CuspAI raises $450M to build an AI-powered materials discovery platform targeting the semiconductor industry

CuspAI raises $450M for AI-driven materials discovery

CuspAI, a Cambridge-based startup building a generative AI platform for materials discovery, has raised $450M. Kleiner Perkins, NEA, and Jeff Bezos are among the investors. The raise follows a $100M Series A the prior year — and Edwards says he's already been preempted on fundraises before, so he wouldn't rule out another round coming sooner than expected.

What CuspAI builds

The platform functions as a search engine for materials, using generative AI to discover entirely new compounds. Edwards frames semiconductors, clean energy, and climate as the core applications — all of which, at their base, are materials science problems. A single semiconductor chip uses more than 60 elements and requires silicon at 99.99% purity. Better materials mean more performant, more energy-efficient chips.

The company started as a horizontal platform, but semiconductor customers pulled it toward that vertical. With AI hardware demand pushing performance requirements higher, chipmakers are actively looking for new material options — and that market pull, Edwards says, determined CuspAI's near-term focus.

“At CUSP, we're building essentially a search engine for materials — a generative AI platform where you can discover entirely new materials. The semiconductor industry has pulled us by all four limbs in that direction. We've proven that the biotech-type royalty model works in material science — every customer contract has royalties baked into it.”

Capital intensity

CuspAI's cost structure has three components: AI and machine learning talent, compute (the company has a partnership with Nvidia), and experimental data generation through a global network of physical labs. Edwards is explicit that CuspAI is not training frontier-scale LLMs — the models are domain-specific and narrower. The lab network, rather than in-house facilities, is the chosen approach for now, partly because those labs carry domain expertise that CuspAI can't easily replicate internally.

Business model

The most distinctive aspect of the company is its revenue model. CuspAI charges customers upfront program fees and bakes royalties into contracts — so when a discovered material enters production, CuspAI shares in the commercial upside. Edwards says he spent the company's first two years proving this model works despite early investor skepticism that large corporates would never agree to royalty structures. They did.

Edwards describes his role as analogous to a VC: running a portfolio of materials bets, some near-term and high commercial potential, others longer-horizon. He acknowledges the royalty model produces lumpy revenue and sees merit in the SaaS subscription approach taken by companies like Chai Discovery, though CuspAI is committed to the royalty structure as the primary value-capture mechanism.

Commercial discipline

Edwards previously founded a quantum computing company, which he describes as an extreme example of pre-revenue deep tech. CuspAI has taken the opposite approach, pursuing paying customers from early on. Revenue is already coming in, and the next fundraise will likely be underwritten against commercial traction rather than scientific milestones alone — though Edwards no longer tries to predict when the next round will happen.