Uptime finds half of 100MW+ datacentre projects slipping

A count in the Guardian this week puts a hard number on something the AI infrastructure conversation has mostly hand-waved. The Uptime Institute identified 250 global datacentre projects over 100MW announced between 2021 and 2024, and roughly half are now either not happening or running late. That lands while hyperscalers are still describing capacity, rather than demand, as the binding constraint on AI growth.

The specific cases the reporting points to were meant to be marquee sites. Project Range in Arizona is cancelled. So is the Cyberjaya campus in Malaysia. In Virginia, the 2,000 acre Prince William Digital Gateway, next to a Civil War battlefield, was halted by a local court ruling after opponents argued the "solemn nature of this historic site would become marred by sitting in the shadow of the monstrous datacentres." California datacentres are reportedly sitting empty waiting for grid connections, and in Amsterdam an Australian developer sued the Dutch grid after being refused one.

The UK example the Guardian dwells on is the £8.2bn CoreWeave and DataVita complex in Lanarkshire, announced in January and pitched as powered entirely from on-site renewables by 2030. DataVita's stated plan is 400MW of solar and 800MW of wind, more than one and a half times the output of Whitelee, currently the UK's largest onshore windfarm. The reporting's analysis is that this would need between 40 and 100 sq km of land, against roughly 2 sq km actually in the planning system, and that government and developers privately acknowledged an "issue" with "power provision" while publicly still committing to up to 1GW of "new energy infrastructure."

Why this matters if you are not building datacentres yourself: AI product roadmaps, capex forecasts and "AI superpower" political pitches have all been priced as if the electron problem gets solved on schedule. Uptime's own line is that "surging datacentre power demands, particularly in North America, cannot be supported by power grids already operating under heavy strain." When roughly half the announced megawatts are slipping, those timelines start to look optimistic in a way valuations are not yet reflecting.

The honest caveats are worth naming. Uptime is one source. JLL's global head of datacentre research Andrew Batson told the Guardian he is "confident that the industry will work through the energy challenges," and the reporting does not tell you which of the stalled projects come back, or on what schedule. What it does establish is that the AI bottleneck has quietly moved from GPUs to grid interconnects, permits and land, which is a slower and more political problem to throw money at.