Summary

• Nebius has signed a 22MW, 10-year AI infrastructure agreement with Kao Data at Harlow.
• The deployment will host Nebius AI Cloud and Nebius Token Factory, with high-density infrastructure for AI workloads.
• The deal links UK AI adoption to the physical limits of compute, power, data centre design, and domestic infrastructure capacity.

Kao Data has signed a 22MW, 10-year agreement with European AI cloud company Nebius for a major AI infrastructure deployment at its Harlow data centre campus.

The deployment will host the Nebius AI Cloud platform and Nebius Token Factory, the company’s managed inference service for deploying and optimising open AI models. Kao Data says the agreement will support AI innovation across the UK’s academic and research communities, enterprise users, and the government’s AI Opportunities Action Plan.

The Harlow campus is being presented as a hub for industrial-scale AI research, enterprise workloads, and GPU-accelerated computing. Kao says the site is powered by 100 percent renewable energy, supported by HVO-powered generators, and equipped with direct-to-chip liquid cooling designed to reduce water use for high-density AI systems.

The agreement forms part of a wider UK push by Nebius. The company has also pledged a £1.7bn UK investment, including three Nvidia infrastructure deployments expected to total 65MW when fully ramped up in 2027.

The deal captures the material reality behind AI adoption. Generative AI and frontier-model development are often discussed as software markets, but the limiting factors are increasingly physical: power, cooling, data centre capacity, grid connections, semiconductor availability, and the location of compute relative to customers, researchers, and regulated data.

Local AI infrastructure can reduce latency, support data-residency preferences, and make access to advanced GPU capacity less dependent on overseas regions. It does not automatically create sovereign AI capability, but it gives developers, enterprises, and public-sector bodies more domestic options for training, inference, and model deployment.

Inference is especially important. The first wave of AI infrastructure debate focused heavily on training large models, which is expensive, compute-intensive, and concentrated among a small number of providers. As businesses move from pilots to production, inference capacity becomes critical. Customer-service assistants, coding tools, document workflows, search systems, design processes, and operational AI services all require reliable compute each time they run.

That changes the economics of data centres. AI workloads require dense racks, advanced cooling, high power availability, and specialised network design. Traditional enterprise data centres were not built for the same thermal and electrical profile. Operators that can provide AI-ready capacity are becoming part of the strategic technology stack rather than neutral landlords.

The environmental pressure will not disappear because a site uses renewable energy or more efficient cooling. AI data centres consume significant electricity and can place pressure on local grids. Kao’s sustainability claims address part of the concern, but the wider sector will still face scrutiny over energy demand, planning, water use, and whether public infrastructure can keep pace with private compute growth.

The UK government has made AI adoption a central economic objective, yet adoption depends on infrastructure that is often slow to build. Planning permissions, grid capacity, local opposition, skills, and supply chains can all delay deployment. The Nebius-Kao agreement shows AI strategy moving from policy documents into long-term data centre contracts.

Market structure is also shifting. AI cloud capacity is becoming a competitive layer between hyperscale cloud, specialist GPU cloud providers, data centre operators, chip suppliers, and enterprise buyers. Nebius is one of several providers trying to offer alternatives to the largest cloud platforms, while still depending on Nvidia infrastructure and scarce high-density facilities.

The Harlow deployment will not by itself solve the UK’s compute constraints. It does show the direction of travel. AI adoption is becoming an infrastructure build-out, and the companies that control power-ready, AI-ready capacity will shape how quickly businesses, researchers, and public services can turn models into usable systems.