Discovered Materials Plays AI Whack-a-Mole to Hunt Cooler Chips

Discovered Materials, a startup founded by Advaith Sridhar and Akash Ramdas, is using swarms of AI agents to find new materials for building more efficient integrated circuits, addressing the heat problem in AI chips. The company recently closed a $9 million seed round led by Lightspeed India Partners, with participation from Peak XV Partners and angel investors Paul Graham, Gokul Rajaram, and Thariq Shihipar, after emerging from Y Combinator. The founders combine Ramdas' PhD in materials science from Stanford with Sridhar's experience working on agents at Persona AI and Luma Labs. Their software pipeline uses Anthropic models in a custom harness to generate material leads, then verifies them using foundational physics models they trained. The company claims to run thousands of guesses per day, compared to about 20 per day during Ramdas' PhD. They released examples of hundreds of new materials and their Material Discovery Bench today. Competitors like MatNex, SandboxAQ, and CuspAI are also working on similar efforts, but Discovered Materials focuses on thermal problems of semiconductor materials. The startup has already found materials matching properties of those used by major chipmakers but can't share details. A challenge is the engineering trade-space, as materials that improve heat dissipation might be hard to manufacture or have compromised electrical properties. Hemant Mohapatra of Lightspeed, who led the round, compares it to whack-a-mole with atomic structures. He expects material prediction will become commoditized, but Ramdas' expertise and rapid lab validation give Discovered Materials an edge. When valuable candidates are found, the company plans to patent the use of materials in GPUs or manufacturing processes and license them to chipmakers. However, no AI-discovered materials have made commercial impact yet; the closest is Insilico Medicine's drug Renterosib in Phase II trials. Mohapatra believes the bottleneck is filtering and synthesizing candidates, not finding them. Sridhar acknowledges the process involves wet labs and cannot be sped up, but their unique data and expertise will help compete with frontier labs.