CuspAI, a Cambridge, UK-based startup applying artificial intelligence to the search for new physical materials, has closed a $450 million Series B round led by Kleiner Perkins and NEA, with Jeff Bezos’s family office Bezos Expeditions, Glade Brook Capital Partners, Lux Capital, AMD Ventures, the UK government-backed Sovereign AI Venture Fund, and investor John Doerr also participating. The round values the company at $2.6 billion — up from roughly $520 million ten months earlier, in September 2025 — one of the sharper valuation climbs of any AI-for-science startup this year.
The core problem CuspAI is attacking is a genuinely old one: discovering a new material — an alloy, a catalyst, a semiconductor compound — has traditionally meant synthesizing and testing thousands of candidates in a lab, a process that can take years and consume enormous amounts of scientist time for what is often a single usable result. CuspAI’s platform, called MIRA, tries to collapse that search space before anything touches a lab bench. It combines Meta’s open-source UMA models, which simulate materials at the atomic level, with CuspAI’s own proprietary toolkit, kUPS, which handles molecular-level simulation and, notably, automates some of the coding work and GPU-efficiency tuning that AI-for-science labs usually still have to do by hand. The company’s own framing is that MIRA can complete material-discovery work that “usually takes years” in about six months, then hand off only the most promising few candidates for actual physical testing.
What separates CuspAI from a typical AI-for-science pitch is the consortium it has built around the technology. The company’s “AI Materials Foundry” now counts more than 45 member organizations — among them Nvidia, Meta, Samsung, Hyundai Motor Group, and chip-equipment maker Lam Research — pooling computing resources, laboratory access, and scientific expertise onto CuspAI’s platform. Rather than selling software licenses one company at a time, CuspAI is positioning itself as shared infrastructure: a member submits a request for a compound with certain target properties, MIRA proposes candidates, and the platform can match a viable candidate to a manufacturing partner with the actual production capacity to make it. That is a meaningfully different business model from most AI-for-science startups, which tend to either license simulation software directly or run an in-house discovery pipeline and sell the resulting materials themselves.
The most immediate commercial target is semiconductors, which the company says will absorb 80% of its research bandwidth this year. Part of that work is aimed squarely at a supply-chain vulnerability the chip industry has quietly worried about for years: heavy reliance on a short list of rare, geographically concentrated metals — ruthenium and iridium among them — in advanced chipmaking. If CuspAI’s platform can identify viable substitutes or reduced-usage processes for those metals, it is solving a problem that sits several rungs below the AI headlines but matters enormously to companies like Samsung and Lam Research, both of which joined the consortium rather than simply watching from outside. That is also, functionally, the pitch to investors: this is not a bet on one flashy discovery, it is a bet that owning the infrastructure layer dozens of manufacturers rely on for candidate discovery is a more durable business than owning any single materials breakthrough.
AI-driven materials discovery has drawn increasing venture interest over the past two years, as models trained on scientific and simulation data start folding years of laboratory trial-and-error into computation that runs in months instead. CuspAI’s bet is that the winner in that category won’t be whichever lab finds the single best new material first, but whichever platform enough competing manufacturers trust to run their searches for them — which is precisely why getting Nvidia, Samsung, Meta, and Hyundai to co-invest resources into the same consortium, rather than build competing internal tools, is the more interesting story here than any individual funding number.
What This Means for Philippine Founders
CuspAI’s own market — advanced semiconductor materials — sits upstream of anything a Philippine startup would build directly, but the Philippines has real, direct exposure to the industry CuspAI is trying to serve. Semiconductor assembly, test, and packaging is one of the country’s largest manufacturing exports, and any shift in how global chipmakers source materials or qualify new compounds eventually filters down into the assembly and testing specifications Philippine-based operations have to meet. The more durable lesson for Philippine founders, though, is about business model rather than sector: CuspAI didn’t just build a better tool, it built a consortium that made adopting the tool a shared, lower-risk decision for competitors who would never normally collaborate. A Philippine startup selling into a fragmented, relationship-driven local industry — logistics, agriculture, retail — could learn from that structure directly: getting several competing operators to co-invest in or co-adopt shared infrastructure is often a stronger moat than convincing them one at a time, even when the initial sales cycle to get there is slower to close.
Share this article