London startup by ex-DeepMind researcher nets $50M for AI-designed non-toxic next-gen GPU cooling
Orbital Industries has secured $50 million in Series B funding to commercialise an AI-designed cooling fluid that tackles the growing heat crisis in AI data centres, replacing toxic chemicals with a PFAS-free alternative while promising to slash development timelines from a decade to months.
The GPUs powering the latest generation of artificial intelligence models generate so much heat that conventional cooling systems are becoming a bottleneck. Water-based cooling is approaching its physical limits, and the dielectric chemicals historically used in immersion cooling are facing tightening environmental regulation in both Europe and the United States. 3M, one of the largest producers of these chemicals, exited its Novec PFAS-based coolant business entirely in 2024, creating a market gap that Orbital is now targeting. The EU is considering a broad restriction proposal that could ban around 10,000 PFAS compounds, while the US Environmental Protection Agency introduced legally enforceable limits for six PFAS in drinking water in April 2024. Against this backdrop, AI workloads already demand power densities exceeding 30 kW per rack, sometimes reaching 100 kW, straining existing infrastructure.
The data centre infrastructure market stands at $344 billion today and is projected to surpass $2 trillion by 2032, according to industry forecasts. Broader estimates for the construction segment alone put the figure at $281 billion in 2025, expected to grow to $431 billion by 2031. Yet the speed at which GPU density is increasing risks outrunning the ability of any single company to design and deploy new physical infrastructure fast enough — a challenge Orbital Industries aims to meet with a vertically integrated approach that combines AI-driven material discovery with hardware manufacturing.
The company, founded in 2022 and headquartered in London with a San Francisco office, employs 50 people and has raised a total of $71 million to date. Its three co-founders bring complementary expertise: chief executive Jonathan Godwin spent nearly a decade at DeepMind, joining around the time of the AlphaFold project, with five of those years focused on AI for science, engineering and advanced materials design; chief technology officer James Gin-Pollock previously sold a company to Shutterstock; and chief operating officer Daniel Miodovnik has a background spanning finance, government AI and advisory work for the United Nations.
Inside Orb: the AI engine accelerating material discovery
At the centre of Orbital’s platform is Orb, a proprietary AI engine that simulates the quantum mechanical behaviour of atoms. Independent benchmarks published in npj Computational Materials in 2026 show that Orb is the only model capable of simulating 100,000 atoms on a single GPU — every competitor crashes at that scale — and it runs ten times faster than its nearest alternative, outperforming models from Microsoft, Meta and leading academic labs. What once took a week of computation can now be done in minutes.
Further benchmarks indicate Orb runs three to six times faster than existing universal interatomic potentials, achieving a 31% reduction in error on the Matbench Discovery benchmark. These capabilities allow Orbital to compress a material-development timeline that traditionally takes close to a decade into a matter of months. The AI engine is also available to Amazon Web Services customers via SageMaker JumpStart and the AWS Marketplace, as part of a multi-year partnership that covers data centre decarbonisation, cooling and water-utilisation technologies.
Orbital’s first commercial product is a dielectric cooling fluid engineered specifically for next-generation GPUs. Unlike most competing technologies, it contains no PFAS forever chemicals, enabling compliance with tightening regulations across the EU and US. The company has also developed a modular data centre system that can be manufactured off-site and deployed in as little as six months, compared with traditional build timelines of up to three years. Orbital describes its approach as an “AI industrial” model, integrating materials discovery, engineering and manufacturing into a single system, and it is more vertically integrated than many AI materials startups, aiming to discover, manufacture and sell products directly rather than merely licensing intellectual property.
Jonathan Godwin, Orbital’s CEO, said: “When people imagine a better future, they think about physical things: technologies that give them more freedom, more time, more life. AI will get us there faster. 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.”
Investors and strategic backing
The Series B round was led by Plural, the early-stage fund co-founded by Wise co-founder Taavet Hinrikus and serial entrepreneur Ian Hogarth, and based in Tallinn and London. Plural previously backed the European defence AI unicorn Helsing and the Munich-based nuclear fusion startup Proxima Fusion. Ian Hogarth, a partner at Plural, said: “AI progress is now constrained by the physical world: by energy, heat and infrastructure. Orbital Industries is tackling those constraints directly, from breakthroughs like its AI-designed cooling fluid, which enables the next generation of GPUs. The ability to discover and deploy these technologies faster than traditional industry will define the next phase of AI.”
New investor NVentures, NVIDIA’s venture capital arm, joined the round, signalling a strategic interest in ensuring that cooling infrastructure keeps pace with the company’s GPU roadmap. Returning investors include Radical Ventures, Compound and Fly Ventures.
Competitive landscape and differentiation
The liquid cooling market for data centres is consolidating rapidly. LiquidStack, previously one of the most prominent independent immersion cooling providers, was acquired by Trane Technologies in February 2026, removing it as an independent competitor. Submer, the Barcelona-based immersion cooling specialist that has raised $55.5 million and counts CERN among its customers, focuses on tank-based immersion systems and direct-to-chip cooling rather than fluid chemistry and modular architecture. Neither company builds the underlying cooling materials using AI. Orbital’s differentiation sits one level deeper: it designs the materials themselves using atomic simulation, then commercialises the hardware built around them.
The company’s long-term ambition extends well beyond data centres. Orbital plans to apply the same AI-to-hardware model across semiconductors, critical minerals, aerospace and energy. Previously known as Orbital Materials, the company initially focused on AI-discovered carbon-capture and sustainable-aviation-fuel chemistry before pivoting to data centre infrastructure in response to strong commercial demand.
The unanswered question for Orbital — and for the broader AI infrastructure market — is whether the pace of GPU density growth will continue to outrun the ability of any single company, however AI-accelerated, to design and deploy new physical infrastructure fast enough to keep the next generation of compute online. Orbital’s answer so far has been to integrate materials discovery, engineering and manufacturing into a single AI-driven system, a model the company describes as “AI industrial” and which it believes can compress decade-long development cycles into months.
