Large-scale datacentre projects around the world are being delayed or cancelled just as demand for the computing power behind artificial intelligence surges. The Uptime Institute, which rates datacentres, found that of 250 global projects above 100MW announced between 2021 and 2024, roughly half will either not go ahead or will be completed late. This backlog matters because AI firms such as OpenAI, Anthropic and Google rely on datacentres to train and run their models — Google has already admitted its cloud business is "compute-constrained" — meaning the physical infrastructure is lagging well behind the pace of AI development.

The obstacles are numerous: securing energy and water, high construction costs, inexperienced developers without committed tenants, chip supply-chain bottlenecks, and opposition from local and environmental groups — one Virginia project, the Prince William Digital Gateway, was halted partly over its proximity to a Civil War battlefield. Power is the central constraint, with strained grids unable to meet demand: datacentres sit empty in California awaiting connections, and an Australian developer has sued the Dutch grid in Amsterdam. Uptime warns of an emerging era of "mega-gigawatt" datacentres, with about 80% of new US-led power demand raising sharp questions over resource trade-offs. The Guardian notes the UK government's AI "superpower" ambitions appear to overlook whether chosen sites even have electricity, while property consultancy JLL remains upbeat, forecasting some 1,200 new datacentres globally by 2030.