Discovered Materials has raised a $9 million seed round to build AI agents that search for new semiconductor materials, with an initial focus on reducing heat constraints in chips used for AI workloads.

The company says the round was led by Lightspeed, with backing from Y Combinator, Peak XV, Paul Graham, Gokul Rajaram, and Thariq Shihipar. Its founders, Advaith Sridhar and Akash Ramdas, describe the product as an attempt to shorten the path from lab work to materials that can be used in semiconductor fabrication.

The technical pitch is narrower than a general AI-for-science platform. Discovered Materials says it is targeting thermal interface and related chip materials, where heat generation and heat dissipation affect power use, cooling systems, and ultimately data center costs. TechCrunch reports that the startup uses Anthropic models in a custom agent harness to generate material leads, then applies physics models and simulations to filter candidates.

The company is also releasing Material Discovery Bench, an open-source benchmark for AI-driven materials discovery. Discovered Materials says it built the benchmark with experts from IBM, imec, Stanford, and Cambridge to evaluate model performance on a real-world materials problem rather than a generic reasoning task.

The caveat is that candidate discovery is only one part of the semiconductor materials pipeline. A material that looks useful in simulation still has to be synthesized, tested, manufactured reliably, and integrated into chipmaking processes. That makes this less a near-term product launch than a bet that agent-driven research can make materials screening faster and less expensive.