Orbital Raises $50M with Nvidia to Design AI Materials 2026

The next breakthrough material inside your data center may not be discovered by a chemist in a lab. It may be designed by an AI model that simulates the quantum behavior of a hundred thousand atoms on a single graphics card. Orbital Industries just raised $50 million to prove that thesis, and the most striking part is who wrote a check: Nvidia, the company whose own GPUs are running hot enough to need exactly the kind of coolant Orbital is using AI to invent.

What Actually Happened

On May 28, 2026, Orbital Industries announced a $50 million Series B led by venture firm Plural, with participation from Nventures (Nvidia's venture arm), Radical Ventures, Compound, and Fly Ventures. The London and San Francisco company, which recently rebranded from Orbital Materials and employs roughly 50 people, is building AI that designs advanced materials from first principles rather than discovering them through years of trial-and-error lab work. The round is small by the standards of 2026 mega-deals, but the strategic backing matters more than the dollar figure.

At the center of the company is a model called Orb, which predicts and simulates the quantum-mechanical behavior of atoms. Orbital claims Orb is the only model that can simulate 100,000 atoms on a single GPU and runs roughly 10 times faster than alternatives, outperforming comparable offerings from Meta and Microsoft. That speed is the whole game in computational materials science, because the bottleneck has always been the staggering cost of accurately modeling how atoms interact. Compress that cost and you can explore a vastly larger space of candidate materials before ever touching a beaker.

Orbital is not selling the model. The company has chosen vertical integration, manufacturing and selling finished materials and hardware directly rather than licensing its intellectual property to chemical giants like BASF or PPG. Its first flagship product is an AI-discovered liquid coolant for GPU server racks that avoids the controversial PFAS "forever chemicals," developed in months for a fraction of the cost rather than the decade and roughly $100 million a conventional coolant program can consume. A second line, branded Orbital IT, offers modular data-center systems deployable in six months versus three years for conventional builds.

Why This Matters More Than People Think

Materials science has been the quiet bottleneck behind nearly every hardware advance of the past century, from semiconductors to batteries to jet engines, and it has resisted acceleration because the underlying physics is expensive to compute. Discovering a genuinely new material has historically taken 10 to 20 years and hundreds of millions of dollars, with most of that time spent on physical synthesis and testing of candidates that mostly fail. If Orbital can compress the discovery loop from decades to months by simulating outcomes before synthesis, it is not improving one industry, it is removing a constraint that sits underneath all of them.

The choice to start with data-center coolant is a tell about where the money is. The AI build-out has created an enormous thermal problem: Nvidia's latest racks draw so much power that air cooling is no longer viable, and liquid cooling has become mandatory at the frontier. The incumbent coolants rely on PFAS chemistry now facing regulatory pressure on both sides of the Atlantic. A PFAS-free coolant designed specifically for next-generation GPUs is therefore not a science-fair project, it is a product with a captive, desperate, fast-growing customer base that includes the very hyperscalers spending hundreds of billions on AI infrastructure.

Nvidia's investment closes the loop in a way that is easy to underrate. Nvidia builds the GPUs that create the heat, sells the systems that need cooling, and now holds equity in the company using AI to design the coolant. That vertical alignment gives Orbital something most deep-tech startups lack: a credible path to a first customer who is also strategically motivated to see it succeed. When the company that defines the thermal envelope of the entire AI industry is on your cap table, the gap between a promising molecule and a qualified, deployed product narrows dramatically.

The timing also matters because regulation is turning the coolant problem from a preference into a mandate. PFAS chemicals, the fluorinated compounds used in many existing immersion and cold-plate coolants, are facing restriction and outright bans across the European Union and several US states, with compliance deadlines arriving over the next few years. That means hyperscalers are not merely looking for a better coolant, they are being forced to find a PFAS-free one on a regulatory clock. A startup that can design a compliant replacement faster than the deadline arrives is selling into a market with a legal gun to its head, which is the rare kind of demand that does not need to be created through marketing because the law manufactures it.

The Competitive Landscape

Orbital is part of a fast-forming cohort betting that AI will industrialize materials discovery. CuspAI raised a $100 million Series A in September 2025 to design materials for carbon capture and beyond. Periodic Labs pulled in a $300 million seed round the same month, one of the largest seeds in the sector, to build AI scientists for physical materials. And Prometheus, backed by Jeff Bezos, has raised a reported $6.2 billion for AI-optimized manufacturing. Against that field, Orbital's $50 million looks modest, which is exactly why its strategy of vertical integration and a wedge product is its differentiator rather than raw capital.

The founding team is built for this fight. CEO Jonathan Godwin is a former Google DeepMind researcher who specialized in AI for science, the discipline that produced AlphaFold, the model that cracked protein structure prediction and rewired drug discovery. CTO James Gin-Pollock is a repeat founder who previously sold a company to Shutterstock, and COO Daniel Miodovnik rounds out the team on operations. The AlphaFold lineage is the relevant historical parallel: it showed that a sufficiently good simulation model can collapse a problem that consumed careers into something that runs in an afternoon, and Orbital is explicitly trying to be the AlphaFold of materials.

