CuspAI

CuspAI is a company in Cambridge, England, that builds software for finding new materials. Its home page calls it "AI-powered materials discovery". It was incorporated in March 2024 and co-founded by Chad Edwards, its chief executive, and Max Welling, a machine-learning researcher. Edwards is a chemist who had worked at Google, BASF and Quantinuum.^[1][2][3] Its first target was a material that captures carbon dioxide from air. By 2026 it was also working on semiconductors and batteries, and its main product was a network of partner laboratories, data owners and software that it calls the AI Materials Foundry.^[3][4]

Molecular Reality Corporation lists CuspAI fifth in the column of companies it calls brain labs, with the chart caption "AI-powered materials discovery".^[5]

Chad Edwards is a chemist who had worked at Google, BASF and Quantinuum before the company began. Max Welling is better known to machine-learning readers: with Diederik Kingma he published the variational autoencoder in December 2013, a generative model that learns a compressed code for its training examples and can decode new samples from it.^[3][6] In 2021 he was a coauthor of a paper on E(n)-equivariant graph neural networks, which are built so that rotating or reflecting a molecule leaves its predicted energy unchanged and rotates its predicted forces, a symmetry that a model of three-dimensional structures would otherwise have to learn from examples.^[7]

Companies House records Cusp AI Limited as incorporated in England and Wales, with a registered office on Station Road in Cambridge.^[2][1] The company announced itself on June 18, 2024, with a seed round of $30 million. TechCrunch's headline called the product a "GenAI-driven search engine for new materials". Edwards described the goal as a reversal of the usual order of work, in which a chemist makes a compound and then measures it: properties go in and materials come out. The first application was carbon capture, in the form of "molecular sponges that selectively absorb carbon dioxide from the air", and the article reported a collaboration with the AI research group at Meta on sorbents for direct air capture.^[3]

On July 20, 2026, CuspAI launched the AI Materials Foundry with more than 45 founding partners. The Foundry has hubs in the United States, Europe and Asia-Pacific, and NVIDIA supplies the computing. It runs Meta's Universal Model for Atoms (UMA) among its atomistic models, and a CuspAI agent called MIRA that predicts properties across millions of candidates, picks the promising ones, drafts a synthesis route and sends the job to a partner facility. With NVIDIA's ALCHEMI team the company wrote kUPS, an open-source toolkit for molecular simulation.^[4]

The company says it holds exclusive AI training rights to two standard collections of crystal structures: the Cambridge Structural Database, kept by the Cambridge Crystallographic Data Centre, and the Inorganic Crystal Structure Database, kept by FIZ Karlsruhe.^[4] Its site lists the Belgian research center Imec as a laboratory partner.^[1]

For the Finnish chemicals company Kemira, CuspAI says it screened 300 trillion molecular structures in six months and delivered 20 validated candidates, a job that had taken the customer years. The ratio is one candidate for every 15 trillion structures examined.^[4]

Company materials place CuspAI in the brain column and also in the list of "Megacap 'Brain' AI Labs" that they expect to buy data from the eye.^[5] Two others in the column, Periodic Labs and Lila Sciences, run autonomous laboratories of their own, while CuspAI describes a network of partner laboratories.^[4]

• Bioptimus
• Brain lab
• Self-driving laboratory