Discovered Materials is playing AI whack-a-mole to hunt cooler chips
TechCrunch Tim Fernholz
Discovered Materials is using AI agents to hunt better chip materials. The bet: smarter materials could mean cooler, less power-hungry chips.
Based on reporting by TechCrunch, Tim Fernholz — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AI has a heat problem, and the data centers feeding it are paying the bill. Discovered Materials thinks the fix starts lower down, at the material level, and it’s using AI to search for better ingredients for integrated circuits instead of just more efficient software tricks.
The startup just closed a $9 million seed round led by Lightspeed India Partners, with Peak XV Partners and angels Paul Graham, Gokul Rajaram, and Thariq Shihipar also backing it. It came out of Y Combinator, and its founders bring a useful mix: Advaith Sridhar worked on agents at Persona AI and Luma Labs, while Akash Ramdas earned a doctorate in materials science at Stanford.
Their setup is a two-step machine. Anthropic models, wrapped in a custom harness, generate candidate materials. Then the company’s own physics models run simulations to check whether those candidates are actually worth chasing. Sridhar says that during Ramdas’ PhD, he might have managed about 20 guesses a day. Now, he says, the system can make thousands of guesses a day, running nonstop in the cloud.
Discovered Materials also released hundreds of new materials examples and a Material Discovery Bench to track how frontier models handle the problem. It’s entering a crowded space: MatNex, SandboxAQ, and CuspAI are all working on similar ideas. But Discovered Materials is narrowing in on semiconductor thermal issues, the messy part where a material can look promising on paper and still fail because it’s too hard to manufacture or its electrical properties fall apart.
The company says it has already found several materials that match the properties of ones used by major chipmakers, though it won’t say which ones. If those candidates hold up, Sridhar says the plan is to patent their use in GPUs or the process for making chips from them, then license that IP to chipmakers. He hopes the first patent-worthy material shows up within a year. Even so, the hard part may still be ahead: not finding candidates, but filtering them correctly and getting them made in wet labs, which the founders admit can’t be sped up.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of AI bet: not another chatbot with a logo, but a search engine for physics. Still, the industry loves “we found lots of candidates” stories because they’re cheap; the real scorecard is whether anything survives the wet lab and gets built at scale. Until then, the hype machine can take a seat by the fume hood.
Read more about this at: TechCrunch