AI startup Discovered Materials hunts cooler chip designs with $9M seed round

Discovered Materials, founded by Advaith Sridhar and Akash Ramdas, is tackling one of the most pressing hardware challenges in AI: the heat generated by chips running AI workloads. Data centers consume enormous amounts of electricity partly because of the cooling systems required to manage this thermal load. Rather than solving the problem through traditional engineering, the startup is using AI agents to discover new materials that could enable more efficient integrated circuits.

The company's approach centers on a custom software pipeline. It uses Anthropic models in a harness to generate potential material leads, then employs foundational physics models that Sridhar and Ramdas trained in-house to run simulations verifying whether candidate materials have practical merit. Sridhar highlighted the speed gain: during his doctorate, Ramdas was able to make roughly 20 educated guesses per day. By deploying AI agents running continuously on the cloud, the team can now test thousands of candidates daily, exploring research directions the founders provide.

The startup today released examples of hundreds of new materials alongside a benchmarking tool called "Material Discovery Bench," designed to track how frontier models tackle this challenge. Discovered Materials has reportedly already found several materials matching the properties of existing semiconductors used by major chipmakers, though the founders have not shared specifics, likely for competitive reasons. The company closed a $9 million seed round from Lightspeed India Partners after graduating from Y Combinator, with additional backing from Peak XV Partners and angels including Paul Graham, Gokul Rajaram, and Thariq Shihipar.

A central challenge is the engineering trade-off: a material that reduces heat or improves dissipation might be too difficult or expensive to manufacture into a functional chip, or its electrical properties may suffer. Lightspeed partner Hemant Mohapatra, who led the round, described this as "a bit of playing whack-a-mole with atomic structures." For a material to be truly useful in the real world, all of its properties—thermal, electrical, manufacturability—must converge. The founders plan to patent either the use of promising materials in GPUs or the manufacturing processes required to produce chips from them, licensing these designs to chipmakers. Sridhar hopes to have patentable materials ready within a year.

However, the path from discovery to commercial deployment remains uncertain. While AI-discovered molecules like Insilico Medicine's Renterosib have reached Phase II clinical trials, and MatNex and Panasonic have identified promising candidates, none have been deployed at commercial scale. Mohapatra cautioned that as AI models improve, predicting novel materials will likely become commoditized, and the real bottleneck is not finding candidates but "filtering them correctly and synthesizing them." Sridhar acknowledged this friction, noting that "a lot of this will involve actually going into wet labs and like making things as well. And this is the process that cannot be sped up." Discovered Materials' edge may ultimately depend not on its AI agent technology—which rivals are adopting—but on the founders' deep domain expertise and their ability to rapidly validate and manufacture prototypes in-house.