NEW YORK: Producing the world’s most in-demand semiconductors requires enormous amounts of energy and access to rare minerals.
CuspAI, a two-year-old British startup, has raised nearly half a billion dollars on its bet that artificial intelligence (AI) can improve that process.
Yesterday, CuspAI launched the AI Materials Foundry, a coalition of more than 48 technology giants, industrial firms and research facilities.
The goal of this body, whose members include Nvidia Corp, Meta Platforms Inc, and Hyundai Motor Group, is to pool computing and scientific resources with the aim of building software that can help researchers develop new materials for chipmakers and other industries at a faster pace and for a lower cost than today’s methods.
To make this work, the Cambridge, a UK-based company, has raised US$450mil in Series B financing from investors including Jeff Bezos’s fund.
Chad Edwards, CuspAI’s co-founder and chief executive, said much of that funding will go towards supporting labs with foundry partners in Cambridge, Singapore and the San Francisco Bay area.
On the research front, he noted, 80% of its efforts this year will go towards material discovery.
One of the company’s goals, he added, was to reduce or eliminate the use in chipmaking of rare metals with supply-chain risks, such as ruthenium and iridium.
The new funding, led by venture firms Kleiner Perkins and New Enterprise Associates, with “significant participation” from Bezos Expeditions, according to CuspAI – gives the startup a US$2.6bil valuation, up from US$520mil last September.
Bloomberg and the Financial Times previously reported on elements of the deal.
CuspAI initially started out with a focus on finding new materials for carbon capture and water purification.
But Edwards said that the company decided to pivot over the past year in response to intense demand from the chip manufacturing supply chain, whose products are critical to powering AI.
Severe shortages of semiconductors, particularly in memory chips, sank shares last week across a range of tech companies dependent on the hardware.
Chipmakers and semiconductor suppliers are “frantically searching for new materials”, Edwards said.
“We’ve been literally pulled by all four limbs.”
Geetika Gupta, a senior director at Nvidia, said the chipmaker is seeing huge demand for new materials across its entire market, from energy storage systems to data centre cooling.
Through the Foundry, Gupta said Nvidia will most likely work on novel materials in “three-way collaborations” between CuspAI and a third partner.
“All of these companies, they treat a lot of this information as a secret sauce.”
CuspAI is among the growing number of AI developers turning to science to open new markets.
Earlier AI systems have shown they can learn from biological data, like protein shapes, to potentially discover new drugs.
Newer models can, in theory, sift through huge swaths of molecular data to generate designs for materials that could be useful for manufacturing and renewable energy.
Investors have piled into the field to back efforts led by former OpenAI and DeepMind researchers, as well as Prometheus, a US$41bil new company launched by Bezos.
Google DeepMind has also indicated that it will soon begin work on material discovery for AI.
CuspAI is trying to set itself apart with talent. Max Welling, its other co-founder, is a respected AI researcher.
Yann LeCun and Geoffrey Hinton, two pioneers of modern AI techniques, are on the startup’s advisory board.
Yesterday, the company also named Abhi Talwalkar, a semiconductor veteran and the director of Advanced Micro Devices Inc, to the board.
AMD’s venture unit joined in the latest round of CuspAI financing.
CuspAI, moreover, said it hired John Giannandrea, a former Google and Apple executive, to work part-time setting up the startup’s California outpost.
Machine learning systems have “certainly” sped up the process of developing new materials, said David Fairen-Jimenez, professor of molecular engineering at the University of Cambridge.
But he cautioned that designing novel materials that radically improve upon existing ones is a long and expensive process, and AI can only be of limited help.
“It’s super useful,” he said. “What I don’t agree with is that there will be some magic coming from machine learning alone.”
Fairen-Jimenez has founded two material science startups but doesn’t work with or advise CuspAI.
CuspAI is already familiar with how much trial and error goes into AI materials discovery.
One of the startup’s early projects involved working with researchers from Meta to find new materials capable of capturing planet-warming carbon dioxide and slowing down climate change.
To do this, CuspAI built a library of 300 trillion possible structures within a class of chemicals called metal organic frameworks, which have a metal atom surrounded by carbon-based molecular chains and are especially good at carbon capture.
The system then whittled that list down to 10. Turning that theoretical work into useful outcomes proved more difficult.
CuspAI was only able to synthesise fewer than 10 of those chemicals, and none outperformed what was already commercially available, said Edwards. “They weren’t state of the art.” — Bloomberg
