CuspAI raises $450M to build a global AI materials foundry

CuspAI, a Cambridge, UK-based materials discovery startup, raised $450 million in Series B funding on July 20, achieving a $2.6 billion valuation two years after founding. Kleiner Perkins and NEA co-led the round, with Bezos Expeditions making a significant investment, as the company plans to build an AI Materials Foundry linking industrial partners with laboratories and computing infrastructure.

Dr Chad Edwards https://www.linkedin.com/in/edwardschad/?ref=runtimewire and Prof Max Welling https://www.linkedin.com/in/max-welling-4a783910/?ref=runtimewire confirmed a $450 million Series B for CuspAI https://www.cusp.ai/?ref=runtimewire on July 20, according to Tech.eu https://tech.eu/2026/07/20/jeff-bezos-and-sovereign-ai-back-cuspai-in-450m-raise/?ref=runtimewire . The round gives the Cambridge, UK-based materials discovery startup a $2.6 billion valuation roughly two years after its founding. Kleiner Perkins https://www.kleinerperkins.com/?ref=runtimewire and NEA https://www.nea.com/?ref=runtimewire co-led the financing. Bezos Expeditions https://www.bezosexpeditions.com/?ref=runtimewire , the family office of Jeff Bezos https://x.com/JeffBezos?ref=runtimewire , made what Tech.eu described as a "significant" investment. The size of that check was not disclosed. New backers also include Glade Brook Capital Partners https://www.gladebrookcapital.com/?ref=runtimewire , Lux Capital https://www.luxcapital.com/?ref=runtimewire , AMD Ventures, Tru Arrow Partners, StepStone https://www.stepstonegroup.com/?ref=runtimewire and the UK's Sovereign AI Fund https://sovereignai.gov.uk/?ref=runtimewire . Existing investors Temasek, Basis Set Ventures https://www.basisset.com/?ref=runtimewire , Giant Ventures https://www.giant.vc/?ref=runtimewire , Phoenix Court https://www.phoenixcourt.vc/?ref=runtimewire and Northzone https://www.northzone.com/?ref=runtimewire participated again. The five-fold valuation increase since CuspAI raised $100 million at a $520 million valuation in September 2025 reflects the scale of the bet Edwards and Welling are making. CuspAI wants to sit between AI-generated material candidates and the laboratories, data providers and manufacturers required to turn those candidates into physical products. A founder pair built around AI materials discovery Edwards is CuspAI's CEO and co-founder. Welling is its CTO and co-founder, and brings the more visible research profile in the supplied materials. A National University of Singapore biography https://ims.nus.edu.sg/events/colloquium maxwelling/?utm source=openai&ref=runtimewire describes Welling as a machine-learning professor at the University of Amsterdam, a Merkin Distinguished Visiting Professor at Caltech, a theoretical high-energy physics PhD from Utrecht University and a former postdoc who worked with Geoffrey Hinton. The research brief also associates him with senior roles at Microsoft and Qualcomm and with work on generative models, including variational autoencoders. CuspAI's adviser list reinforces that scientific positioning. The company names Hinton, Yann LeCun, Kristin Persson, Verity Harding, Martin Van Den Brink and Lord John Browne among its advisers. CuspAI starts with the properties a customer needs, then generates and evaluates possible materials against those requirements. Tech.eu identifies ASML and Meta as customers and says CuspAI is targeting applications across clean energy, semiconductors, manufacturing, water treatment and carbon capture. The round buys the physical layer CuspAI says the Series B will fund expansion in the US, Asia-Pacific and Europe, along with an AI Materials Foundry https://www.cusp.ai/?ref=runtimewire that links industrial partners with laboratories, scientific data and computing infrastructure. Tech.eu says the foundry has more than 40 founding partners, including Nvidia, Meta, Samsung and Hyundai. That network is central to the investment case. Generative models can produce thousands of plausible structures, while synthesis, testing, safety review and manufacturing determine whether any of them becomes useful. CuspAI's model works only if it can connect software-generated candidates to laboratories and industrial operators fast enough for customers to trust the results. A project with water-treatment chemicals producer Kemira https://www.kemira.com/news-and-stories/newsroom/releases/new-ai-designed-materials-show-promising-potential-to-remove-forever-chemicals-from-drinking-water-in-industry-first-breakthrough/?ref=runtimewire shows the attrition between design and validation. According to the research brief, the project generated more than 5,000 designs for removing PFAS from water, with about 20 candidates advancing to development and testing. The public evidence stops at candidates entering development and testing. None of the supplied sources establishes that a CuspAI-designed material has entered industrial production. The foundry gives Edwards and Welling a way to address that gap by bringing labs and industrial operators into the same network as CuspAI's models. Competitors are organizing around the same bottleneck. Periodic Labs https://periodic.com/?ref=runtimewire , which launched with a $300 million founding round https://a16z.com/announcement/investing-in-periodic-labs/?ref=runtimewire , is building autonomous laboratories where AI systems can commission and learn from experiments. Orbital Industries https://www.orbitalindustries.com/news/press/orbital-industries-series-b-50-million?ref=runtimewire raised $50 million in May to combine materials discovery with the engineering and manufacturing of data-center cooling systems. Altrove https://www.altrove.ai/resources/seed-announcement?ref=runtimewire raised a $10 million seed round and says it is working toward kilogram-scale production of AI-designed materials. CuspAI is choosing a network model across data, compute, labs and established manufacturers. That can extend CuspAI into several industries without requiring Edwards and Welling to build every laboratory or factory themselves. It also makes execution dependent on partners whose testing schedules, manufacturing constraints and commercial priorities CuspAI does not control. A $2.6 billion price before production proof The financing values CuspAI at five times the $520 million valuation reported for its September 2025 Series A. In its Series A announcement https://medium.com/%40CuspAI/securing-our-100m-series-a-to-revolutionise-materials-discovery-with-ai-e6705e8c8fd3?ref=runtimewire , CuspAI said it had raised more than $100 million. CuspAI has not disclosed revenue, annual recurring revenue, headcount, commercial terms for its customer relationships or the share of foundry participants that pay CuspAI. The valuation therefore prices the reach of the partner network, the founders' scientific credentials and the possibility that CuspAI becomes a common design layer across semiconductors, energy and climate technology. The cumulative funding figure also lacks a full reconciliation in the supplied materials. Tech.eu reports more than $670 million raised. CuspAI previously disclosed a $30 million seed and a $100 million-plus Series A, but the supplied sources do not identify the structure or timing of all capital beyond the headline rounds. The Series B gives Edwards and Welling the resources to test CuspAI's model at industrial scale. Its next measure of progress will come from the laboratory and manufacturing floor: how many generated candidates survive validation, how quickly partners can scale them and whether those materials perform economically outside a model.