DataVita switches its 500 MW AI data center project to grid power

A major AI project near Glasgow has run into a familiar infrastructure constraint: computing capacity can be announced quickly, but electrical capacity cannot. DataVita has confirmed that its planned 500 MW data center in Lanarkshire will be connected entirely to the UK National Grid. The company previously spoke of more than 1 GW of wind and solar generation and storage that would supposedly supply the site through private direct lines.

The wording has now changed. The shared power system will be the primary source, while the company's own solar and wind facilities may be added later—if they clear commercial agreements, planning and permitting. This is a substantial revision for a project promised £202 million in government guarantees and expected to create around 3,400 jobs.

Where the promise broke down

A data center cannot be powered in the same way as an office with rooftop solar panels. Servers operate around the clock, while wind and solar deliver variable output. If cloud cover reduces generation or the wind dies down, a load of hundreds of megawatts does not simply disappear. The project needs the grid, storage on an enormous scale, or separate dispatchable generation.

Consider a site with constant consumption of 500 MW. Even power plants with 1 GW of installed capacity cannot guarantee that they will provide the required 500 MW at every moment: actual output depends on the weather, technical availability and transmission constraints. A direct connection to wind and solar farms therefore does not, by itself, solve the problem of uninterrupted power.

DataVita says it will purchase electricity under a renewable power purchase agreement. Such an agreement can lock in the origin and price of the energy, but the data center still physically receives electricity from the shared grid. If the grid lacks capacity, the contract does not add cables, substations or available megawatts.

Why this matters to AI product teams

Energy becomes a product constraint long before a model is launched. Available power determines the number of accelerators, the speed at which a cluster can expand, training costs and the ability to meet commitments to customers. In the UK, connecting large facilities can take years, while electricity costs more than in many European countries. An impressive GPU procurement plan therefore means nothing without a confirmed connection point and power delivery date.

Founders and CTOs should distinguish between three different claims: a site can buy “green” energy under contract; new generation is planned nearby; and the facility has already secured a guaranteed physical connection at the required capacity. These points often appear side by side in presentations, even though their risk profiles are entirely different.

What to verify before announcing a project

The first document to examine is not the press release but the technical connection terms: how many megawatts are available, exactly when they will be available and what grid work will be required. Next come an hourly load model, a backup scenario, transmission costs and confirmation of who will pay for the substations and lines. It is also important to check separately whether the announced power plants actually exist or remain at the statement-of-intent stage.

The practical conclusion is simple: for AI infrastructure, a megawatt matters more than the promised number of GPUs. Platform developers must incorporate power limits into the architecture and scaling plan. Product teams should not promise timelines until the power arrangements have been confirmed. Investors and governments should tie support not to announcements about future renewable facilities, but to permits, grid agreements and a clear construction schedule.