Jeff Bezos believes that the next major breakthrough in artificial intelligence will not come from any chatbot, but from the discovery of new materials that could transform chip manufacturing, clean energy, and advanced manufacturing. This is precisely the kind of thinking that attracted investors to the British startup CuspAI, which announced on Monday that it had raised $450 million, sending its valuation soaring to $2.6 billion. Bezos invested in the company through his investment vehicle Bezos Expeditions, joined by Nvidia, Meta, and dozens of other technology and industrial companies.
What CuspAI Does
The company has taken on a problem that has held back science for decades. Developing a new material used to mean years of laboratory experiments, numerous dead ends, and enormous costs. CuspAI and its artificial intelligence models first calculate how a new material is likely to behave, and only then does someone produce it. This allows them to select a few promising candidates from millions of theoretical possibilities that are actually worth testing in the laboratory.
Investors particularly appreciate one thing about this approach. It shortens the path from scientific theory to a product that can actually be sold. Most major technological leaps ultimately depend on some kind of material, and many of them are now waiting for materials that no one has discovered yet. CuspAI has built a tool that could change that.
The company is based in Cambridge, UK, has only been operating for two years, and originally focused on something entirely different: materials for carbon capture and water purification. Over the past year, however, it has redirected its efforts in response to demand from chip manufacturers. CEO and co-founder Chad Edwards said that semiconductor manufacturers and their suppliers are desperately seeking new materials and, in his words, are literally tearing his company apart. As a result, 80 percent of its research this year is focused specifically on materials for chips.
The company launched one of its first projects together with researchers from Meta. They built a library of 300 trillion possible structures in a class of chemicals suitable for capturing carbon dioxide. They eventually narrowed that enormous list down to ten candidates. Edwards admitted, however, that they managed to produce only a few of them, and none outperformed what is already available on the market today. The company is now using the same library to search for substances capable of removing so-called forever chemicals from water. Finnish chemical company Kemira Oy is expected to produce and test 20 new structures this year.
New Materials Are Key to Chips
Manufacturing the most sought-after semiconductors consumes enormous amounts of energy and relies on rare metals. One of CuspAI’s main goals is to reduce or completely eliminate dependence on metals at risk of supply disruptions, such as ruthenium and iridium. Both are commonly used in chip manufacturing today. However, both metals are extremely scarce worldwide, are obtained mainly as byproducts of platinum and nickel mining, and their resources are largely concentrated in a single country, meaning that a single disruption can cause supply problems.
According to Nvidia, demand is enormous. Geetika Gupta, a director at the company, said that customers across the entire market are seeking new materials, from energy storage to data center cooling. Nvidia will most likely develop new substances through three-way collaborations involving itself, CuspAI, and another partner. Most companies closely guard this information as a trade secret.
A Project Called AI Materials Foundry
Alongside the funding, the company has launched another initiative. It introduced a project called AI Materials Foundry, a coalition of more than 45 technology, industrial, and research organizations. In addition to Nvidia and Meta—or more specifically, its Fundamental AI Research team—the coalition also includes automaker Hyundai Motor Group and Samsung. Nvidia provides computing power, and the goal is to give scientists a faster path from computer predictions to real-world laboratory validation. To support this effort, the company is opening an office in Singapore and expanding its teams in the United Kingdom, the Netherlands, Germany, Japan, and the United States. Laboratories will operate in Cambridge, Singapore, and the San Francisco Bay Area.
To gain credibility, CuspAI has brought on board prominent names in the world of artificial intelligence. Co-founder Max Welling is a respected researcher, while AI pioneers Yann LeCun and Geoffrey Hinton serve on its advisory board. The company has also appointed semiconductor industry veteran Abhi Talwalkar, who sits on AMD’s board of directors, to its leadership team, and hired John Giannandrea, a former Google and Apple executive, on a part-time basis to launch its California office.
Not Everyone Shares the Enthusiasm
Experts nevertheless caution against excessive expectations. David Fairen-Jimenez, a professor of molecular engineering at the University of Cambridge, acknowledges that machine learning has noticeably accelerated materials development. At the same time, however, he disagrees with the idea that any magic comes from machine learning itself. Welling has raised a similar point. According to him, people underestimate how many difficulties a real-world experiment involves. It looks simple, but it is anything but simple. The company has not yet introduced any material ready for commercial deployment.
Sources: ft.com and bloomberg.com
