Discovered Materials uses AI agents to cool down the chip industry

The startup, founded by Advaith Sridhar and Akash Ramdas, utilizes a software pipeline that leverages Anthropic models to generate candidates before subjecting them to custom-trained physics simulations. While Ramdas once manually performed roughly 20 guesses per day during his doctoral research at Stanford, the new system operates 24/7 on the cloud to run thousands of simulations daily. The company intends to patent these discoveries and license them to major chipmakers, aiming for viable, patentable substances within the next year.

Investors like Hemant Mohapatra of Lightspeed India Partners are backing the venture, citing the difficulty of the search as a classic optimization problem. However, the path to commercialization remains unproven. While companies such as MatNex and SandboxAQ explore similar frontiers, the industry has yet to see AI-discovered materials reach widespread manufacturing deployment. Sridhar acknowledges that the ultimate bottleneck is not just the digital discovery process, but the physical synthesis of these materials in wet labs, a stage that remains resistant to acceleration.