# CuspAI's Chad Edwards lines up $400M at a $2.6B valuation > Source: > Published: 2026-06-19 01:10:06+00:00 [Dr Chad Edwards](https://northzone.com/insights/material-revolution-a-portrait-of-cuspais-chad-edwards?ref=runtimewire) and [Prof. Max Welling](https://www.turing.ac.uk/people/guest-speakers/max-welling?ref=runtimewire) are lining up the biggest test yet for [CuspAI](https://www.cusp.ai/?ref=runtimewire): a reported $400 million financing at a $2.6 billion valuation for a two-year-old materials discovery company whose promise is to make industrial chemistry move more like software. [TechStartups](https://techstartups.com/2026/06/17/jeff-bezos-backs-ai-materials-startup-cuspai-in-400m-round-at-2-6-billion-valuation/?ref=runtimewire), citing the [Financial Times](https://www.ft.com/content/4c479227-567c-435e-a4d3-795dd957b347?ref=runtimewire), reported on June 17 that CuspAI is raising the round with participation from [Jeff Bezos (@JeffBezos)](https://x.com/JeffBezos?ref=runtimewire)'s family office, [Bezos Expeditions](https://www.bezosexpeditions.com/?ref=runtimewire), and [Kleiner Perkins](https://www.kleinerperkins.com/?ref=runtimewire). The important detail is not just the size. Parallel reporting in [The Next Web](https://thenextweb.com/news/cuspai-bezos-400-million-materials-ai?ref=runtimewire) said term sheets have been signed but the financing has not formally closed, which makes this a live fundraising signal rather than a completed balance-sheet event. If the round closes on the reported terms, it would mark a roughly 5x step-up from the $520 million valuation [Fortune reported](https://fortune.com/2025/09/10/cuspai-raises-100-million-in-new-venture-capital-funding-ai-for-chemistry/?ref=runtimewire) around CuspAI's September 2025 Series A. That is the kind of valuation expansion investors usually reserve for companies that can show software-like compounding. CuspAI is trying to earn it in a market where the hard part starts after the model proposes a molecule: synthesis, testing, qualification, manufacturing and customer adoption. ### Edwards is selling speed to an industry built around patience Edwards did not arrive at CuspAI as a lab-only founder. Northzone's profile describes him as a chemist from rural Wales who became the first in his family to attend university, then moved into deep-tech commercialization. Before CuspAI, he was the first commercial hire at Cambridge Quantum Computing, later part of Quantinuum, where his job was to figure out how a research-heavy quantum business would make money. That operator background matters because CuspAI is not pitching a narrow research tool. Edwards's thesis, as he has described it, is that industry keeps running into the same bottleneck: the materials it needs for energy, climate, computing and manufacturing do not exist yet, or take too long to discover. In Northzone's telling, he started asking why generative AI could create text, images and code from prompts, but not candidate molecules or materials from desired properties. Welling gives the company its technical center of gravity. The Alan Turing Institute lists him as a Distinguished Scientist at Microsoft Research AI4Science and Professor of Machine Learning at the University of Amsterdam, with prior roles including VP at Qualcomm Technologies. His work has touched variational autoencoders, graph neural networks and equivariant neural networks, all relevant to the problem CuspAI is attacking: representing physical structures in ways machine-learning systems can reason about. CuspAI's own site describes a mission to unlock materials breakthroughs in "months, not millennia" and lists an advisory bench that includes Geoffrey Hinton, Yann LeCun, Kristin Persson, Verity Harding, Martin van den Brink and Lord John Browne. The company calls its founding group among the most cited in AI, chemistry and engineering. That is company framing, but the roster explains why investors have been willing to price CuspAI as more than a conventional chemistry software startup. ### The real proof is outside the model CuspAI's product is often described as a search engine for materials: a customer specifies properties, and the system proposes candidate structures that may satisfy them. The useful version of that product is not a pretty generated molecule. It is a candidate that can be synthesized, tested and eventually used in a product or process where performance, cost and manufacturability all matter. That is why CuspAI's work with [Kemira](https://www.kemira.com/news-and-stories/newsroom/releases/kemira-and-cuspai-forge-strategic-partnership-to-pioneer-ai-driven-materials-innovation/?ref=runtimewire) is more important than the slogan. The Finnish chemicals company announced