Discovered Materials: AI Agents Screen Thousands of Chip-Cooling Candidates Per Day

Y Combinator graduate Discovered Materials uses a swarm of AI agents built on Anthropic's models to discover new semiconductor materials. While a traditional PhD-level scientist would evaluate around 20 candidates per day, the system generates thousands and immediately runs them through proprietary physics simulations.

1The Target: Cooling AI Processors

The primary focus is thermally conductive materials for chips under AI workloads. Overheating limits performance: the hotter a chip runs, the lower its clock speed and the more expensive the data center cooling. The team has already characterized hundreds of new materials and launched Material Discovery Bench — a public leaderboard showing how well leading LLMs perform on materials-science tasks.

2Funding and Team

The startup raised $9M from Lightspeed India Partners, with participation from Peak XV Partners and angel investors Paul Graham and Gokul Rajaram. The founders are Akash Ramdas (PhD in materials science, Stanford) and Advait Sridhar (previously built AI agents at Persona AI and Luma Labs). Applying agentic systems to the discovery of physical materials means AI is beginning to accelerate not just the software layer of infrastructure, but the hardware layer as well.