AI searches for new materials that keep microchips from overheating

Modern chips that perform AI tasks generate so much heat that data centers consume large amounts of electricity and require complex cooling systems. According to TechCrunch, entrepreneurs are once again turning to AI itself to solve this problem. The startup Discovered Materials has announced plans to use AI agents to find new materials for creating more efficient integrated circuits. Techcrunch.com reports .

The new company announced that it had successfully completed the Y Combinator program and closed a $9 million seed investment round led by Lightspeed India Partners. The project also received funding from Peak XV Partners and prominent angel investors Paul Graham, Gokul Rajaram, and Tariq Shipar. The company was founded by Advayt Sridhar and Akash Ramdas, who combined their experience: Ramdas holds a PhD in materials science from Stanford University, while Sridhar worked on agents at Persona AI and Luma Labs.

Searching for materials with AI

The founders developed a pipeline for generating ideas for new materials using Anthropic models in specialized software. They then run fundamental physics models they trained themselves to determine whether the candidate materials are genuinely promising. According to Sridhar, Ramdas tested around 20 hypotheses per day during his doctoral research, whereas today these agents are testing thousands of ideas because they operate around the clock in the cloud.

Discovered Materials has now presented hundreds of new material samples, as well as the Material Discovery Bench platform, designed to track how advanced models perform this task. Although companies such as MatNex, SandboxAQ, and CuspAI have launched similar initiatives, Discovered Materials hopes to succeed by focusing specifically on the thermal challenges of semiconductor materials.

Engineering challenges and prospects

The startup claims to have already discovered several new substances whose properties match those of existing materials used by major chip manufacturers, but it is not disclosing the details. One major engineering challenge is that even if a material is found that reduces heat generation or dissipates heat, manufacturing a chip from it may be extremely difficult, or its electrical properties may deteriorate.

Lightspeed partner Hemant Moxapathra, who led the investment round, said that this resembles a kind of puzzle involving atomic structures. A material is useful in the real world only when all its properties align simultaneously. Moxapathra believes that as models improve, predicting new materials will become a standard service, but Discovered Materials' advantage lies in its founders' deep expertise and ability to conduct experiments quickly.