Discovered Materials is playing AI whack-a-mole to hunt cooler chips

The massive energy demands of modern AI workloads have created a thermal crisis in data centers. As chips run hotter than ever, the cooling infrastructure required to keep them operational has become a primary driver of electricity consumption. Now, a new startup is turning the power of artificial intelligence back on the very problem it helped create.

Discovered Materials, a startup emerging from the Y Combinator ecosystem, is leveraging swarms of AI agents to accelerate the search for novel materials capable of building more efficient integrated circuits. The company recently announced a $9 million seed funding round led by Lightspeed India Partners, with additional backing from Peak XV Partners and notable angel investors including Paul Graham, Gokul Rajaram, and Thariq Shihipar.

The Human-AI Hybrid Approach

The company was founded by Advaith Sridhar and Akash Ramdas, a duo that combines expertise in software engineering and materials science. Ramdas brings a doctorate in the field from Stanford, while Sridhar contributes a background in agentic AI from his time at Persona AI and Luma Labs.

Their core innovation is a proprietary software pipeline that utilizes Anthropic’s large language models within a custom harness. This system generates potential material candidates, which are then vetted by foundational physics models trained by the team to verify if the substances hold real-world promise.

“[Ramdas] was doing maybe 20 guesses a day during his PhD,” Sridhar explained. “We’re able to do thousands of guesses a day now by having these agents run 24/7 on the cloud, exploring research directions that he gives them.”

Navigating the Atomic Trade-Space

The challenge of discovering new materials is often compared to a game of "whack-a-mole" at the atomic level. Finding a substance that effectively dissipates heat is only half the battle; that material must also be manufacturable and maintain the necessary electrical properties for high-performance computing.

Hemant Mohapatra, the partner at Lightspeed who spearheaded the investment, emphasized the complexity of this search:

“A material is only useful in the real world if all of them converge at once, which is what makes this a really interesting search problem.”

To track the industry's progress in this domain, Discovered Materials has launched the "Material Discovery Bench," a tool designed to monitor how various frontier models handle the complexities of materials science. While competitors like MatNex, SandboxAQ, and CuspAI are also exploring this space, Discovered Materials is betting that a laser-focused strategy on semiconductor thermal management will provide a competitive edge.

The Path to Commercialization

The startup claims to have already identified several materials that mirror the properties of those currently used by major chip manufacturers, though they are keeping the specific details under wraps for now. Their business model centers on patenting these discoveries—either the materials themselves or the novel processes used to integrate them into GPUs—and subsequently licensing the intellectual property to chipmakers.

The founders hope to have patentable materials ready within the next year. However, the industry remains in a "wait and see" phase regarding AI-driven discovery. While companies like Insilico Medicine have pushed AI-discovered drugs into clinical trials, and others have identified promising magnets or semiconductor components, true large-scale commercial deployment remains elusive.

Key Challenges Ahead

• The Synthesis Bottleneck: As Mohapatra noted, the industry is currently better at generating candidates than it is at filtering and synthesizing them.
• The Wet Lab Reality: Despite the speed of AI, the physical process of creating and testing materials in a laboratory cannot be fully digitized.
• Manufacturing Feasibility: A material that performs well in a simulation may prove impossible to scale in a semiconductor fabrication plant.

Sridhar remains pragmatic about these hurdles. While he believes the company’s unique data and deep technical expertise will allow them to compete with larger, well-funded labs, he acknowledges that the final stages of the process are inherently physical.

"A lot of this will involve actually going into wet labs and making things as well," Sridhar said. "And this is the process that cannot be sped up."

As the AI industry continues to grapple with the physical limitations of its own hardware, Discovered Materials is positioning itself as a critical bridge between generative software and the hard, atomic reality of next-generation chip manufacturing.