A $9M bet that AI agents can hunt down cooler-running chips before data centers melt

Data centers are burning through electricity at staggering rates, and a big chunk of that power goes straight to cooling chips that can’t handle their own heat. Discovered Materials, a fresh Y Combinator grad, just closed a $9 million seed round to throw AI at the problem. Lightspeed India Partners led the round, with backing from Peak XV Partners and angels Paul Graham, Gokul Rajaram, and Thariq Shihipar.

Founders Advaith Sridhar and Akash Ramdas built a software pipeline that deploys Anthropic models as agents to generate material leads, then runs them through custom physics models for verification. The scale difference is stark. Ramdas spent his Stanford PhD making maybe 20 guesses a day. “We’re able to do thousands of guesses a day now by having these agents run 24/7 on the cloud,” Sridhar told TechCrunch. The startup dropped hundreds of new material candidates and a “Material Discovery Bench” today to track how frontier models perform on this challenge.

But the real-world physics of chipmaking makes this a brutal optimization problem. A material that reduces heat might be impossible to manufacture, or it might wreck electrical performance. “It’s a bit of playing whack-a-mole with atomic structures,” said Lightspeed partner Hemant Mohapatra, who led the round. The startup is betting that Ramdas’s lab experience gives them an edge in rapidly testing and validating candidates rather than just generating more of them.

Here’s the reality check: no AI-discovered material has made a dent commercially yet. Insilico Medicine got a drug into Phase II trials, and MatNex found promising rare-earth-free magnets, but that’s as far as it’s gone. Mohapatra sees candidate generation becoming commoditized; the bottleneck is filtering and synthesis. Sridhar acknowledged the hard truth that “a lot of this will involve actually going into wet labs and making things as well. And this is the process that cannot be sped up.” The company plans to patent useful discoveries and license them to chipmakers, with hopes of filing something worthwhile within a year. Whether silicon valley’s patience matches the timeline of actual lab science is another question entirely.