The AI boom has a physical cost, and it’s now large enough to see from the electricity grid. Data centres — the vast warehouses of servers that power everything from cloud storage to large language models — now consume 6% of all electricity generated in the United States. New figures suggest that milestone has been quietly crossed, and communities are starting to push back.
The Scale of the Problem
Total global data centre power consumption has hit 67.7 gigawatts, a 36% increase in just two years. The US accounts for nearly half of that on its own.
The biggest individual facilities now draw as much electricity as small cities, and spending on new data centre construction globally is approaching $1 trillion annually — with up to $700 billion of that anticipated in the US alone in 2026.
The US isn’t even the worst case. Germany’s data centres now consume 9.5% of the country’s electricity. The UK sits at 5.8%. According to a new report from the International Data Center Authority, there appears to be a consistent threshold at around 5% of national grid consumption where significant community and political resistance begins to emerge. Several major economies have now crossed it.
The Backlash Is Real and Growing
Hundreds of state-level bills regulating data centres have been introduced across the US. In Maine, the legislature passed a bill that would have halted construction of large facilities until 2027 — the governor vetoed it, but subsequently signed an executive order establishing an advisory council to investigate their impact. In Northern Virginia’s Data Center Alley, already one of the most concentrated clusters of facilities on Earth, new permits have been frozen until 2032 simply because the local grid has no remaining capacity.
Water is an equally charged issue. Most existing data centres rely on water-cooled systems that can consume millions of gallons daily — a single large facility can draw as much water as 6,500 households. While newer AI facilities increasingly use more efficient closed-loop cooling systems, these remain a small fraction of the overall fleet.
Some of the hostility, the report notes, is partly self-inflicted. Developers routinely register projects through locally incorporated shell companies with generic names, deliberately obscuring which tech giant is actually behind the development. In communities already anxious about grid strain and water use, that opacity breeds distrust.
The Hidden Waste
Beyond the headline consumption figures, the report surfaces two less-discussed problems that make the situation worse than it appears.
The first is so-called zombie workloads — abandoned test environments, forgotten applications, and unused services that continue drawing power without doing anything useful. These account for an estimated 13% of US cloud consumption, roughly 3 gigawatts of electricity consumed for no productive purpose whatsoever.
The second is the vast number of smaller data centres embedded in corporate offices and regional buildings that rarely feature in consumption estimates focused on large hyperscale campuses. These smaller, older, and considerably less efficient facilities account for at least 15% of total data centre power consumption — a significant hidden load that tends to be overlooked entirely.
The Tension at the Heart of the AI Boom
There is an obvious and uncomfortable irony here. The same technology companies arguing that AI will solve humanity’s biggest challenges — including, in some cases, energy and climate — are driving electricity demand at a pace that is straining grids, frustrating communities, and in some regions, pulling investment back into fossil fuel generation to keep up.
Whether the electrical grid can absorb the continued expansion of AI infrastructure, and how hard communities and regulators choose to resist it, may turn out to be one of the defining constraints on the AI industry’s trajectory. The technology can scale as fast as the investment allows. The grid, and the people who live next to it, cannot.