The global race to produce high-performance semiconductors requires vast amounts of energy and rare, difficult-to-source minerals. In response to these pressing constraints, the British startup CuspAI has launched a major initiative called the AI Materials Foundry. Backed by nearly half a billion dollars in fresh capital, the two-year-old firm aims to use artificial intelligence to accelerate the discovery of new materials for chipmaking and other industrial applications.

The new coalition brings together more than 48 technology giants, industrial firms, and research facilities, including major players like Nvidia Corp., Meta Platforms Inc., and Hyundai Motor Group. By pooling their computing and scientific resources, members hope to build software that develops new manufacturing materials faster and cheaper than traditional methods. To fund this effort, CuspAI raised 450 million dollars in a Series B round, drawing significant participation from Jeff Bezos’s investment fund, Bezos Expeditions. The round, led by Kleiner Perkins and NEA, propels the startup’s valuation to 2.6 billion dollars.

Chad Edwards, the co-founder and chief executive of CuspAI, explained that much of the capital will support labs located in Cambridge, Singapore, and the San Francisco Bay area. The startup plans to dedicate 80 percent of its research efforts this year to material discovery, with a particular focus on reducing or eliminating the use of supply-chain-risk metals like ruthenium and iridium. CuspAI initially focused on carbon capture and water purification but pivoted due to intense demand from the semiconductor supply chain. Commenting on the urgent need from manufacturers, Edwards noted, “We’ve been literally pulled by all four limbs.”

While early AI models successfully analyzed biological structures to aid drug discovery, newer models are designed to sift through vast molecular datasets to engineer entirely new materials. CuspAI has assembled a high-profile team to tackle this challenge, including co-founder and respected AI researcher Max Welling, alongside prominent industry advisors like Yann LeCun, Geoffrey Hinton, and semiconductor veteran Abhi Talwalkar.

Despite the enthusiasm, experts urge caution regarding the immediate impact of machine learning. David Fairen-Jimenez, a professor of molecular engineering at the University of Cambridge, noted that creating revolutionary materials remains a long, expensive process. “What I don’t agree with is that there will be some magic coming from machine learning alone,” he stated.

CuspAI’s own early experiments illustrate these difficulties. An early collaboration with Meta screened a library of 300 trillion possible structures to find candidates for carbon capture, yet the handful that were successfully synthesized failed to outperform existing products. “They weren’t state of the art,” Edwards admitted.

None of the startup’s molecular candidates have left the lab yet, but the company remains optimistic about bridging the gap between digital design and real-world testing. Welling emphasized that the industry frequently struggles with a lack of reliable data and physical testing facilities, observing that “people underestimate the friction of actually doing the experiment.”