AI Chips Overheating: The Problem and the Technological Solution
According to a report in TechCrunch, chips running artificial intelligence (AI) workloads suffer from significant overheating during their operation. This overheating represents one of the primary reasons why data centers worldwide consume such massive amounts of electricity and require highly complex and expensive cooling systems. Inevitably, many entrepreneurs in the industry are now turning to AI technologies to solve the overheating problem that AI itself created in the first place. The startup company Discovered Materials is the latest player to enter this arena, with definite plans to utilize swarms of AI agents to locate and discover new materials that can be used to construct more efficient and cooler integrated circuits.
A $9 Million Seed Funding Round and Backing from Leading Investors
Discovered Materials recently announced the closing of a successful seed funding round totaling $9 million. The investment round was led by the venture capital firm Lightspeed India Partners, and it took place after the startup graduated from the renowned accelerator Y Combinator. Peak XV Partners also participated in the funding round, alongside a series of prominent and well-known angel investors in the industry, including Paul Graham, Gokul Rajaram, and Thariq Shihipar.
Company Founders: Merging Materials Science with AI Agents
Behind the establishment of the company are founders Advaith Sridhar and Akash Ramdas, who joined forces to launch this joint venture. The two combine a rich academic and technological background: Ramdas brings substantial practical and research experience accumulated during his doctoral studies in materials science at Stanford University, while Sridhar contributes his professional expertise in developing intelligent agents acquired during his work at Persona AI and Luma Labs.
The co-founders have developed a unique software pipeline that utilizes Anthropic's models inside a custom-built harness to generate material leads. Following this step, the system turns to foundational physics models that the founders trained themselves, in order to run complex simulations designed to verify whether the proposed candidate materials are indeed of scientific and practical interest.
In an interview with TechCrunch, Sridhar described the dramatic difference in search capabilities that this technology brings: "[Ramdas] was doing maybe 20 guesses a day during his PhD. Today, we are able to do thousands of guesses a day by running these AI agents, which run 24 hours a day, 7 days a week on the cloud, exploring the scientific research directions he defines for them."
Revealing Hundreds of Materials and the Material Discovery Bench
Alongside the funding announcement, Discovered Materials published examples of hundreds of new materials discovered through its proprietary system, and also unveiled its benchmark tool, known as the "Material Discovery Bench". This tool is designed to track and continuously examine how frontier models handle the complex scientific challenge of material discovery.
While other companies in the industry, such as MatNex, SandboxAQ, and CuspAI, have already launched similar initiatives and efforts in this space, Discovered Materials is betting that a laser-focus on the thermal and heating problems of semiconductor materials is the correct path that will lead them to commercial success. The startup reports that it has already managed to discover several new materials that match the properties of existing materials currently used by major chipmakers in the industry; however, at this stage, the company cannot disclose or share further details about them.
The Engineering Trade-Space and the Atomic Structure "Whack-a-Mole"
One of the greatest challenges facing the company is the engineering trade-space. The founders explain that even if they manage to find a new material that might significantly reduce heat generation on a chip or improve heat dissipation, the material might be too difficult or complex to actually physically manufacture a chip in fabrication plants, or its electrical properties might be compromised and degraded as a result of structural changes.
Hemant Mohapatra, the partner at Lightspeed who led this current seed funding round, explained the scientific difficulty to TechCrunch: "It’s a bit like playing 'whack-a-mole' with atomic structures. A material is only useful in the real world if all of its properties and characteristics converge at the same time, which is exactly what makes this search problem a particularly interesting challenge."
Mohapatra expects that the industry of predicting novel substances will become commoditized as technological models continue to improve and evolve. In his view, Discovered Materials' competitive advantage lies in Ramdas’ deep professional experience in the field, alongside the capability to operate a physical lab that can perform rapid experiments and practical validation of candidate materials—an action the two have already actually performed with several new materials.
A Business Model Built on Patents and Licensing to Manufacturers
Regarding the company's business model, Sridhar notes that when they identify high-value candidate materials, the company will work to patent the use of these materials inside graphics processing units (GPUs), or alternatively, patent the manufacturing process that allows chips to be produced from that material, and then license those rights to various chipmakers in the market. He expresses hope that within the coming year, the startup will have new materials worthy of commercial patenting.
The Industry's Commercial Challenge and the Physical Lab Bottleneck
Despite the excitement and significant interest surrounding the technology, it is important to note that so far, no real commercial impact of drugs or materials discovered using AI has been registered in the open market. The closest example of success is likely the drug Renterosib by Insilico Medicine, which is the first drug discovered with generative AI to make it to Phase II clinical trials.
In the materials science sector, several promising candidate materials have been discovered in the past, such as rare-earth-free permanent magnets developed by MatNex, or new semiconductor materials worked out in a collaboration between Panasonic and Citrine Informatics. However, these materials have not yet been commercially deployed or implemented at scale in the industry.
While these search methods may be coming into their own now as AI technologies continue to improve, Mohapatra emphasizes that in his view, finding more candidates is not the main hold-up in AI-based materials science. Instead, he argues that "filtering them correctly and synthesizing them is the bottleneck." Sridhar acknowledges that although Discovered Materials' unique data and expertise will help the startup compete against deep-pocketed frontier labs, the reality is that a significant part of the work will require physically entering wet labs and actually making things as well—"and this is the process that cannot be sped up."