LONDON — The secretary-general of the United Nations used the stage of London Climate Action Week to put a new item on the agenda: the environmental cost of artificial intelligence. António Guterres called on AI companies to disclose the environmental impact of their data centers and to commit to powering them with clean energy, according to a transcript of his remarks and a Reuters report.
The message was blunt. AI data centers consume electricity on a scale that has surprised even the industry’s own planners, and their carbon emissions are rising exponentially as training runs grow larger and more frequent, Guterres said. The companies building this infrastructure, he said, have a responsibility to account for its effects, and the current level of disclosure is not adequate for a sector growing this fast.
The speech moves the AI-environment question from the pages of environmental advocacy to the program of an international organization. That matters, analysts said, because the U.N. has convening power that no environmental group possesses: when the secretary-general speaks, governments listen, and when governments listen, regulation follows. The framing of AI’s environmental impact as a global governance issue is new, and it arrives at a moment when data center growth is accelerating around the world.
The numbers behind the speech are the reason it carried weight. Electricity demand from data centers has become a visible constraint on power grids in several countries, and utilities in regions with heavy data center concentration have warned that capacity is tightening. The International Energy Agency has projected that data center electricity consumption will grow substantially through the end of the decade, driven primarily by AI workloads, and the projections have become a fixture in utility planning documents.
The AI companies have not been silent. Microsoft has committed to becoming carbon-negative by 2030, Google has pledged to run its data centers on carbon-free energy around the clock by the same year, and Amazon has said it will match its electricity consumption with renewable energy. The companies have also signed a wave of nuclear power agreements, including deals with reactor developers that would dedicate new plants to data center campuses, a development that has reshaped the nuclear industry’s outlook.
The question is whether voluntary commitments are enough. Guterres’s speech suggested they are not, and his call for disclosure implies a standard that would make the commitments verifiable. Environmental groups have pressed for standardized reporting of data center energy use, water consumption and emissions, arguing that current disclosures are inconsistent and incomplete. The secretary-general’s endorsement gives that campaign its most prominent voice yet.
The politics are delicate for the companies involved. AI is a strategic priority for the governments of the United States, Europe and Asia, and none of them wants to slow its development, but public concern about the environmental effects of data centers has grown in communities hosting the facilities. Water use, in particular, has become a flashpoint, with data centers drawing criticism in regions facing shortages. Disclosure requirements would force the industry to answer for those effects in public.
The technology companies have begun to adapt their strategies. Several hyperscale operators have changed how they site data centers, favoring locations with surplus renewable power, and some have redesigned cooling systems to reduce water consumption. The shift to nuclear power, once seen as a distant option, is now a central part of several companies’ plans, and the deals signed this year have given the nuclear industry its strongest pipeline in decades.
Guterres’s speech lands in the middle of this adjustment. The industry can point to genuine progress on clean energy commitments and to a pipeline of nuclear projects that would transform its power mix, but it cannot yet point to a uniform standard of disclosure or to emissions data that investors can compare across companies. The secretary-general’s call effectively asks the industry to finish what it started.
The industry’s own numbers are beginning to align with the secretary-general’s message. Several hyperscale operators have published reports showing that their data center energy use is rising faster than their revenue, a trend that reflects the intensity of AI workloads, and utilities have begun asking regulators for permission to build new generation specifically to serve data center demand. The nuclear agreements signed this year, which would dedicate new reactors to individual campuses, represent a response at industrial scale, but they will take years to deliver power. In the meantime, the companies have been buying renewable energy contracts and, in some cases, paying premiums to secure capacity, costs that will eventually appear in the prices they charge for AI services.
What a disclosure standard would look like is already being debated. Environmental groups have proposed requiring companies to report data center electricity use, water withdrawal and emissions on a per-facility basis, with third-party verification, while the industry has pushed for aggregated reporting that protects commercially sensitive details. The U.N.’s endorsement of disclosure gives the stricter camp momentum, and the secretary-general’s speech is likely to be cited in the rulemaking processes under way in the European Union, which has been drafting energy-efficiency requirements for data centers. The gap between what the companies report voluntarily and what regulators would require is the real subject of the debate, and Guterres’s intervention was designed to narrow it.
For investors, the direction of travel is clear. Companies that disclose their environmental impact and demonstrate progress on clean power will face fewer regulatory surprises and lower reputational costs, while laggards will carry the risk of mandates imposed from outside. The U.N. has no enforcement power, but it has something almost as useful: the ability to set the terms of the public conversation, and to make the question of AI’s environmental cost one that executives cannot avoid answering.
