Discovered Materials Raises $9M Seed to Let AI Agents Hunt for Cooler, More Efficient Chips
Discovered Materials, a startup applying AI agents to materials discovery for integrated circuits, 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 TechCrunch. The company, founded by Advaith Sridhar and Akash Ramdas, emerged from Y Combinator and is simultaneously open-sourcing Material Discovery Bench, a benchmark for evaluating how well AI models can discover new materials.
Lightspeed India Partners partner Hemant Mohapatra led the round. The founders met over a decade ago at IIT-Madras. Ramdas earned a doctorate in materials science from Stanford, where his work on new nanoscale interconnects — the tiny wires that connect components inside a chip — became Stanford Engineering's most popular story of 2025. Sridhar previously worked on AI agents at Persona AI and Luma Labs TechCrunch.
The technical premise is easy to state, if hard to pull off: deploy AI agents to search the vast space of possible chemical compounds for ones that can improve the thermal and power efficiency of integrated circuits. The company frames the urgency in concrete terms. Current GPUs handle heat fluxes of approximately 140 watts per square centimeter, a figure that exceeds what a space shuttle nose cone endures during atmospheric re-entry. Discovered Materials also notes that today's chips are at least 10,000 times less power efficient than the human brain.
During its three-month Y Combinator batch, the company simulated, synthesized, and tested thermal interface materials that matched the performance of trade-secret products held by the world's largest chemical companies for over two decades. Thermal interface materials sit between a chip and its heatsink, filling microscopic gaps to transfer heat away from the silicon. Going from simulation to physical synthesis to benchmarked testing within a single accelerator cohort is the operational claim that caught investor attention.
Material Discovery Bench, built in collaboration with experts from IBM, IMEC, Stanford, and Cambridge, tests how well frontier AI models perform on real-world materials problems. The benchmark includes multiple verifiers for grading model ability, giving the field a standardized way to evaluate AI-driven materials discovery.
The broader context here matters. The semiconductor industry's thermal management problem has intensified with the spread of large-scale AI training and inference workloads. As accelerators push past 1,000 watts per package and hyperscale deployments scale into the hundreds of thousands of units, the thermal interface layer between a chip's die and its heatsink is no longer a commodity component. It is a systems-level bottleneck. Materials discovery has historically been a slow, trial-and-error process dominated by large chemical conglomerates with decades of institutional knowledge. The claim that an AI-agent-driven approach can compress that timeline into months and match guarded trade-secret performance is worth scrutinizing. The company has published its benchmark, which gives the community a mechanism to do exactly that.
What separates credible claims from hype in AI-for-materials will be reproducibility. Open-sourcing Material Discovery Bench is a meaningful step in that direction. If frontier models can be evaluated on standardized, real-world materials problems with multiple independent verifiers, the field gains a shared yardstick. That matters because materials science has fewer natural feedback loops than software or even protein folding. A language model can generate a candidate compound, but validating it requires physical synthesis, characterization, and testing under real thermal and electrical loads. The benchmark's inclusion of multiple verifiers suggests the company understands this gap between prediction and physical proof.
The investor lineup also tells a story about market positioning. Lightspeed India and Peak XV bringing institutional capital, combined with Graham's angel participation, places Discovered Materials at the intersection of deep-tech materials work and the AI-agent investment thesis. Ramdas's Stanford credentials in nanoscale interconnects and Sridhar's background in agentic systems at Persona AI and Luma Labs form a founder profile that maps directly onto the problem space: one founder understands the materials science, the other understands how to build and deploy autonomous AI agents.
The thermal management layer of the semiconductor stack has not historically attracted this kind of founder or this kind of capital. The fact that it is doing so now reflects the pressure that AI compute demand is placing on every layer of the hardware stack, from silicon fabrication through packaging through data center cooling. If AI agents can accelerate materials discovery by even a fraction of what Discovered Materials claims, the downstream effects extend beyond any single company. New thermal interface materials that improve efficiency by even modest margins, deployed at hyperscale, translate into meaningful reductions in data center power consumption. The open benchmark will determine how quickly the field can verify and build on these claims.
