The startup Discovered Materials has raised $9 million in a seed round led by Lightspeed India Partners, with the goal of using swarms of AI agents to search for new materials that could help build more efficient integrated circuits that generate less heat.
The move comes at a time when chips running AI workloads have become an increasing source of heat, raising data centers’ electricity consumption and increasing the need for cooling systems. The company emerged from the Y Combinator program, and the round also attracted investments from Peak XV Partners and angel investors Paul Graham, Gokul Rajaram, and Thariq Shihipar.
A Research and Simulation Pipeline
The company was founded by Advaith Sridhar and Akash Ramdas, drawing on Ramdas’s materials science experience after earning his Ph.D. from Stanford University, and Sridhar’s experience with agents during his work at Persona AI and Luma Labs.
The founders developed a software pipeline that uses Anthropic models within a custom runtime environment to generate proposals for new materials, then uses foundational physics models trained by the company to run simulations verifying the feasibility and properties of those materials.
Sridhar said Ramdas tested around 20 hypotheses per day during his doctoral studies, while the agents can now test thousands of hypotheses per day by working in the cloud around the clock and exploring research paths identified by the team.
On August 10, 2026, Discovered Materials announced examples of hundreds of new materials, along with the Material Discovery Bench, a tool designed to track how advanced models handle the challenge of materials discovery.
The Problem Is Not Just Finding Candidates
The company is competing with similar efforts launched by MatNex, SandboxAQ, and CuspAI, but it is focusing specifically on thermal problems associated with semiconductor materials. It says it has already discovered several materials that match the properties of existing materials used by major chipmaking companies, but it has not disclosed additional details about them.
However, finding a promising material does not mean it will be easy to use in an actual chip. The material may be difficult to manufacture, or its thermal improvements may weaken its electrical properties. Hemant Mohapatra, the Lightspeed partner who led the round, described the process as a continual balancing act involving the properties of atomic structures, since a material only becomes practically useful when all requirements align at the same time.
Mohapatra believes that predicting new materials could become a commoditized service as models improve, but he believes Ramdas’s expertise and the company’s ability to operate a laboratory for conducting experiments and rapidly validating candidates give it an advantage. The company said the founders have already tested several new materials.
A Patent and Licensing Plan
When it finds valuable candidates, Discovered Materials plans to try to obtain patents for using the materials in graphics processing units, or for manufacturing processes that use them to make chips, and then license those patents to chip manufacturers. Sridhar hopes the company will have new materials worthy of patent applications within the next year.
However, the track record of AI-assisted drug and materials discovery has not yet produced broad commercial impact. Insilico Medicine’s drug known as Renterosib is one of the closest examples, after entering a Phase 3 clinical trial in July, while other promising examples in materials, such as the rare-earth-free magnets developed by MatNex and semiconductor materials worked on by Panasonic and Citrine Informatics, have not yet been used commercially at scale.
According to Mohapatra, the bottleneck is not necessarily increasing the number of candidates, but properly screening and manufacturing them. Sridhar acknowledges that the process will require moving to wet laboratories and physically manufacturing the materials, a stage that cannot be accelerated through software alone.