchmarks, but emphasize robotics needs new scaling laws, real-world data, and deployment infrastructure. They also stress that robotics is not just one product category—it’s an ecosystem of vertical solutions, tooling, and data pipelines.
Software’s center of gravity is shifting from tools to agent-run end-to-end “harnesses.”
They argue SaaS isn’t dead; instead, “system of record” products must become “harnesses” where agents execute work end-to-end. YC sees “full task” companies rising from ~10% to 25%+ of the batch, reflecting a shift from point tools to products that complete workflows (e.g., clinical intake, insurance brokering, billing).
WORDS WORTH SAVING
5 quotesThings that actually touch atoms and not just bits.
— Garry Tan
I mean, that's the true bull case for Hard Tech. It's that it's not just that people are shying away from funding software businesses, but it's actually that the super smart models that we have now are actually accelerating scientific research and making it possible for startups to have bigger research breakthroughs earlier, and that therefore these di- deep tech companies will actually work better.
— Jared Friedman
We right now are about almost a year since agentic coding started to work, since Opus 4.5, that we're seeing these workflows fully blossom, and the result are basically people want their job just be done and are willing to buy software that just gets the job done.
— Diana Hu
One of the things that we started experiencing this year that we never experienced in the past is we have companies breaking from zero to seven figures in revenue during the batch, and that is in a span of three months, and that's shocking.
— Diana Hu
What a w- weird moment we are in history where you wake up in the morning, you, like, wire up a new model, and then these things that even a month ago you're just like, "Why isn't it working?" It just starts working.
— Garry Tan
High quality AI-generated summary created from speaker-labeled transcript.
