This startup uses AI to hunt for chip materials that run cooler

Key points

• Discovered Materials closed a $9 million seed round led by Lightspeed India Partners in 2025.
• The startup uses AI agents, software that can carry out tasks on its own around the clock, to test thousands of candidate chip materials per day.
• Its founders believe the real bottleneck in AI-assisted materials science is not finding candidates but verifying and physically making them.
• No AI-discovered material has reached large-scale commercial production yet, in chips or in any other industry.

Chips get hot. That heat is why data centres, the vast warehouse-sized buildings full of computers that power everything from streaming video to AI chatbots, burn enormous amounts of electricity and need heavy-duty cooling systems. Discovered Materials, a startup that emerged from the well-known startup school Y Combinator, thinks AI can help fix the problem.

The company announced this week that it has closed a $9 million seed round, with backing from Lightspeed India Partners, Peak XV Partners, and angel investors Paul Graham, Gokul Rajaram, and Thariq Shihipar.

What does the company actually do?

Discovered Materials uses swarms of AI agents running on cloud computers to search for new materials that could make computer chips generate less heat. Co-founder Akash Ramdas earned a doctorate in materials science from Stanford; his partner Advaith Sridhar previously worked on AI agents at two other tech companies.

Ramdas could test around 20 material ideas a day by hand during his PhD research. The startup's system now runs thousands of tests a day, continuously, by having AI agents explore ideas he points them toward.

The system works in two stages. First, it uses models built by AI company Anthropic, the maker of the Claude chatbot, to generate lists of promising materials. Then it runs those candidates through physics simulation models the team trained itself, to check whether the materials are actually worth pursuing.

Why is finding cooler chip materials so hard?

A material that keeps chips cool might be impossible to manufacture, or it might work thermally but fail electrically. As Hemant Mohapatra, the Lightspeed partner who led this funding round, put it to TechCrunch: "It's a bit of playing whack-a-mole with atomic structures. A material is only useful in the real world if all of them converge at once."

Discovered Materials released examples of hundreds of new candidate materials this week, along with a public benchmarking tool called the "Material Discovery Bench" that tracks how well frontier AI models, meaning the most capable AI systems available, perform on this search problem.

Several other companies, including MatNex, SandboxAQ, and CuspAI, are working in the same space. Discovered Materials is betting that focusing tightly on the heat problem in semiconductors, the class of materials chips are built from, gives it an edge.

What does this mean in practice?

For now, patients, consumers and chip buyers should not expect immediate change. No drug or material found with AI has reached commercial scale yet. The nearest example is a cancer drug called Renterosib, developed by Insilico Medicine, which became the first drug discovered using generative AI to reach a Phase II clinical trial, a mid-stage human safety and effectiveness test.

Mohapatra believes the hold-up is not a shortage of candidate materials. "Filtering them correctly and synthesizing them is the bottleneck," he said. Sridhar agrees: "A lot of this will involve actually going into wet labs and making things as well. And this is the process that cannot be sped up."

If the startup does find materials worth protecting, the plan is to patent them and license their use to chipmakers. Sridhar hopes to have patentable discoveries within the next year.

Common questions

Does this affect my electricity bill or the cost of AI services?

Not yet. The startup is at an early research stage, and any new chip material would take years to reach mass production. Cheaper, cooler chips could eventually lower the cost of running AI services, but that outcome is still well in the future.

Is AI actually good at discovering new materials?

AI can rapidly narrow down a vast space of possibilities, but the final validation still requires physical laboratory work that takes time and cannot be automated away. Promising results exist across the industry, but none have reached large-scale commercial use yet.