AI workloads have made chip heat a central problem for data centers. Hotter chips mean more electricity use and more demand for cooling systems, so a startup called Discovered Materials is trying to use AI to find materials that could make integrated circuits more efficient.
Why Cooler Chips Are Now A Materials Problem
The starting point is simple: chips running AI workloads are too hot. The source of the problem is not only software or data center design. It also reaches down into the materials used to build integrated circuits.
Discovered Materials is focused on that layer. Its goal is to identify new semiconductor materials that could reduce heat generation or improve heat dissipation while still being useful inside real chips.
That is a narrow but difficult target. A candidate material cannot be judged by one attractive property alone. If a substance looks promising thermally but is hard to manufacture, or if its electrical properties are compromised, it may not be useful for chipmakers.
How Discovered Materials Uses AI Agents
The company was founded by Advaith Sridhar and Akash Ramdas. Ramdas brings experience from earning a doctorate in materials science from Stanford, while Sridhar previously worked on agents at Persona AI and Luma Labs.
Together, they built a software pipeline that uses Anthropic models inside a custom harness to generate material leads. Those leads are then checked with foundational physics models the company has trained to run simulations.
The goal is not just to produce a long list of possibilities. The system is designed to narrow attention toward candidate materials that may actually deserve deeper investigation.
“[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.”
Discovered Materials also released examples of hundreds of new materials, along with its “Material Discovery Bench,” which is meant to track how frontier models approach this kind of materials search challenge.
The Funding And The Competitive Field
The startup recently closed a $9 million seed round from Lightspeed India Partners after emerging from from Y Combinator. Peak XV Partners also invested, along with angel investors Paul Graham, Gokul Rajaram, and Thariq Shihipar.
Discovered Materials is not the only company trying to apply AI to materials science. MatNex, SandboxAQ, and CuspAI have launched related efforts. Discovered Materials is trying to stand apart by concentrating on thermal problems in semiconductor materials.
The company says it has already found several materials that match the properties of existing materials used by major chipmakers. It has not shared more details about those materials.
Hemant Mohapatra, the Lightspeed partner who led the round, described the difficulty as a search across many competing constraints.
“It’s a bit of playing whack-a-mole with atomic structures,” Hemant Mohapatra told TechCrunch. “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 Discovery Alone Is Not Enough
The hard part is that materials science does not end when software produces a candidate. A useful substance still has to be filtered correctly, synthesized, and tested.
Mohapatra expects the business of predicting novel substances to become commoditized as models improve. In his view, Discovered Materials has an edge because of Ramdas’s field experience and the ability to operate a lab that can quickly experiment with and validate candidates.
That distinction matters because AI-generated lists do not automatically become commercial products. The source notes that AI-discovered drugs and materials have not yet shown commercial impact at scale.
Insilico Medicine’s Renterosib is described as perhaps the closest example on the drug side, as the first drug discovered with generative AI to make it into a Phase II clinical trial. On the materials side, examples include MatNex’s rare-earth free permanent magnets and new semiconductor materials worked out by Panasonic and Citrine Informatics, but those have not been commercially deployed at scale yet.
That is why Mohapatra argues that simply finding more candidates is not the main bottleneck for AI materials science. Instead, he said, “filtering them correctly and synthesizing them is the bottleneck.”
What The Company Wants To Patent
If Discovered Materials finds valuable candidates, Sridhar says the company plans to try to patent either the use of those materials in GPUs or the process for making chips out of the substance. The business model would then be to license those patents to chipmakers.
Sridhar hopes the company will have new materials worth patenting in the next year. But he also acknowledged that software can only accelerate part of the process.
“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 sp ed up.”
That tension defines the opportunity. AI agents may be able to explore research directions far faster than a human researcher working manually. But the final test still depends on whether the material can be made, measured, and used in a real semiconductor manufacturing context.
For data centers and chipmakers, the promise is clear: cooler chips could help address one of the practical limits of AI infrastructure. For Discovered Materials, the challenge is to prove that AI-guided discovery can move beyond promising candidates and into materials that the semiconductor industry can actually use.