CuspAI Ltd., a startup using artificial intelligence to discover new materials, today announced that it has raised $450 million in funding.
The U.K.-based company also formed a chemical research consortium that includes Nvidia Corp. and Samsung Electronics Co.
CuspAI’s Series B round values it at $2.6 billion. The deal was led by Kleiner Perkins and NEA with a “significant contribution” from Bezos Expeditions. The funds were joined by more than a half-dozen others, including prominent tech investor John Doerr.
Discovering a new compound requires scientists to navigate trillions of material and molecular structure combinations. In the initial phase of a chemistry project, research teams narrow down the possibilities to a relatively small number of promising compound candidates. They then test the candidates in a simulation to evaluate their properties.
CuspAI has developed an AI platform called MIRA that speeds up the process. According to the company, the software can complete tasks that usually take years in six months. It does so using two open-source projects called UMA and kUPS.
UMA is a series of AI models that Meta Platforms Inc. released last year. The algorithms can simulate compounds at the atomic level to map out their electronic properties. Those electronic properties, in turn, can help researchers understand important details such as how a newly discovered compound interacts with other materials.
The other open-source technology that underpins MIRA, kUPS, was developed by CuspAI. Whereas UMA simulates materials at the atomic level, kUPS does so at the molecular level. The latter approach lends itself better to tasks such as understanding the internal structure of crystals.
In addition to analyzing molecules, kUPS can automate certain coding tasks. Graphics cards sometimes struggle to process datasets that contain information collected at both the atomic and molecular levels. Developers can use kUPS to increase the efficiency with which such data is processed. Additionally, the toolkit makes it easier to troubleshoot simulation errors.
CuspAI developed MIRA using several academic datasets to which it bought exclusive AI training rights. The company also used information from popular scientific journals.
MIRA underpins the AI Materials Foundry, a new research consortium that CuspAI launched in conjunction with its funding announcement. The group comprises more than 45 members, including Nvidia, Samsung, Meta and other tech industry heavyweights. Its goal is to speed up material discovery.
