Power & policy now central to the AI build-out
Analysts and policy groups warn that AI’s energy demands are becoming a core regulatory and infrastructure issue — from grid strain to emissions — and that overbuilt data-centre capacity could become stranded if demand softens. That framing makes electricity supply, permitting and regional policy first-order risks for AI investments. (brookings.edu) (policyalternatives.ca)
The new choke point in the artificial intelligence boom is not chips. It is electricity, because the International Energy Agency estimates data centres used about 415 terawatt-hours in 2024, or roughly 1.5% of global electricity consumption. (iea.org) That load is rising fast, because the International Energy Agency says data-centre electricity use grew about 12% a year over the last five years and could climb past 1,000 terawatt-hours by 2030 in its base case. (iea.org) A data centre is basically a warehouse full of computers, and artificial intelligence training packs more power-hungry chips into each building than older cloud-computing sites did. The Electric Power Research Institute says utilities are now dealing with both more data centres and bigger single-site loads tied to artificial intelligence. (restservice.epri.com) That changes the investment math, because a company can buy servers in months but a new transmission line or gas plant can take years to permit and build. Brookings wrote on April 10, 2026 that electricity supply, water use, siting and international governance are now central constraints on artificial intelligence expansion. (brookings.edu) Utilities are reacting like landlords asked to wire an entire new suburb overnight. A 2024 Electric Power Research Institute utility survey said clusters of large new point loads are testing how fast electric companies can expand generation, transmission and distribution. (restservice.epri.com) Regulators are starting to treat that as a consumer-bill issue, not just a tech-sector issue. Morningstar DBRS reported in March 2026 that several United States legislative proposals were aimed at shielding ratepayers from higher bills as hyperscale data-centre demand pushes up power and water needs. (dbrs.morningstar.com) The local fights are getting sharper, because data centres do not just use electricity. Brookings said the April 2026 dialogue focused on rising electricity and water consumption worldwide, which turns zoning boards, utility commissions and environmental agencies into gatekeepers for artificial intelligence projects. (brookings.edu) There is a second risk hiding behind the power scramble: overbuilding. The Canadian Centre for Policy Alternatives argued on March 26, 2026 that if artificial intelligence demand cools, communities could be left with oversized data-centre infrastructure and stranded costs after utilities and governments spent heavily to serve it. (policyalternatives.ca) That stranded-asset problem is simple: a grid upgrade lasts decades, but an artificial intelligence spending cycle can turn much faster. The Canadian Centre for Policy Alternatives says governments should require developers to cover more of the cost of grid upgrades and new generation so residents are not left paying for infrastructure built around speculative demand. (policyalternatives.ca) So the artificial intelligence race now runs through county permits, public utility commissions and power-plant queues. In 2026, the winning regions are not just the ones with the best chips or biggest models, but the ones that can deliver firm electricity, water access and political approval before the financing window closes. (brookings.edu)
