Data Centre Buildouts Are Outrunning the Power Grid, and the Gap Is Getting Real
The cloud industry has a numbers problem, and it is not the sort that a bigger budget solves. Around 250 data centre projects that each demand more than 100 megawatts were announced globally between 2021 and 2024, and research firm Uptime Institute reckons roughly half of them will be delayed or never built at all. Analysts at Uptime crunched the figures for a report that Cloud Computing News detailed this week. The headline is blunt: grid connections, land, substations, and transformers have become the binding constraint on where AI compute can actually land.
100MW sounds abstract until you frame it another way. Uptime says it is roughly the electricity use of about 300,000 homes. So each one of those announced projects is a small city's worth of demand, and the grid is being asked to add dozens of them at once.
The buildout that never showed up
The most visible casualty is in Virginia, the state that hosts much of the world's internet traffic. The Prince William Digital Gateway, a planned 2,000-acre campus, has run into a court challenge over its proximity to a Civil War battlefield. Opponents argued in a legal filing that the scale of the data centres and the electrical work around them would change the character of the historic ground. A local court has halted the project, a key backer has pulled out, and Uptime now lists it among cancelled schemes. Two more big names that dropped last year: Project Range in Arizona and the Cyberjaya campus in Malaysia.
It is not just legal fights. In Santa Clara, California, two data centres with a combined planned load of nearly 100MW could sit idle for years because the municipal utility cannot finish the grid upgrades until 2028. The city is managing 57 facilities in operation or development, and the upgrades carry a price tag of around $450 million. Meantime in Amsterdam, an Australian developer has sued the Dutch grid operator after its connection request was rejected.
Jay Dietrich, a research director at Uptime, points to a stack of reasons beyond grid strain. Some developers have no track record in data centres, and some projects have no tenant signed up yet. Energy and water needs are rising, projects are clustering into the same corridors, and the supply chain cannot keep pace.
"The global supply chain cannot support the number of proposed projects within the timelines being presented," Dietrich said. Chips and electrical gear set the delivery schedule, and so do cooling systems, skilled construction labour, and grid infrastructure. Tenant risk is real as well: several proposals come from developers without a committed customer.
A demand curve the grid can't follow
The trend is up, and the numbers are climbing faster than new capacity can arrive.
The International Energy Agency expects global electricity consumption from data centres to more than double, to roughly 945 terawatt-hours by 2030, equal to just under three percent of all global electricity that year. Uptime points out that data centre power hunger, especially in North America, is growing on grids that are already under strain. Projects announced last year alone, the institute estimates, would consume about 1.3 percent of projected 2025 global electricity if they ran at a quarter of planned capacity, which is nearly double current data centre demand. About 80 percent of that new demand sits in US projects.
Reuters has reported that capacity charges in the PJM Interconnection, the region's grid operator covering much of the eastern US, rose more than 1,000 percent between 2024 and 2026, and manufacturers blame data centre growth for pushing costs higher. Meanwhile the US Energy Information Administration expects electricity consumption to hit record levels in 2026 and 2027, with AI data centres, crypto farms, and electrification among the drivers.
Newer campuses only make it harder
Buildouts are getting bigger rather than smaller, and that is tightening the squeeze. Six projects announced last year each sought at least 5 gigawatts of power. Five were in the US and one in the UAE. To put it in scale, Ireland's entire peak power demand sits around 6GW. Uptime says the seven largest planned data centres globally are proposing a combined 45GW of on-site power, with natural gas as the main source. That 45GW is about the same as the UK's total peak demand.
Google has said its cloud business is "compute-constrained," because companies and consumers want more AI capacity than the data centres can deliver. That is the core irony of the moment: the cloud is being throttled not by demand or software, but by substations and transformer lead times.
Battery storage and on-site generation are being tested as workarounds. Property consultancy JLL expects about 1,200 data centres to be built globally by 2030, with AI as the main driver. Andrew Batson, JLL's global head of data centre research, told reporters that lease signings and groundbreakings in the first half of 2026 ran slightly ahead of his estimates. He said the industry has been developing ways to handle energy constraints, including battery storage and on-site power generation.
What breaks first
The US Department of Energy has flagged that rising data centre power use will need a mix of generation, storage, grid expansion, energy efficiency, and demand-side resources. No single lever fixes it. Grid operators cannot flip a switch and conjure new transmission capacity, and hyperscalers cannot shrink the physical laws of cooling.
For the cloud and edge sector, this is the real constraint story of 2026. Cloud and Edge Computing buildout plans now live or die on access to power and the patience of local grids, every bit as much as on chip supply. AI infrastructure demand is the fuel, but the grid is the throttle.
Individual projects will sort themselves out differently. Some will pair with community opposition and fold. Others will find battery storage and gas-turbine sites to ride out the gap. A handful, like the cancelled Prince William site, demonstrate what happens when the maths stops adding up long before the servers arrive.
The one thing everyone agrees on is that the compute is not slowing down. The utilities, the courts, and the supply chains are where the industry's next logjam will be decided, not at the chip foundry or in a rack in Dallas.
(Sources include the Uptime Institute report as covered by Cloud Computing News.)