CuspAI Raises $450M to Speed Up Materials Discovery

The Search Space Problem, and a $450M Answer

Cambridge-based CuspAI announced a $450 million Series B on July 20, 2026, led by Kleiner Perkins and NEA, at a $2.6 billion valuation. The company builds models that simulate how a candidate material will behave before anyone makes it — the point being to narrow an effectively infinite search space down to a shortlist worth putting in a furnace. Alongside the raise it launched the AI Materials Foundry, a coalition of more than 45 organizations pooling compute, lab access, and expertise.

• $450M Series B led by Kleiner Perkins and NEA at a $2.6B valuation, up from $520M last September
• Investors include Bezos Expeditions, Lux Capital, AMD Ventures, Glade Brook, StepStone, the UK Sovereign AI Venture Fund, Invest-NL, and John Doerr
• The AI Materials Foundry spans 45+ organizations including NVIDIA, Meta, Samsung, Hyundai Motor Group, and Lam Research
• Semiconductors will take about 80% of the company's research bandwidth this year

Why Simulate a Material Before Making One?

Materials science has always been bounded by how fast you can run physical experiments. Synthesizing a candidate compound, characterizing it, and testing it under real conditions takes weeks, and the space of plausible candidates is astronomically larger than any lab schedule. CuspAI's approach is to use learned models of material behavior as a filter: predict performance across a very wide candidate set, then spend scarce lab time only on the compounds the models rank highly.

That is the same shape as several other AI-for-science efforts we have covered — MIT's GIFT system turning 2D designs into CAD models uses a comparable "narrow it computationally, verify it physically" loop. The payoff is not a replacement for experiment; it is a much better guess about which experiment to run first.

What Is the AI Materials Foundry?

It is a shared-resource coalition rather than a product. Members contribute compute, laboratory capacity, and domain expertise, and the roster is notably cross-industry: NVIDIA and Lam Research from the semiconductor side, Meta and Samsung from the platform and device side, Hyundai Motor Group from automotive. Consortium structures like this tend to work best where no single participant can afford the full stack of simulation compute plus wet-lab validation, which describes materials research fairly precisely.

The Semiconductor Focus

CuspAI says semiconductors will consume roughly 80% of its research bandwidth this year, and one named target is removing supply-constrained rare metals such as ruthenium and iridium from chipmaking processes. That is a concrete, checkable goal rather than a general promise — and it has an obvious industrial logic behind it. Materials that depend on scarce inputs are a structural constraint on how much of anything can be manufactured, so finding substitutes has a compounding effect downstream.

The funding round itself follows a $100M+ Series A less than a year earlier, and the valuation has moved from $520 million last September to $2.6 billion now. Fast, but consistent with where investor attention has gone across the wider AI field this year.

What to Watch

The interesting metric will not be the raise. It will be how many Foundry-originated candidate materials make it through physical validation and into a real process line, and how long that takes. Simulation-first pipelines earn their keep at exactly that handoff.

Sources: SiliconANGLE — July 20, 2026; Quartz — July 20, 2026; CuspAI — July 20, 2026.