UN warns data centers will double

The United Nations warning matters because it shifts the AI infrastructure debate from chips and model size to physical inputs: electricity, water, land and grid access. United Nations University researchers said on June 3 that data centres serving AI demand could consume 945 terawatt-hours of electricity by 2030, roughly double current levels, with water use rising on the same trajectory. That number is large enough to move the story out of the tech sector and into power markets. The International Energy Agency said in April that electricity demand from data centres rose 17% in 2025 and that overall data-centre consumption is set to double by 2030, while AI-focused facilities could triple their power use. (unu.edu) ### Why are power and water rising together? Data centres use electricity to run servers and networking equipment, and they use water directly or indirectly to keep systems cool and to support the power generation behind that electricity. United Nations University’s June 3 note said the environmental cost of AI is often undercounted because each kilowatt-hour carries not only carbon emissions but also water and land footprints. (iea.org) The same UN University release said the associated water footprint by 2030 could equal the basic annual domestic water needs of 1.3 billion people in sub-Saharan Africa. Reuters, citing the researchers, reported that data-centre power and water consumption are both expected to double by 2030 as AI demand expands. ### What is driving the jump? AI is the main accelerant because newer workloads are more compute-intensive than traditional cloud tasks. (unu.edu) The IEA said AI use is broadening even as power consumed per task improves, and that rising adoption of more energy-intensive tools such as AI agents is offsetting efficiency gains. Capital spending is following that demand. The IEA said the capital expenditure of five large technology companies topped $400 billion in 2025 and is set to rise another 75% in 2026, driven by data-centre investment. ### Why does this become a grid problem so quickly? Utilities do not just need more generation; they need transmission, substations, transformers and approvals in place before a data centre can take load. (iea.org) The IEA said tightening supply chains for gas turbines, transformers, chips and other components, along with planning and regulatory delays, are already slowing grid connections and other approvals. That helps explain why developers are looking for interim power strategies instead of waiting for full utility service. The IEA said many U.S. projects are advancing with onsite natural-gas generation because grid connections are slow, though it added that many of those projects remain in early stages and face technical and financial hurdles. (iea.org) ### What does that mean on the ground for new projects? Owners trying to hold construction schedules are likely to break capacity into phases rather than wait for one full energization date. In practice, that points toward staged load ramps, battery energy storage systems, temporary generation and other bridge solutions while permanent utility infrastructure catches up. That is an inference from the documented grid bottlenecks and onsite-generation push, not a direct quote from the UN report. (iea.org) The constraint is not only megawatts. Water availability, cooling design and local permitting can become deciding factors in where projects go and how quickly they can scale, because the UN researchers said AI’s footprint also extends to water use, land use and electronic waste. ### What should readers watch next? The next hard signals will come from utility filings, interconnection queues and developer power-supply announcements rather than from AI model launches. (iea.org) The IEA said the tech sector is already signing large volumes of renewable power purchase agreements and expanding conditional offtake deals tied to small modular reactor projects, while also pursuing onsite generation in the United States. (unu.edu) By 2030, the benchmark to watch is whether new data-centre capacity arrives with firm power and cooling plans attached. The UN University report published on June 3 and the IEA’s April 16 update now give policymakers, utilities and developers a common reference point for that buildout. (unu.edu) (iea.org)