a strategic partnership with CuspAI in July 2025 focused first on PFAS removal from water. In May 2026, [Kemira and CuspAI said](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) the project had explored about 300 trillion possible material structures, generated more than 5,000 novel designs with property data, and narrowed the field to about 20 priority candidates moving into further development and testing. That result still sits on the early side of commercialization. It is not a commercial PFAS remediation product in the market. But it is the kind of industrial proof point CuspAI needs: a defined customer problem, constraints around stability and manufacturability, and candidates moving from computation toward testing. For a company being valued in the billions before broad public revenue metrics are established, these are the details that matter. The same discipline should apply to customer language. TechStartups reported that CuspAI's customer list includes ASML, Meta and Hyundai. Publicly verifiable partnerships include Kemira, and earlier reporting and deal announcements have pointed to Meta-related direct air capture work and Hyundai involvement. The public record still does not establish revenue, pricing or a paying-customer count. ### A $400 million round would change the clock CuspAI announced a $30 million seed round in June 2024 in a [TechCrunch](https://techcrunch.com/2024/06/18/cuspai-raises-30m-to-create-a-gen-ai-driven-search-engine-for-new-materials/?ref=runtimewire) story that framed the company as a generative AI search engine for new materials. That round was led by Hoxton Ventures, with participation from Basis Set Ventures and Lightspeed Venture Partners, among others. By September 2025, CuspAI had moved into a $100 million Series A. [Mishcon de Reya](https://www.mishcon.com/news/mishcon-de-reya-advises-cuspai-on-100-million-series-a-funding-round?ref=runtimewire), which advised on the deal, said the round was led by [New Enterprise Associates](https://www.nea.com/?ref=runtimewire) and [Temasek](https://www.temasek.com.sg/en/index?ref=runtimewire), with participation from NVentures, Samsung Ventures, Hyundai Motor Group, existing investors and angels. Mishcon said the capital would go toward expanding the team, opening new offices with a focus on Asia, and accelerating synthesis-aware generative AI models. The reported new round would be different in kind. A $400 million financing would give CuspAI the capital to hire deeper scientific teams, fund validation work, expand industrial partnerships and absorb the long lag between a promising candidate and a deployed material. It would also put pressure on Edwards and Welling to show that CuspAI is not merely a high-status research organization wrapped in venture packaging. Materials discovery does not naturally obey SaaS timelines. Even when AI compresses the search process, customers still have to trust the output in physical systems: water treatment plants, semiconductor manufacturing, batteries, carbon capture systems or advanced industrial processes. That makes CuspAI's central question sharper than a standard AI startup's: can better models shorten the slowest parts of the physical economy enough to justify venture-scale returns? ### Bezos is circling physical AI The Bezos name gives this round a second storyline. It comes days after RuntimeWire reported that [Jeff Bezos's Prometheus is a $41 billion bet on AI for physical engineering](/article/jeff-bezos-prometheus-ai-engineer-41-billion), with Prometheus saying it had raised $12 billion to build AI systems for designing and manufacturing complex products such as jet engines. CuspAI is not Prometheus. It is a separate Cambridge company built around materials, not a broad engineering lab. But the proximity of the two reported bets is notable. Bezos's capital is pointing at the same broad thesis: the next major AI market is not limited to chat interfaces and coding assistants. It is AI that can improve the design of things that have to work in the physical world. For Edwards, that creates an opening and a burden. The opening is obvious: investors are looking for founders who can translate frontier AI into atoms, factories and industrial supply chains. The burden is that CuspAI must prove a company can capture enough value from materials discovery without being slowed to death by the very industries it wants to accelerate. The reported $2.6 billion valuation is therefore less a victory lap than a deadline. Edwards and Welling have assembled the scientific credibility, the venture syndicate and the early industrial examples. The next phase is whether CuspAI can turn model-generated candidates into materials customers will buy, qualify and depend on.