The bear case, however, is that materials are harder to commercialize than software, and the graveyard of computational-materials startups is real. A model that designs a promising molecule still has to clear synthesis, scale-up, qualification, regulatory approval, and manufacturing, each a slow and capital-hungry stage where AI provides little leverage. Critics argue that Orbital's vertical-integration strategy, while it captures more value, also saddles a 50-person startup with the brutal economics of physical manufacturing, an arena where chemical giants like BASF have a century of process expertise. The risk is that the AI design step, however brilliant, is the easy 10% of a 100% problem.

Hidden Insight: The First AI-Designed Molecule on the Market Is the Real Prize

Orbital has set a target that, if hit, would be a genuine milestone: commercializing its AI-designed cooling fluid alongside the next generation of GPUs in 2027, which the company says would mark the first time an AI-designed molecule has reached the commercial market in any industry. That claim is worth taking seriously, because it would convert the entire "AI for science" narrative from a research curiosity into a shipping product with revenue attached. There is a vast difference between an AI that proposes a molecule in a paper and an AI whose molecule is circulating through racks in a live data center, and crossing that line changes how the whole field is valued.

The deeper insight is about what kind of moat this creates. In software, AI capabilities are increasingly commoditized: models leapfrog each other monthly and yesterday's breakthrough is today's open-weight baseline. In materials, the moat is different and more durable, because a qualified, patented, manufactured product protected by regulatory approval and customer integration cannot be cloned by a competitor who fine-tunes a better model next quarter. Orbital is betting that the defensible value is not the model but the physical thing the model produces, which is a fundamentally different theory of competitive advantage than most AI startups hold.

That theory is why vertical integration, despite its costs, may be the correct call rather than a reckless one. If Orbital licensed Orb to BASF, it would capture a thin slice of value and hand the durable, defensible part, the manufactured product and customer relationship, to an incumbent. By manufacturing itself, Orbital keeps the part of the value chain that compounds. The trade is real and dangerous, more capital intensity and operational risk in exchange for a moat that survives model commoditization, but in a world where AI capability is racing toward parity, owning the physical output may be the only thing that stays scarce.

It is worth separating Orbital from the broader hype around AI scientists, because the distinction is where the value lives. Many AI-for-science efforts stop at prediction: they output a ranked list of candidate molecules and leave the hard physical work to someone else. Orbital is structurally betting that prediction alone is a commodity and that the money is in carrying a candidate all the way through to a shipped, qualified product. That is a far harder business to build, but it is also far harder to copy, and in a sector where a dozen well-funded teams are all training similar simulation models, the willingness to own the unglamorous downstream steps may be the cleanest form of differentiation available.

There is a second-order implication for the AI infrastructure boom itself. The market has treated the build-out as a story about chips, power, and data centers, the visible, capital-heavy layers. Orbital points at a less visible layer: the materials science underneath the hardware, the coolants, the substrates, the thermal interfaces that determine whether a rack can actually run at the densities the chips demand. As GPUs push past the limits of conventional cooling, the companies that can design new materials faster than the hardware roadmap advances become a hidden chokepoint. Whoever solves the thermal and materials wall captures leverage over an industry that has so far been focused entirely on compute.

What to Watch Next

In the next 30 to 90 days, the signal to watch is whether Orbital announces a named pilot customer for its coolant, ideally a hyperscaler or a server OEM qualifying the fluid in real hardware. A funding round is a promise; a qualification program is proof. Given Nvidia's stake, watch specifically for any indication that Orbital's coolant is being tested against Nvidia's next-generation rack designs, which would convert the strategic-investor relationship into a concrete commercial validation and de-risk the 2027 commercialization target.

Over the next 180 days, track whether Orbital can demonstrate that Orb's simulation advantage translates into a second product beyond coolant. A single AI-designed material could be luck or a narrow win; a pipeline of candidates across coolants, substrates, and other data-center materials would prove the model is a general discovery engine rather than a one-hit tool. Also watch the competitive responses from CuspAI and Periodic Labs, and whether Meta or Microsoft, whose materials models Orbital claims to beat, respond by open-sourcing competitive simulation tools that erode Orbital's speed advantage.

On the longer horizon, the defining question is whether 2027 actually delivers the first AI-designed molecule on the commercial market. If Orbital ships, it sets a precedent that reprices the entire AI-for-science sector and pulls forward capital into materials startups across batteries, catalysts, and pharmaceuticals. If the 2027 target slips, as deep-tech timelines often do, it will validate the skeptics who argue that the synthesis-and-scale-up wall is where AI materials dreams go to die. The mental model for readers is to watch the gap between simulation and shipping product: that gap, not the quality of the model, is where this thesis will be won or lost.

For investors and operators watching the AI infrastructure trade, Orbital is a reminder to look below the chips. The visible layers of the build-out, the GPUs, the power deals, the data centers, are crowded and richly valued, but the materials layer underneath them is thin, slow-moving, and ripe for exactly the kind of acceleration AI now offers. The next decade of hardware progress may hinge less on the next process node and more on whether someone can design the coolants, substrates, and thermal materials fast enough to keep pace, and the teams that win that race will be invisible to anyone watching only the chip roadmap.

The durable moat in AI is not the model anymore, it is the physical thing the model makes. Orbital is betting that the first AI-designed molecule to ship will be worth more than any model that designed it.

---