Jeff Bezos backs CuspAI in $400 million AI materials round

Reporters at the Financial Times said Wednesday that Jeff Bezos will invest in CuspAI as part of a $400 million financing round. Investors would value the Cambridge, England, AI materials company at $2.6 billion, up from $520 million in September.

Chad Edwards and Max Welling founded CuspAI in 2024. The company sells software that helps engineers search for materials with desired traits, then test those candidates in computer simulations before chemists make them in a lab. CuspAI calls the approach inverse design: start with a target, then work backward toward molecules or structures that may meet it.

The round has not closed, according to the FT. Investors have signed term sheets, and the reported group includes Kleiner Perkins, Bezos Expeditions, Temasek and New Enterprise Associates. Bezos also runs Prometheus, a separate industrial AI company that targets engineering and manufacturing work. The CuspAI investment appears personal, the FT reported.

The funding fits a broader pattern in AI for science. Investors have spent the past three years moving from chat interfaces toward systems that change physical work: drug design, robotics, chip design and materials discovery. Google DeepMind pushed that story into public view with GNoME, a materials model that proposed 2.2 million crystal structures in 2023. CuspAI now wants to turn that research pattern into a paid industrial platform.

CuspAI has adoption signals that many AI science startups lack. In September, the company said it closed a $100 million-plus Series A, with NEA and Temasek co-leading the round. It named Hyundai, Kemira and Meta as partners, and it said customers use its platform across automotive, water treatment and carbon capture work. The company also lists Geoffrey Hinton, Yann LeCun, Kristin Persson and former ASML President and CTO Martin van den Brink among its advisers.

Kemira gives the clearest public case study. In a May 21 announcement, Kemira said its team worked with CuspAI to design materials that target PFAS removal from drinking and process water. Kemira said the project searched about 300 trillion possible structures, produced more than 5,000 designs with property data and narrowed the list to about 20 candidates for more testing. The company said the discovery phase took six months.

That example shows why investors care. Materials discovery can block progress in chips, batteries, clean water and energy systems. A company may know the property it wants, such as higher heat resistance or selective adsorption of a pollutant, yet spend years testing compounds. If CuspAI helps customers cut that search window, customers can ship better filters, coatings or device components sooner.

The counterargument starts in the lab. Software can propose candidates at scale, but chemists still have to make the material, test it under industrial conditions and prove that manufacturers can produce it at a sane cost. A material that looks strong in simulation can fail during synthesis, degrade in water or require ingredients that break the business case. Investors fund the search engine; customers pay for the material that survives contact with equipment, regulation and procurement.

Developers should watch the infrastructure question. AI materials work needs curated scientific data, simulation engines, high-performance computing and feedback from experiments. A generic language model cannot solve that alone. CuspAI talks about molecular simulation as part of its stack, and its April post on kUPS points toward a company building domain tools rather than wrapping a chat model around chemistry papers.

Open science advocates will also press CuspAI on access. Google DeepMind released GNoME predictions to researchers through the Materials Project. CuspAI, as a venture-backed company, has stronger incentives to keep models, data and customer discoveries closed. That tension will shape how researchers judge the company. Industrial customers may accept secrecy if CuspAI delivers materials; academic users may prefer open databases and reproducible workflows.

Bezos' role adds attention, but customers will decide the outcome. The reported round gives CuspAI capital, status and hiring power. Kemira's PFAS work gives it a more useful benchmark: a narrow industrial problem, a defined design space and candidates moving into tests. If CuspAI can repeat that pattern across semiconductors and energy, AI materials discovery will look less like a research demo and more like a procurement line item.