AI Data Centers Are Becoming a Power-Bill Political Problem
The Guardian reports that data centers are now using about 6% of electricity in both the United States and the United Kingdom, according to research from the International Data Center Authority. The group says global data-center power use has risen 15% in two years as AI demand accelerates and annual investment approaches $1 trillion. The report warns that once data centers reach roughly 5% of a national grid, community and political resistance tends to intensify. In the U.K., developers already face multiyear grid-connection delays, and the queue for connections reportedly grew 460% in the first half of 2025. The Guardian also notes recent scrutiny over whether Google-linked developers understated carbon emissions for proposed U.K. AI data centers. Greenpeace warned that an unchecked AI buildout could push higher energy bills, increase pressure on water supplies, and extend the life of fossil-fuel infrastructure. This is no longer a narrow tech-industry story; it is becoming a utility, zoning, climate, and cost-of-living story.
The important thing about the AI power story is not that data centers use electricity. Every serious industrial transition has a physical footprint. The important thing is that the industry still talks as if the AI boom lives in the cloud, while the bill shows up in substations, water systems, zoning fights, and eventually monthly utility statements.
That mismatch is where politics starts. A data center is not just a warehouse full of servers. It is a claim on the local grid. It is a claim on transmission capacity. It is a claim on cooling, land, tax incentives, emergency generation, and public patience. When research says data centers are already consuming around 6% of electricity in the U.S. and U.K., we are past the stage where this can be treated as a background cost of innovation. It is now a visible allocation fight.
The public was sold a story about frictionless intelligence. The real product requires concrete, copper, turbines, gas contracts, water rights, and years of interconnection work. That does not make AI bad. It makes it normal. The problem is that normal industries have to defend their costs in public. Tech companies prefer to defend their benefits in press releases while negotiating the costs in utility filings and local development meetings.
Jordi’s lens here is incentives. The incentive for the platform is to move fast, secure power first, and leave the politics to arrive later. The incentive for politicians is to promise jobs and future growth while hiding the near-term tradeoffs. The incentive for utilities is to build for large customers if regulators let them recover costs broadly. The incentive for ordinary ratepayers is much simpler: do not make me subsidize someone else’s server farm under the banner of progress.
That is why the 5% threshold matters. Once a load is large enough to be felt, the narrative changes. Below that line, data centers look like local economic development. Above it, they start to look like competition for scarce infrastructure. The same project that sounded futuristic in a ribbon-cutting can sound very different when residents hear about delayed grid connections, backup fossil generation, or higher bills.
The AI companies will argue that their systems will generate productivity gains that justify the buildout. Maybe they are right. But if the gains are private and the infrastructure risk is socialized, the public will see the deal for what it is. The smarter path would be transparency: who gets the power, who pays for the grid upgrades, who bears the water and emissions costs, and what happens if the projected AI revenues do not show up on schedule.
This is the second-order consequence that gets missed in the AI hype cycle. The bottleneck is not only chips. It is trust. Communities will tolerate industrial expansion when the bargain is honest. They revolt when they suspect the numbers were massaged, the costs were buried, and the winners were chosen before the hearing began.
AI may be virtual in the demo. It is not virtual on the grid. The sooner policymakers admit that, the less likely the next phase of AI politics becomes a backlash against both the technology and the institutions that smuggled its costs into public life.