The Hidden Bedrock of NHS AI

Outside Glasgow, in the flat industrial land of North Lanarkshire, concrete is going down for server halls that will eventually hum with graphics processors leased to CoreWeave, the American cloud computing firm that has become one of the defining infrastructure suppliers of the generative AI boom. DataVita, the Scottish operator building the capacity, this week confirmed a £25m loan from digital bank OakNorth to push the work forward. Buried in the announcement was a detail that matters rather more to health policy than the financing round itself. DataVita lists NHS bodies among its public sector customers.

That relationship is not incidental. Since 2019 DataVita has sat on the Scottish Government's Cloud Services Framework, the procurement route through which NHS Scotland, local authorities and other public bodies access cloud and colocation services. It is a formal, governed arrangement rather than a passing contract, and it means a meaningful slice of NHS Scotland's digital workload already runs through infrastructure that DataVita owns, finances and now expands primarily to serve a private American hyperscale client.

The company's own framing of the deal is instructive. William Hill, chief executive of parent group HFD, described the expansion as reinforcing Scotland's position as a destination for AI and cloud infrastructure. OakNorth's Fraser McPhail went further, arguing that data centre investment is spreading beyond traditional hubs as operators chase the power, land and renewable energy that AI workloads demand. Neither executive mentioned the NHS. There was no reason for them to. But the pattern they describe, commercially financed regional data centres built chiefly to satisfy hyperscale AI demand and then offering spare capacity to public bodies almost as an afterthought, deserves more scrutiny than it currently gets in health policy circles.

The NHS's digital ambitions increasingly depend on exactly this kind of compute. Ambient voice technology in consultations, imaging AI tools moving through NICE and MHRA approval, and the analytics layer underpinning the Federated Data Platform all require processing capacity that has to sit somewhere physical. Much of the debate around that dependency to date has focused on software governance, who has access to patient data and under what contractual terms, as the long-running arguments over Palantir's FDP contract show. What gets far less attention is the layer beneath the software, the buildings, racks and power contracts that make any of it run at all, and who actually owns and prioritises that capacity when demand outstrips supply.

DataVita is not Palantir, and it would be wrong to treat a regional colocation provider as a contractual risk on the same scale as a strategic national data platform vendor. The dependency here is structural rather than contractual. NHS bodies are tenants inside a market where private capital is chasing AI-driven demand from clients with far deeper pockets than any integrated care board. If a facility is expanded principally to host CoreWeave's workloads, and NHS or local authority use sits alongside that as one customer segment among several, the question of whose priorities shape future capacity allocation is not an abstract one. It becomes sharper as government pushes NHS trusts toward AI-enabled discharge planning, diagnostic support and administrative automation under the productivity agenda that Wes Streeting's successors at DHSC have inherited.

There is also a financing dimension worth noting. Debt-financed expansion through commercial lenders such as OakNorth means the underlying infrastructure sits as collateral on a private balance sheet, its future shaped by the commercial performance of hyperscale contracts rather than by NHS planning cycles or capital budgets. That is a perfectly ordinary way to build a data centre business. It is a less ordinary foundation on which to build critical national health infrastructure by default, almost without anyone deciding it should happen that way.

None of this argues for panic about a single loan agreement in Lanarkshire. It argues for a more basic question that NHS digital strategy has yet to ask with any rigour, which is how much of the physical capacity that AI adoption in health care will require is actually being built with public health bodies in mind, and how much is being built for American cloud clients with NHS bodies picking up whatever room is left over.