AINEWS
The International Energy Agency (IEA) counts more than 2,500 gigawatts of projects in global connection queues. AI is the largest new load, and grid operators, the biggest AI and cloud companies, and regulators now negotiate over wires and permits.
More than 2,500 gigawatts of power projects are now waiting in global grid-connection queues, the lineup where utilities schedule the wires, transformers, and substation upgrades needed to hook new factories, battery farms, and data centres onto the network. The AI data centre boom is the single largest new load pushing that backlog wider, according to the International Energy Agency's 2026 Electricity Outlook. The grid queue is now the binding constraint on AI buildout.
Europe's FLAP-D cluster, the data-centre belt that runs through Frankfurt, London, Amsterdam, Paris, and Dublin, is where the constraint shows up most clearly. Consultancy Interface, cited in industry publication MEED's coverage of the IEA outlook, estimates new data-centre grid connections in those markets are taking seven to ten years to clear. The Netherlands and Frankfurt have responded by effectively banning new connections through at least 2030. Ireland is moving the other way, allocating €18.9 billion, roughly $21.8 billion at the euro's mid-2026 rate, for grid investment between 2026 and 2030 to keep up with planned AI and industrial demand.
The mismatch is now visible in dated, named projects. The IEA's 2026 grid chapter puts the planning, permitting, and build time for new grid infrastructure at five to fifteen years, against one to three for a typical hyperscale data centre.
In April 2026, OpenAI paused work on a planned multi-billion-pound UK data centre, citing regulatory constraints and high energy costs. Three months later, UK cloud provider Nscale announced the grid connection for its new AI campus will not be ready for its targeted 2027 opening, the second major UK AI buildout in three months delayed by grid connection while chip supply stayed on track.
The same constraint has appeared across the Atlantic. PJM, the grid operator covering 13 mid-Atlantic and Midwest states, received emergency approval from the Department of Energy to curtail data centres and other large loads during hot-weather emergencies, the first federal go-ahead a US system operator has had to cut power to AI-adjacent demand during grid stress. Two months earlier, Reuters reported that Google had told US policymakers the transmission system itself is the biggest challenge for connecting new data centres, a rare on-record admission from a hyperscaler that wires, not chips, are the binding constraint.
Three responses are in motion. Ireland's €18.9 billion plan is the most concrete, but the IEA flags similar build-out gaps in the United States, Germany, and the United Kingdom, and the agency's Europe commentary on data-centre energy constraints warns the gap will widen through 2030 even under optimistic build scenarios. Queue reform is the second lever: most of the 2,500-gigawatt backlog is permitting and connection-process, and regulators from Ofgem to FERC are beginning to triage projects by readiness as well as by first-come, first-served. Siting is the third: hyperscalers are starting to chase grid headroom over tax breaks, putting new campuses near existing substations and retiring fossil plants, sites that usually have spare grid headroom.
The next test lands in early 2027, when Nscale's UK campus is supposed to switch on and when PJM's emergency curtailment order is up for renewal.