Data centers guzzle electricity, and a big chunk of that power goes to cooling chips that run AI workloads. Now, a Y Combinator-backed startup is turning AI loose on the problem itself, deploying swarms of agents to discover new materials that could make semiconductors run cooler and more efficiently. Main Developments Discovered Materials, founded by Advaith Sridhar and Akash Ramdas, has raised a $9 million seed round led by Lightspeed India Partners, with participation from Peak XV Partners and angel investors Paul Graham, Gokul Rajaram, and Thariq Shihipar. The startup emerged from Y Combinator and is now building a software pipeline that combines AI agents with physics-based simulations to generate and validate new material candidates. Ramdas, who holds a doctorate in materials science from Stanford, provides the domain expertise, while Sridhar brings experience working on AI agents at Persona AI and Luma Labs. Their system uses Anthropic models in a custom harness to generate leads, then runs simulations using their own trained foundational physics models to verify which candidates are worth pursuing. Read also: Why Venmo on Google Play Could Change How You Pay "Ramdas was doing maybe 20 guesses a day during his PhD," Sridhar told TechCrunch. "We're able to do thousands of guesses a day now by having these agents run 24/7 on the cloud, exploring research directions that he gives them." The company has released examples of hundreds of new materials and launched its Material Discovery Bench, a benchmark designed to track how frontier models handle this challenge. Background Discovered Materials is not alone in applying AI to materials science. Companies like MatNex, SandboxAQ, and CuspAI have launched similar efforts, but the startup is betting that a narrow focus on semiconductor thermal problems will give it an edge. The founders claim they have already found several materials that match the properties of those used by major chipmakers, though they declined to share specifics. The engineering trade-space is a major hurdle. A material that reduces heat generation might be too difficult to manufacture into a chip, or its electrical properties could be compromised. "It's a bit of playing whack-a-mole with atomic structures," said Hemant Mohapatra, the Lightspeed partner who led the round. "A material is only useful in the real world if all of them converge at once, which is what makes this a really interesting search problem." Why It Matters The stakes are high. Data centers consume enormous amounts of electricity, and cooling is a major contributor. If AI-discovered materials can make chips run cooler, they could slash energy use and reduce the environmental footprint of the AI boom. But the industry has yet to see a commercially deployed material or drug discovered by AI. The closest example is Insilico Medicine's Renterosib, the first AI-discovered drug to reach Phase II clinical trials, while materials like MatNex's rare-earth free magnets and new semiconductors from Panasonic and Citrine Informatics remain unproven at scale. Mohapatra believes the bottleneck isn't finding candidates—it's filtering and synthesizing them. He expects the business of predicting novel substances will become commoditized as models improve, but Ramdas's deep field experience and the ability to run a lab for rapid validation give Discovered Materials an edge. Still, Sridhar acknowledges that much of the work will happen in wet labs, a process that can't be rushed. What's Next Discovered Materials plans to patent valuable candidates for use in GPUs or the processes to make chips from them, then license those patents to chipmakers. Sridhar hopes to have materials worth patenting within the next year. The company is also releasing its Material Discovery Bench to the public, inviting other researchers to test frontier models against standardized challenges. Open questions remain about whether any AI-discovered material will achieve commercial impact, and how quickly the startup can move from simulation to physical validation. The company's success will depend on its ability to navigate the trade-space where thermal, manufacturing, and electrical properties all converge.