Finally confirmed today after plenty of leaks, CuspAI's $450m Series B, at a $2.6bn valuation, is a significant moment, both for trends in AI, as well as European venture. Unlike so much of what comes out of Silicon Valley these days, the Cambridge, UK-based company (founded by Dr Chad Edwards, CEO and Co-Founder and CTO Professor Max Welling) is not just another application-layer AI business, but assembling what amounts to a sort of production system for industrial materials. It mashes together proprietary scientific data, frontier AI models, GPU compute, and customers with problems, the solutions to which could rake in billions.
CuspAI’s AI Materials Foundry brings more than 45 organisations (including Nvidia, Meta, Samsung, Hyundai, Applied Materials and Lam Research) into a network coordinated by its MIRA platform. Here’s how it works: a company specifies the properties it needs; MIRA generates candidates, predicts performance, plans how they could be made, routes them to suitable laboratories and then learns from the results. The result is an acceleration of problem-solving which would have seemed inconceivable just a few years ago.
For instance, CuspAI claims a project with Finnish chemicals group Kemira screened 300 trillion molecular structures and produced 20 validated candidates in six months rather than several years, as would have been the case pre-AI. The pace of change this implies is absolutely jaw-dropping.
But what’s interesting isn’t just CuspAI’s news, but the general trend it's part of, especially across the UK/Europe, which appears to be leaning into its ability to turn DeepTech, science-based ideas into industrial-scale startups.
Apoha's emergence (from the UK) with $36m recently (as we interviewed on the Pathfounders podcast) is further evidence of this trend. Founded in 2021 by Shamit Shrivastava and Anshika Srivastava, the London and San Francisco company is building Liquid State Intelligence: a data layer describing how molecules, materials and formulations behave under stress and real-world conditions.
Its first product, VIBE, uses tiny samples to produce more than 1,000 empirical descriptors within minutes. In work with Boehringer Ingelheim, Apoha says the system identified high-risk antibody candidates with more than 90% precision from as little as eight micrograms of material. It is targeting pharmaceuticals first but says the same data could be used in food, materials and physical-world AI.
While Apoha is not exactly a direct CuspAI competitor, what it’s working on could end up making it either a supplier, partner or, perhaps even one day, a rival in Physical AI.
What both companies are showing us is that while AI can now generate almost unlimited hypothetical physical structures, the data about how matter behaves (whether it be for a machine part or a pharma product) is the moat these companies are assembling, together with these deeply integrated partners. This suggests that the strongest companies in this arena will own the loop between AI model predictions and the physical validation produced, in concert with their customers, and not just an AI model elegantly hanging in space.
CuspAI's defensibility rests on spanning more of that loop by securing access to important experimental datasets, and working with industrial groups capable of validating discoveries, with every completed programme potentially improving its models.
Admittedly, the network effects are not automatic, as partners can keep their most valuable results private, but then that surely is the draw for working with CuspAI in the first place. CuspAI’s, for its part, can now prove that its ‘Foundry’ can become shared infrastructure which benefits all its partners, in turn. But CuspAI’s fundraising isn’t the only game in town. This is a rapidly developing field.
London/San Francisco-based Orbital Industries recently raised a $50m Series B led by Plural, with NVentures, Radical Ventures, Compound and Fly Ventures participating. Orbital combines materials discovery, engineering and manufacturing, initially producing cooling fluids and modular infrastructure for data centres.
Meanwhile, US-based Periodic Labs raised a $300m founding round led by Andreessen Horowitz, with Felicis, DST, NVentures and Accel among its backers. Founded by former OpenAI and Google DeepMind researchers, it is building AI scientists and autonomous laboratories that generate proprietary experimental data, beginning with materials problems in semiconductors, energy and advanced manufacturing. And Nvidia-Backed, Cambridge, Massachusetts-based Lila Sciences has raised $550m, including a $350m Series A, to build broader AI Science Factories across materials, chemistry and biology. Its models generate hypotheses, control instruments and learn from physical experiments.
Europe has smaller, specialised challengers. Paris-based Entalpic raised an €8.5m seed round to apply generative AI, atomistic simulation and experimental feedback to catalysts, batteries and industrial chemistry. Imperial College, UK spinout Polaron raised $8m from Racine2, Speedinvest and Futurepresent to connect manufacturing processes, material microstructures and performance. UK-based MatNex designs proprietary materials to specification and owns the resulting IP, starting with rare-earth-free magnets. Glasgow-based Chemify has raised more than $90m across its Series A and Series B to combine AI, robotic laboratories and a chemical programming language for designing and synthesising molecules and materials.
While these startups won’t all compete in exactly the same fields or for the same customers. Collectively, they are building overlapping layers of a new industrial stack: generation, simulation, behavioural measurement, laboratory automation, synthesis and manufacturing. However, CuspAI's ambition is notable for the scale of its ambition to orchestrate more of those layers through one platform and partner network.
For Britain, the Sovereign AI Fund's participation in the round is strategically important, even if its likely to be a smaller partner.
The question is whether the UK can keep its foot in the door, given that CuspAI's lead investors are American, Nvidia will supply the key compute, and the company is expanding across the US, Europe and Asia.
What the Sovereign AI Fund could do is reinforce its position by providing compute access, dataset funding and procurement pathways.
For the venture capital industry, CuspAI’s raise simply confirms the trend of AI’s incursion into materials. Investors are betting that scientific data and experimental infrastructure can produce software-like network effects. Let’s see.
