Data centres are on track to double their electricity and water consumption by 2030 as AI workloads scale, according to UN researchers, putting the infrastructure buildout on a collision course with utility-level clean energy commitments across the United States.

The numbers frame the problem precisely. Data centres consumed roughly 4.4% of total U.S. electricity in 2023, per the Department of Energy. The IEA projects that sector-specific consumption will grow at approximately 15% per year from 2024 to 2030 — more than four times the rate of total electricity demand growth. Meanwhile, the EIA's Annual Energy Outlook 2026 puts national demand growth at 2.1% annually over the past five years, a trajectory the agency expects to hold. Renewables are projected to lift their share of the U.S. generation mix from roughly 24% in 2025 to 27% by 2027, with nuclear holding steady — but that incremental clean capacity is being absorbed before it can move the needle on carbon intensity, because the demand curve is steeper.

The regulatory consequence is already visible at the state level. Nevada's largest utility has signalled it may fail to meet its 2030 clean energy mandate, attributing the shortfall directly to data centre load growth, according to AP News. Nevada is not an isolated case — it is an early indicator. The state sits at the intersection of a land-cheap, politically accommodating environment for hyperscaler campuses and a grid that was sized for a different demand profile. What plays out there will be replicated in other Sun Belt and Mountain West markets where the same site-selection calculus applies.

What the Maths Actually Mean

The 15% annual growth figure for data centres deserves scrutiny. Compound that rate over six years and sector consumption roughly doubles — consistent with the UN researchers' headline projection. But the baseline matters. If data centres were already consuming 4.4% of U.S. electricity in 2023, doubling that share by 2030 would push them toward 8–9% of a national grid that is itself expanding. That is a structural load, not a cyclical one, and it shapes long-term resource adequacy planning in ways that short-term demand forecasts do not fully capture.

Utilities operate on integrated resource plans with 10- to 20-year horizons. When a hyperscaler signs a large-load agreement, the utility must either procure new generation — ideally clean, to meet state RPS targets — or serve the incremental demand from existing, often fossil-fuelled, baseload. The lag between signing a power purchase agreement, permitting a wind or solar facility, and energising it is typically three to five years. The data centre can be operational faster than the clean power can be built to serve it. That gap is where the emissions accumulate and where state clean energy targets slip.

Water is the less-discussed constraint. Most large data centres use evaporative cooling — effectively trading electrical efficiency for water consumption. Doubling the power draw roughly doubles the water draw, in arid states where water rights are already contested. Nevada, Arizona, and parts of Texas are simultaneously building out data centre capacity and managing chronic water stress. The two trajectories are not currently being reconciled in any coordinated federal framework.

The broader tension here is structural rather than cyclical. The U.S. energy transition has been architected around a demand-growth assumption — roughly flat, with modest increases — that no longer holds. AI infrastructure is injecting a new, geographically concentrated, always-on load class that stress-tests both the pace of renewable build-out and the resilience of state-level clean energy policy. Utilities that made RPS commitments in a lower-demand environment now face a genuine choice between load growth and compliance.

What happens next will depend substantially on federal permitting reform, transmission investment, and whether major technology companies follow through on their stated 24/7 carbon-free energy commitments at the portfolio level — or whether those commitments quietly soften as the capacity gap widens. The Nevada signal is worth watching not as an anomaly, but as a leading indicator for the dozen or so states now absorbing the heaviest hyperscaler build.