Discovered Materials is playing AI whack-a-mole to hunt cooler chips
Discovered Materials, a Y Combinator-backed startup, is using swarms of AI agents to search for new materials that could help chips run cooler, addressing a key driver of data centres' huge electricity and cooling costs. The company has closed a $9 million seed round led by Lightspeed India Partners, with backing from Peak XV Partners and angels including Paul Graham, reflecting growing investor interest in AI-driven materials discovery as a way to tackle the thermal limits of semiconductor design.
Founded by Advaith Sridhar and Akash Ramdas, the startup pairs Anthropic-based AI agents that generate material candidates with custom-trained physics models that simulate and verify their properties, running "thousands of guesses a day" compared with the roughly 20 Ramdas managed manually during his PhD. It has released examples of hundreds of new materials and a "Material Discovery Bench" to benchmark AI models on the task, and says it has already found candidates matching properties of materials used by major chipmakers, though details remain undisclosed. The firm faces competition from similar ventures such as MatNex, SandboxAQ and CuspAI, and acknowledges that manufacturability and electrical trade-offs make real-world validation, described as "whack-a-mole with atomic structures", difficult; it hopes to patent commercially viable materials within a year, though AI-discovered materials have yet to achieve large-scale commercial deployment anywhere in the industry.
- Discovered Materials raised $9m to find cooler chip materials using AI agents
- Combines AI-generated leads with physics simulations to verify candidates
- No AI-discovered chip material has yet reached commercial-scale deployment
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Advocates for AI-driven materials discovery argue that pairing generative AI agents with physics-based verification represents a genuine leap beyond traditional trial-and-error research, allowing thousands of candidate materials to be screened daily rather than the mere handful a PhD researcher could manage by hand. They see this as a rational response to a real and pressing problem, since cooling accounts for a substantial share of data centres' soaring energy costs, and even incremental improvements in thermal materials could yield significant efficiency gains at scale. Investors backing the approach, including experienced technology financiers and angels, view the speed and breadth of AI-assisted discovery as a legitimate way to compress years of laboratory work into a fraction of the time, justifying early-stage bets despite the absence of proven commercial deployment.
The case against
Sceptics of the approach would argue that generating large numbers of candidate materials is not the same as delivering usable ones, and that the company's own acknowledgement of manufacturability and electrical trade-offs, likened to whack-a-mole, underscores how far the technology remains from real-world validation. They would note that no AI-discovered material has yet achieved large-scale commercial deployment anywhere in the industry, and that specific claims of matching major chipmakers' material properties remain undisclosed and therefore unverifiable. Given a crowded field of well-funded competitors making similar claims, a cautious observer might reasonably ask whether the promise of AI-accelerated materials science is currently outpacing demonstrated results, and whether investor enthusiasm reflects proven capability or the broader appeal of AI-branded ventures.