San Francisco Materials Startup Raises $9 Million to Speed Chip Cooling Design
Discovered Materials, a San Francisco startup using artificial intelligence to develop thermal materials for semiconductors, has announced $9 million in seed funding led by Lightspeed Venture Partners, with backing from Y Combinator, Peak XV Partners, and angel investors including Y Combinator founder Paul Graham, Gokul Rajaram, and Thariq Shihipar.
The company is tackling a problem that has become increasingly urgent as AI chips generate more heat. Modern GPU systems in data centers already manage heat fluxes around 140 watts per square centimeter, with experimental immersion cooling configurations reaching 155.6 watts per square centimeter, according to research published last July in the International Journal of Heat and Mass Transfer. The materials that sit between those chips and their cooling systems—thermal interface materials, in industry parlance—have become a bottleneck.
Discovered Materials uses AI agents to simulate, synthesize, and test new materials, a process the company says can compress development timelines from months to days. During a three-month stint in Y Combinator's spring batch, the startup claims it created thermal interface materials matching the performance of products that major chemical companies have protected as trade secrets for over two decades. It's a bold claim for such a young operation, though the company declined to name paying customers or provide third-party validation.
The startup's target customers are packaging and thermal engineers at chip heavyweights including NVIDIA, AMD, Groq, Cerebras, Google, Intel, Samsung, and TSMC, according to its Y Combinator profile. Whether any of those companies are actually buying remains unclear.
Two Founders, Deep Technical Roots
The founding team consists of Advaith Sridhar, who holds a master's degree in AI from Carnegie Mellon and previously worked as a research engineer at Luma Labs and founding applied scientist at Persona AI (later acquired by Luma Labs), and Akash Ramdas, who completed his PhD and postdoctoral work in materials science and engineering at Stanford. Ramdas has published research on ultrathin conductors and alternative interconnects, including a January paper in Science on surface conduction in noncrystalline semimetals.
Y Combinator lists the team size as two people. The company, which operates legally as Matforge Inc., recently rebranded from Matforge to Discovered Materials, according to a LinkedIn post from Sridhar.
Testing the Limits of AI Models
Alongside the funding announcement, Discovered Materials launched the Material Discovery Bench, an open benchmark measuring how frontier language models discover thermally conductive dielectrics for 3D chip stacks. The leaderboard evaluates various AI models against the task, though the specific model versions listed appear to be placeholder designations that may evolve as the benchmark matures.
The results reveal both promise and limitation. Models can identify computationally novel, dynamically stable candidates, but synthesis remains a stubborn hurdle. "Of the 500+ materials discovered, only 1 (one) material has a plausible synthesis pathway," the company noted. Token budgets per run range from 30 million to 100 million tokens, reflecting the computational intensity of the work.
The benchmark lists acknowledgments from researchers at IBM, IMEC, Micron, Stanford, and Cambridge, suggesting some level of advisory or review collaboration. Among them: IBM's Dr. Daniel Edelstein, IMEC's Dr. Zsolt Tokei and Dr. Geoffrey Pourtois, Micron's Dr. Gurtej Sandhu, Stanford's Prof. Krishna Saraswat, and Cambridge's Prof. Judith MacManus-Driscoll.
A Crowded Field
Discovered Materials enters a materials discovery space that has drawn substantial venture attention, particularly in recent months. CuspAI reportedly raised roughly $450 million at a $2 billion valuation in July, according to eWeek and The Guardian. Orbital Industries closed a $50 million Series B in May, Fortune reported. Earlier rounds include alqem's €8 million pre-seed in July and Arcturus' $8 million seed in June, among others.
Government laboratories are also moving into the territory. Lawrence Berkeley National Laboratory announced its FORUM-AI project for agentic AI in materials research in February. In July, the National Science Foundation awarded nearly $20 million each to the University of Tennessee Knoxville and Rice University for automated materials discovery nodes.
The crowding suggests either genuine opportunity or hype-driven capital chasing a compelling narrative. Probably some of both.
Hiring Plans
The company is now hiring founding engineers in Mountain View, California. Job postings on its Y Combinator page list salary ranges from $150,000 to $250,000 with equity grants of 1 to 2 percent. Open roles include process engineer, computational materials engineer, and equipment engineer positions—the kind of hires that signal a shift from algorithm development to physical production.
For a two-person team claiming to match the output of chemical giants, scaling the human side of the operation may prove as challenging as the materials science itself.
