Voters' fury over data centres rattles the AI boom

Democrats took control of both chambers of Congress in Tuesday's US midterm elections, and the artificial intelligence industry is counting the cost. An NBC exit poll, reported by Time, found that 56% of voters called AI and data centres a crucial or major factor in how they voted for Congress. Well-funded pro-AI political groups, the magazine noted, struggled to turn their spending into wins.

Markets reacted at once. On Wednesday Reuters reported that cash-rich AI giants rose while chipmakers and data-centre operators fell, after Democrats made gains in states such as Texas and Georgia where opposition to the buildings has grown. By Thursday CNBC said the selling had intensified, with chip stocks leading it, as senior Democrats including Bernie Sanders called for a halt to new data-centre construction. Nvidia fell about 9% on the day.

What changes, and what does not

It would be a mistake to read this as the imminent end of the build-out. Time cautions that sweeping federal AI legislation is not expected soon, and the White House remains in Republican hands. TechCrunch sees the likelier outcomes as state-level moratoriums and rules forcing developers to pay for their own grid upgrades rather than pass the cost to neighbours. It also records Goldman Sachs predicting construction delays in Georgia and Texas. Congress, meanwhile, could haul executives before committees and unpick Republican efforts to block state AI laws.

The industry has seen this coming. Axios reported last weekend that OpenAI had prepared a post-election policy campaign, built around what it calls mandatory "speed limits" for the most powerful systems. Its chief global affairs officer, Chris Lehane, said: "You're going to see us talk more and more about the concept of speed limits." On Thursday the company published a community compact and a policy blueprint. According to MLex's summary, OpenAI will sign local agreements for each campus it owns and operates, and urges utilities to make large electricity customers cover the costs they add to the grid. The pledges apply only to OpenAI's own sites, and the company concedes they are a starting point. The tactic is plain enough: a business that enjoyed a friendly Washington for two years now wants to help write the rules before hostile legislators do it for them.

Britain has met the same problem

The British version of this argument is about power rather than ballots, but it is the same argument. Microsoft recently withdrew from plans for what would have been Britain's largest AI data centre, at Loughton in Essex, City AM reported, after grid constraints put its supply date out of reach. Ofgem figures cited there show the connection queue growing from 41GW to 125GW in about seven months, with data centres accounting for at least 80GW of applications. The same report says a taskforce including Nvidia, Microsoft and Cisco has warned ministers that Britain has roughly a year to head off American-style opposition.

The government's answer is a data centre covenant, announced in last month's Budget, which offers priority grid access in return for local jobs, investment and water protection. Yet the Chancellor also said developers without a connection could generate their own power. The BBC's File on 4 found that four planned hyperscale sites are weighing gas-fired generation, with emissions that could exceed those of Manchester, Liverpool and Southampton combined. Ofgem told the programme that in many places the grid is already at capacity. Toby Perkins, who chairs the Environmental Audit Committee, said a big rise in gas use cannot be the future. Ministers are now caught between wanting the investment and wanting to keep faith with their net zero commitments.

What it means for UK businesses

Few British firms will ever build a data centre, but nearly all of them will rent the output. Their exposure is therefore to price and availability. If American construction slows and British sites wait years for power, the assumption that AI capacity will keep growing and getting cheaper deserves scrutiny. That is our inference rather than a forecast from any source above, but it is a prudent planning assumption for any firm whose budgets depend on heavy use of the newest models.

Three questions are worth putting to suppliers now. Where is the work actually processed? What happens to price and service if capacity tightens? And how dependent is a given workflow on a single provider's most advanced model, which could face tighter safeguards if OpenAI's speed limits, or something like them, reach the statute book? Firms that have built AI into daily operations on the belief that supply is unlimited should test that belief before the politics test it for them.