Frontier AI Unlocks Material Innovation for Next-Gen Chips

The semiconductor industry faces a growing challenge: managing the escalating heat output from increasingly powerful chips. As Nvidia’s H100 consumes 700W and Blackwell approaches 1.2kW, with future Rubin chips projected at 2.3kW, efficient heat dissipation becomes critical for datacenter power consumption.

Discovered Materials (YC P26) is leveraging AI agents to address this challenge by accelerating the discovery of novel materials that can enhance chip performance and thermal management. Our approach tackles a key bottleneck in material innovation - the time and cost required to move from lab concept to commercial application, often referred to as the “lab-to-fab valley of death.”

We’ve been testing leading AI models (Anthropic, OpenAI, Kimi) and have observed remarkable capabilities. These agents can computationally identify dynamically stable materials with promising properties at a fraction of the time compared to traditional methods - in some cases, discovering materials that would typically take PhD students weeks to find in just 8 hours.

While computational discovery is only the first step, we’re already seeing tangible results: our team has successfully synthesized and tested thermal interface materials (TIMs) matching performance characteristics of compounds developed by industry leaders over decades - demonstrating the potential for AI-driven breakthroughs.

We’re releasing a comprehensive benchmark and dataset of discovered materials at https://discoveredmaterials.com/research, including insights into model behavior such as Claude’s tendency to reward “hacky” solutions and occasional anomalies with GPT-5.6.

Business Model & Future Directions

Our primary goal is to license IP for discovered materials, while also exploring a secondary model of providing our discovery platform to semiconductor and chemical companies - enabling them to innovate in-house. We’re particularly focused on applications like 3D chip packaging (which could improve data transfer efficiency by 10-50x) and advanced thermal management solutions.

With expertise spanning material science and AI, our team is uniquely positioned to bridge the gap between computational prediction and real-world implementation - accelerating the development of next-generation semiconductor materials.