This startup is using AI to find cooler computer chips

AI is now being used to fix a problem it helped create. Computer chips that power AI systems get extremely hot, forcing data centers to use massive amounts of electricity for cooling. Discovered Materials, a new startup, is using swarms of AI agents to hunt for better materials to build more efficient chips.

The company just raised 9 million dollars to scale up its approach. Their system uses AI to generate thousands of potential new materials each day, then runs physics simulations to check if they might actually work. During his PhD, one founder could only test about 20 ideas a day. Now their AI can test thousands nonstop.

They’ve already found materials that match what major chipmakers use today, though they’re keeping the details secret for now. The challenge is that a material might solve the heat problem but be impossible to manufacture, or it might have other flaws. As one investor put it, it’s like playing whack-a-mole with atomic structures. A material has to be good at everything at once to be useful.

Discovered Materials was started by Advaith Sridhar and Akash Ramdas, who combined expertise in AI agents and materials science. Other companies like MatNex and SandboxAQ are working on similar ideas, but Discovered Materials is focusing specifically on the heat issues in chips. They plan to patent and license any useful materials they find.

The bigger picture here is that AI has been promised for years as a way to discover new drugs and materials, but we haven’t seen many real-world successes yet. There are some promising examples, like a drug in clinical trials and new magnets, but nothing at large scale. The bottleneck isn’t finding ideas it’s testing them and turning them into real products.

If this works, it could mean chips that use less energy and generate less heat, which would make everything from smartphones to data centers more efficient. But the process of actually making and testing these materials in labs can’t be sped up, so progress might still be slow.

What happens next is anyone’s guess. The company hopes to have materials worth patenting within a year. But the real test will be whether these materials can move from the lab to actual chips in our devices.

Could AI really design materials that outperform what humans have created so far

Is the hype around AI-driven discovery justified, or are we still years away from real impact

---

Filed under: AI, Chips, MaterialsScience, Startup, TechInnovation