Can AI help solve the overheating computer chip problem?

A startup called Discovered Materials has raised nine million dollars to hunt for these new materials. Instead of human scientists working in a lab, the company uses swarms of AI agents—independent pieces of software designed to perform specific tasks—to propose and test thousands of new chemical combinations every day. They are specifically focusing on the thermal problems that plague modern chips, hoping to find materials that can dissipate heat better than the ones currently used by major manufacturers.

Finding new materials traditionally involves a slow, manual process of trial and error. To speed this up, the startup uses two layers of software. First, they use commercially available AI models to brainstorm thousands of possible material structures. Then, they use custom-built, scientifically grounded computer simulations to test these ideas. This process acts like a digital filter, discarding thousands of poor options instantly so that only the most promising candidates are flagged for real-world testing. It is a game of probability: the AI performs a massive amount of guesswork to narrow down the periodic table to a handful of substances that might actually work as parts for a chip.

We are currently in an era where software can design complex structures, but the physical world remains a significant bottleneck. Even if an AI identifies a perfect, heat-resistant material, manufacturing it is a different, much harder hurdle. It might be too brittle to handle, too expensive to produce, or physically impossible to integrate into a standard chip. These startups are betting that by automating the guessing process, they can find the one in a million material that changes the efficiency of hardware. However, the real work will still happen in a traditional wet lab—a physical space where chemicals and materials are tested by hand. The race is on to see if these digital discoveries can lead to real hardware in our devices, or if the transition from simulation to a physical chip remains a challenge too big for AI to solve alone.