Discovered materials raises $9 million seed to cool next-gen AI chips

Discovered Materials, a startup founded by Advaith Sridhar and Akash Ramdas, closed a $9 million seed round led by Lightspeed India Partners after graduating from Y Combinator with participation from Peak XV Partners and angels Paul Graham, Gokul Rajaram and Thariq Shihipar. The company tackles a bottleneck that limits every modern data center: AI chips emit excessive heat, forcing power-hungry cooling systems.

Ramdas, a Stanford PhD in materials science, and Sridhar, who previously worked on Persona AI and Luma Labs, built a software pipeline that stitches together Anthropic models fine-tuned to generate candidate materials and then runs in-house physics simulations to evaluate and filter them. “During his PhD he made maybe 20 guesses a day,” Sridhar told TechCrunch. “We can make thousands of guesses a day with the agents running 24/7 in the cloud, exploring directions he points them to.”

Today the firm released hundreds of new material samples along with the “Material Discovery Bench,” a tool designed to track how frontier models meet the challenge. Competitors such as MatNex, SandboxAQ and CuspAI operate in similar spaces, but Discovered Materials bets that a laser focus on the thermal properties of semiconductor materials offers a faster route to practical benefit. The startup claims to have already identified materials that match the specifications of existing chip-maker offerings, though it has not disclosed details.

“It’s a bit like playing whack-a-mole with atomic structures,” said Hemant Mohapatra, the Lightspeed partner who led the round. “A material is only useful in the real world if all the properties line up, and that’s what makes it a genuinely interesting search problem.” Mohapatra estimates that material prediction will become a commodity as models improve, and he points to Ramdas’s deep expertise and the company’s ability to run a lab that validates candidates quickly—a capability the duo has already demonstrated with several new substances.

The business model centers on filing patents on material use in GPUs or on manufacturing processes and licensing them to chip manufacturers. Sridhar hopes to have patent-worthy materials within a year.

Despite the enthusiasm, no AI-discovered material has yet achieved significant commercial impact. The closest analogue is Renterosib from Insilico Medicine, the first generative-AI drug to reach a phase-2 clinical trial. In the materials arena, promising candidates include rare-earth-free permanent magnets from MatNex and semiconductor compounds from Panasonic and Citrine Informatics, but none have been deployed at scale. Mohapatra emphasizes that candidate discovery is not the bottleneck; “proper filtering and synthesis are the bottleneck.”