Jeff Bezos just backed a two-year-old Cambridge startup that turned a $520 million valuation into $2.6 billion in ten months, but the money is only half the story.
CuspAI closed a $450 million Series B this week, led by Kleiner Perkins and NEA with participation from Bezos Expeditions, Jeff Bezos's family investment vehicle. Bloomberg reported that the round values the AI materials startup at $2.6 billion, up from $520 million last September. That's roughly a five-fold jump in under a year. Fast, by any startup's standard. The more useful signal is who lined up beside the money: Nvidia, Meta, Samsung, Hyundai Motor Group, Applied Materials, Tokyo Electron and Lam Research are among 48 founding partners in CuspAI's new AI Materials Foundry.
You should pay attention to the Foundry before you pay attention to the valuation. Venture rounds can get silly. Industrial coalitions are harder to fake when the names come from chipmaking, memory, manufacturing tools and car batteries. CuspAI is trying to make itself part of the plumbing for materials discovery. Not just another AI company with a big number attached to its name.
Founded in 2024 by chemist Chad Edwards and AI researcher Max Welling, CuspAI describes its product as a search engine for the material world. Give the software the properties you want, conductivity, strength, heat resistance, carbon capture, whatever the problem demands, and its models suggest molecular candidates that can then be tested in simulation and the lab. No years of blind bench chemistry first. The company says its approach can find usable materials up to ten times faster than traditional research and development. That claim now has to survive contact with customers.
The Foundry Is the Hard Part
Nvidia and Meta don't join a startup's coalition for goodwill, and neither does Lam Research. Materials discovery is a real bottleneck in chip manufacturing. Applied Materials said in March that one reaction discovery run of about 15,000 pathway searches, which previously took one and a half weeks on 400 CPU cores, could be completed in five hours on an Nvidia B200 GPU. That's not a branding problem. That's a physics and compute problem sitting inside the AI supply chain.
CuspAI is positioning the Foundry as shared infrastructure. The pitch is to pool compute, data, lab access and scientific expertise so large companies stop duplicating research programs in private corners. The Guardian reported that the group is aimed at materials for chips, energy networks, batteries and related industrial systems, including work that could reduce reliance on rare metals such as iridium and ruthenium. That's a useful list because it shows where the commercial pressure is coming from. Chips need better materials. Batteries need better materials. Grid infrastructure needs better materials. AI made the compute problem visible, but the material problem was already there.
The company is also widening its footprint. Reports this week said CuspAI is adding a Singapore office to hubs in Cambridge, Amsterdam, Berlin and Tokyo, while expanding in the US with John Giannandrea, the former Apple and Google executive, helping build American operations. It has also added Abhi Talwalkar, an AMD board member and Lam Research chairman, to its advisory board.
Good. Now the burden rises.
Edwards previously helped grow Quantinuum, the quantum computing company. Welling was a distinguished scientist at Microsoft Research before running technology for Qualcomm. The advisory board already had unusual weight for a company of roughly 65 employees: Geoffrey Hinton, who won the 2024 Nobel Prize in Physics, and Yann LeCun, the Turing Award winner who left Meta to build AMI Labs, both sit on it. Names like that open doors. They don't ship a new semiconductor material by themselves.
Britain Gets a Rare AI Win
Frankly, the UK angle matters here. CuspAI is headquartered in Cambridge and has been held up as one of the flagship companies in the British government's AI for Science push. The Guardian reported that the UK government backed the latest round, which came to about 330 million pounds. The company also has access to Isambard-AI, the Bristol supercomputer built with Nvidia and HPE and backed by 225 million pounds in public investment.
Britain's problem has not been a lack of scientists. It has been the habit of watching promising technical companies drift toward deeper American capital, larger compute budgets and bigger commercial partners. CuspAI is still in Cambridge, opening a King's Cross office and pulling Jeff Bezos, Kleiner Perkins and NEA into a British deep-tech company. That's money with a job.
CuspAI's funding history shows how quickly the story changed. The company raised a $30 million seed round in June 2024. It followed with a Series A of more than $100 million in September 2025, co-led by NEA and Temasek, which put the valuation at $520 million. Now it has $450 million more at $2.6 billion. Few startups post that kind of markup twice in twelve months, let alone one working on physical chemistry rather than a chatbot wrapper.
Here's the thing: investors have overpaid for hard science before. Plenty of impressive coalitions have looked serious on launch day and then quietly dissolved into pilots, white papers, missed deadlines and private disappointment. CuspAI gets no exemption from that just because Bezos wrote a check or Nvidia joined the room. Its next test isn't another funding round. It has to help partners discover materials they can test and put into real products.
If it does, CuspAI becomes one of the more important AI companies in Europe. If it doesn't, this week will look like another moment when capital mistook scientific promise for industrial proof.
Also read: The EU just ordered Google to hand Android and Search data to its AI rivals • China's Kimi K3 and Qwen3.8 Are Rattling Silicon Valley's AI Giants • Zhongji Innolight Raises Hong Kong IPO Target to $7 Billion
