Discovered Materials deploys AI agent swarms to hunt for cooler chip materials
Startup Discovered Materials has raised $9 million to use AI agents in search of new materials for more efficient computer chips. The company hopes its discoveries can help ease data center overheating and power consumption.
AI chips generate so much heat that cooling has become one of the main reasons data centers consume vast amounts of electricity. Startup Discovered Materials wants to tackle that problem using the very technology that created it, deploying swarms of AI agents to search for new materials for building more efficient integrated circuits.
The company, which came out of Y Combinator, recently closed a $9 million seed round led by Lightspeed India Partners, with participation from Peak XV Partners and several angel investors. It was founded by Advaith Sridhar and Akash Ramdas, combining Ramdas's materials science doctorate from Stanford with Sridhar's background working on AI agents at other companies.
Thousands of guesses a day
The founders built a software pipeline that uses Anthropic's language models to generate candidate materials, then verifies them using custom-trained physics models that run simulations. While Ramdas could only test around 20 hypotheses a day during his PhD, the company's AI agents now run continuously in the cloud, generating thousands of candidate materials daily.
Discovered Materials released examples of hundreds of newly identified materials, along with a new benchmark tool designed to track how well different AI models perform at this kind of discovery task. Several other companies are pursuing similar goals, but Discovered Materials says its focus is specifically on solving the thermal challenges of semiconductor materials.
Hurdles before real-world use
The bigger challenge lies in the trade-offs: a material might reduce heat or improve dissipation, but still be too difficult to manufacture or have compromised electrical properties. According to the investor who led the funding round, a material only becomes useful once all these properties align simultaneously.
The company plans to patent promising materials and their use in chip manufacturing, then license the technology to chipmakers, with hopes of having patentable results within the next year. Still, no AI-discovered material or drug has yet achieved major commercial deployment, and the founders acknowledge that the lab experimentation stage of the process cannot be sped up.
