SAN FRANCISCO — Discovered Materials closed a $9 million seed round to develop artificial intelligence agents that identify more efficient integrated circuit materials. Lightspeed India Partners led the investment, with participation from Peak XV Partners and angel investors Paul Graham, Gokul Rajaram and Thariq Shihipar.
The Y Combinator alumnus targets the thermal problem eating into data center economics: modern AI chips generate enough heat that electricity and cooling costs have become a primary operating expense for hyperscalers.
Co-founders Advaith Sridhar and Akash Ramdas lead the company. Sridhar previously built AI agents at Persona AI and Luma Labs; Ramdas holds a doctorate in materials science from Stanford.
The startup runs a software pipeline that feeds Anthropic models into a custom harness. Those models generate candidate materials, which are passed to proprietary physics models that run simulations to test viability.
The throughput advantage is the core pitch. During his Ph.D. research, Ramdas could evaluate roughly 20 material candidates per day. The company's agents run continuously on cloud infrastructure, processing thousands of candidates daily.
Discovered Materials on Tuesday released a set of hundreds of new materials alongside its Material Discovery Bench, a platform designed to track how frontier AI models perform on material discovery tasks.
The competitive field is growing. MatNex, SandboxAQ and CuspAI are pursuing similar objectives. Discovered Materials distinguishes itself with a specific focus on the thermal properties of semiconductor materials.
CuspAI, a direct competitor, recently closed a $450 million Series B backed by Jeff Bezos's fund, Glade Brook Capital Partners, Lux Capital, AMD Ventures and Britain's Sovereign AI Venture Fund. CuspAI also holds partnerships with Nvidia, Meta Platforms and Hyundai Motor Group.
"Major technological advances often depend on new materials," said Josh Coyne, a partner at Kleiner Perkins. "The next generation of innovations, from carbon capture to semiconductors, awaits the discovery of materials not yet known."
Discovered Materials says it has already identified several materials matching properties of those currently used by major chipmakers, though it has not disclosed specifics. The central engineering challenge is convergence: a material that improves heat dissipation may be difficult to manufacture into a chip or may compromise electrical performance.
"Finding a useful material requires multiple properties to converge simultaneously," said Hemant Mohapatra, the Lightspeed partner who led the seed round. He also expects the general business of predicting novel substances to become commoditized as AI models improve.
Mohapatra said Ramdas's depth in materials science and the company's ability to run a lab for rapid validation are the key differentiators. Discovered Materials has already confirmed several new materials through laboratory testing.
The company plans to patent newly discovered materials for use in GPUs, or the processes for manufacturing chips from them, then license that intellectual property to chipmakers.
