Orbital Industries Raises $50M to Plug a Cooling-Fluid Gap

Orbital Industries, a London startup founded by Google DeepMind alumni, has raised $50 million in a Series B round led by venture firm Plural, with NVIDIA’s investment arm and Radical Ventures returning. The money is aimed at one unglamorous problem: stopping the next wave of AI chips from cooking themselves, using a coolant the company designed with artificial intelligence rather than a chemistry lab’s decade of trial and error.

Behind the round sits a supply shock most data-centre operators saw coming. In December 2022, manufacturing giant 3M said it would stop making the fluorochemical fluids that two-phase immersion cooling was built on, and the last orders shipped through 2025. That left a hole in a market growing as fast as AI compute itself, and Orbital Industries is one of the firms sprinting to fill it.

3M’s Exit Left a Coolant-Sized Hole

For a decade, the liquids that made immersion cooling work came mostly from one source. 3M’s Novec and Fluorinert lines, including Novec 7100, Novec 649 and Fluorinert FC-72, were the dielectric fluids that let servers sit fully submerged without shorting out. They are also PFAS (per- and polyfluoroalkyl substances, the “forever chemicals” that do not break down in the environment).

When 3M announced its plan to exit PFAS manufacturing by the end of 2025, it was walking away from roughly $1.3 billion in annual net sales. The last day to place a new Novec order was the spring of 2025. The lines went quiet by year end.

That timing collided with a boom. The global immersion cooling fluids market size and forecast from research firm Mordor Intelligence puts the category at $2.51 billion in 2025, rising toward $4.04 billion by 2030. The demand is there. The dominant supplier is not.

What Orbital Industries Is Selling

The company rebranded from Orbital Materials alongside the raise, a tell about where it thinks the value sits. It does not license software. It sells hardware, built on top of an in-house simulation engine, through a commercial arm called Orbital IT. Three product lines carry the strategy.

The PFAS-Free Coolant

Orbital’s first product is a dielectric cooling fluid paired with a refrigeration system, designed for high-density GPU (graphics processing unit) racks and built to be PFAS-free. Chief executive Jonathan Godwin, who spent five years on AI for science at DeepMind, has said developing a new coolant the old way would take “10 years and $100 million.” Orbital says it synthesized an entire molecular family in months. The fluid is in qualification with chip providers and is slated to ship with next-generation GPUs in 2027.

The Orb Simulation Engine

Underneath the products sits Orb, an AI model that simulates the quantum-mechanical behaviour of atoms. The company says its Orb-v3 model for atomistic simulation at scale can handle roughly 100,000 atoms on a single GPU and runs about ten times faster than comparable models from Meta and Microsoft. A free version trained on public data is open-sourced under a permissive licence on the Orb forcefield models repository, while a proprietary version stays internal for product work.

Modular Data-Centre Units

The third line is modular data-centre infrastructure, manufactured off-site and delivered as ready-to-deploy units. Orbital claims this can cut deployment of new compute capacity to about six months, against the roughly three years a conventional build takes.

The Heat Wall Behind the Bet

None of this matters without the thermal problem driving it. As AI training clusters pack more power into each rack, the heat they throw off is outrunning what air can carry away. That is why operators are shifting toward liquid and immersion architectures in the first place.

The pressure is not theoretical. Hyperscalers are already reaching past the grid, with some, including the partners behind a push to put AI data centres in orbit, exploring exotic ways to add capacity and shed heat. On Earth, the bottleneck shows up as power draw, cooling cost and how long it takes to stand up a new site.

The fluid market tracks that strain closely. The numbers below come from Mordor Intelligence’s forecast.

• $2.51 billion in immersion cooling fluid sales in 2025
• $4.04 billion projected by 2030
• $551.9 million of that market sitting in Europe in 2025

Mordor also notes that synthetic hydrocarbon oils have become the preferred fluid type, helped by material compatibility and exactly the regulatory pressure on PFAS that pushed 3M out. That is the lane Orbital is trying to leapfrog.

The Regulatory Clock in Brussels and Washington

The case for a PFAS-free coolant is not just commercial. It is being written into law on two continents, and the timeline is tightening rather than easing.

1. December 2022: 3M says it will exit PFAS manufacturing by the end of 2025, citing accelerating regulation and shifting stakeholder expectations.
2. August 2025: the European Chemicals Agency (ECHA, the EU’s chemicals regulator) publishes an updated EU-wide PFAS restriction proposal under REACH, expanding derogated uses from 26 to 74, some with phase-outs of up to 13.5 years.
3. First half of 2026: a public consultation on the draft opinion, with ECHA set to consult on the PFAS draft opinion in spring 2026 before committees finish their scientific assessment by year end.

In the United States, the Environmental Protection Agency has been moving on PFAS in drinking water and reporting rules, adding to the compliance case for anyone designing new industrial fluids today. A coolant built clean from the molecule up sidesteps a regulatory question its incumbent rivals still have to answer.

Why NVIDIA’s Fund and Plural Wrote the Cheque

Orbital had raised more than $20 million before this round, from backers including Radical Ventures, NVIDIA and Toyota, and signed a partnership with Amazon Web Services in December 2024. The Series B brought in Plural as lead, alongside returning investors NVentures, Radical Ventures, Compound and Fly Ventures. The team is about 50 people, split between London and San Francisco, with a chief technology officer, James Gin-Pollock, who previously sold a company to Shutterstock.

The thesis the investors are buying is bigger than coolant. Godwin frames it as proof that frontier AI lets a tiny team do what once needed an industrial research division.

Frontier AI gives us PhD-level expertise across every discipline, meaning small, agile teams can move from materials discovery to commercial hardware in a way that simply wasn’t possible before, so what used to take a decade, we can now do in months.

Godwin has said the ambition is to build the largest industrial conglomerate in Europe, an AI-native answer to the old chemicals giants. That fits a wider rotation of capital toward hard-tech founders, the same current running through deals like a European fund targeting energy and industrial AI startups. The pitch is that software-speed iteration can finally reach physical products.

For NVIDIA’s fund, there is a narrower logic too. Anything that cools its chips faster and cheaper protects the demand for the chips themselves.

The Part That Isn’t Proven Yet

The gap is real, but so is the distance Orbital still has to cover. The coolant is in qualification, not in production, and chipmakers are conservative about anything touching a rack full of expensive silicon. A 2027 ship date is more than a year out, which leaves room for slips and for rivals to move.

Those rivals are not standing still. Synthetic hydrocarbon oils already dominate new immersion designs, and replacement fluorochemical fluids such as Galden remain on the market for operators who just want a like-for-like swap. Orbital’s harder challenge is the one every materials startup hits: scaling a molecule from a lab batch to industrial volume without losing the economics that made it attractive.

If Orbital’s fluid clears chipmaker qualification before the 2027 GPU wave, a 50-person team will have beaten the chemicals majors to a market they chose to leave. If it slips, operators keep buying synthetic oil and stopgap fluorochemicals, and the AI-industrial thesis waits a little longer for its first hardware proof.