Nine months ago, CuspAI was a $520 million bet on an idea: that a search engine for molecules could do for materials science what web crawlers did for information. On Monday the Cambridge, England startup confirmed a $450 million Series B that values it at $2.6 billion — a five-fold jump in under a year, with Jeff Bezos writing a check through his family office and roughly 45 industrial giants signing on as partners.
The round, larger than the $400 million first reported by the Financial Times in June, was led by Kleiner Perkins and NEA, with significant participation from Bezos Expeditions. It brings CuspAI’s total raised past $650 million for a company founded in 2024. New backers include Lux Capital, AMD Ventures, Glade Brook Capital Partners, StepStone, Tru Arrow Partners, Britain’s Sovereign AI Venture Fund, the Netherlands’ Invest-NL and, notably, John Doerr investing personally. Existing investors Temasek, Prosus, Northzone, Basis Set Ventures, Giant Ventures, Touring Capital and Phoenix Court all returned.
What CuspAI Actually Does
CuspAI was co-founded by Dr Chad Edwards, previously commercial co-founder of Cambridge Quantum Computing and later global head of strategy at Quantinuum, and Professor Max Welling, a University of Amsterdam machine-learning researcher and former VP of technology at Qualcomm AI Research. Welling is one of the more decorated names in the geometric and probabilistic deep learning world, and CuspAI is his attempt to point that machinery at atoms rather than pixels.
The product is best described as a search engine over the space of possible materials. A customer specifies target properties — a sorbent that grabs CO2 at a given partial pressure, a dielectric that survives a specific thermal budget, a filter that strips PFAS from drinking water — and CuspAI’s generative models and molecular simulation pipeline propose candidate structures, then screen them computationally before anything reaches a lab.
Alongside the funding, the company launched what it calls the AI Materials Foundry: a global network of data, labs, compute and domain expertise with more than 45 founding partners, among them Nvidia, Meta, Samsung, Hyundai Motor Group, Henkel, Applied Materials, Tokyo Electron and Lam Research. Bloomberg reported the same day that CuspAI is working with Nvidia specifically on materials for chipmaking. Headcount is spreading fast — a new Singapore office joins Cambridge, Amsterdam, Berlin, Tokyo and the United States, with former Apple and Google executive John Giannandrea helping stand up US foundry operations and AMD board member Abhi Talwalkar joining the advisory board.
The Pitch, In The Founders’ Words
The framing from Edwards and Welling is deliberately civilisational. ‘If we don’t make progress fast, the next 50 years of industrial progress will be constrained by a single challenge: the world needs materials that don’t yet exist,’ the pair wrote in the post announcing the round. ‘That’s what we’re on a mission to solve — combining frontier agentic AI with deep domain expertise, exclusive data access and close customer partnerships.’
Investors echoed the bottleneck argument, but with a telling emphasis on buildability. ‘Most big leaps in technology come down to a material, and the next set, from cheaper carbon capture to semiconductors and cleaner water, is stuck waiting on materials nobody has discovered yet,’ said Josh Coyne, the Kleiner Perkins partner who co-led the round. ‘What sets Max, Chad, and the team apart is that they design for materials that can actually be built, not just ones a model can dream up.’
NEA’s Philip Chopin, who also co-led the Series A, put the differentiator the same way: CuspAI’s edge is ‘their ability to bridge the gap between frontier AI and the physical world, winning them industry leaders as their early partners.’
Analysis: AI For Science Grows Up
Two things are happening in this deal at once, and it is worth separating them.
The first is a category shift. For three years, ‘AI for science’ has largely meant papers — DeepMind’s GNoME announcing hundreds of thousands of hypothetical stable crystals, protein-structure models racking up citations. Venture capital funded the demos politely. A $450 million round at $2.6 billion for a two-year-old company is a different posture: this is growth-stage capital underwriting an industrial supply chain, not a research programme. Bezos Expeditions, Doerr’s personal money and two sovereign funds arriving on the same cap table signal that scientific discovery is now being priced as infrastructure.
The second is the thing that makes materials the hardest test case for that thesis. Software AI is validated by a benchmark; materials AI is validated by a furnace. A model can propose a metal-organic framework with beautiful simulated CO2 selectivity, and it can still be unsynthesisable at scale, unstable under humidity, or dependent on a precursor with a fragile supply chain. Note how carefully both lead investors chose their language around candidates that ‘can actually be built.’ That is not marketing filler. It is the diligence question, stated in public.
Which is what the AI Materials Foundry is really for. Forty-five partners with fabs, pilot lines and proprietary process data are not just a distribution channel. They are the closed loop — the source of experimental results that turn a generative model from a plausibility generator into something with a calibrated hit rate. The exclusive data access Edwards and Welling mention is arguably a harder moat than the model architecture.
What To Watch
Three markers over the next twelve months. First, whether any CuspAI-designed material reaches qualified production with a named partner — a Samsung or Lam Research disclosure would be worth more than another funding headline. Second, whether the company publishes a hit rate: how many computational candidates survive synthesis. Third, revenue mix. Partner-funded research contracts and product royalties are very different businesses, and only one of them justifies $2.6 billion. The capital has arrived. The furnace is next.
“What sets Max, Chad, and the team apart is that they design for materials that can actually be built, not just ones a model can dream up.”— Josh Coyne, Partner, Kleiner Perkins