US Data Center Electricity Use Could More Than Double by 2030 - and That's a Grid-Planning Emergency

A new report from the Kansas Health Institute landed this week, and it deserves more attention than it's getting in the AI-hype cycle. The headline number is stark: US data centers consumed an estimated 183 TWh of electricity in 2024 - more than 4% of the country's total electricity use - and that figure is projected to grow 133% by 2030, reaching 426 TWh, according to Lawrence Berkeley National Laboratory data cited in the report. That's not a rounding error. That's a structural shift in the load profile of the American grid, arriving faster than most utilities planned for.

The KHI report frames this explicitly as a public health and community planning problem, not a technology story. That framing matters. It shifts the question from "how much power does AI need?" to "who decides how that power gets built, who pays for it, and what happens to the communities in the way?"

The Numbers Behind the Projection

Lawrence Berkeley National Laboratory has been the most rigorous tracker of data center energy use over time, and its findings keep getting revised upward. Data center power consumption jumped 14% between 2023 and 2024 alone - and while hardware has grown more efficient, the scale of computational demand more than offsets those gains. That's the core dynamic: efficiency improvements are real, but they're being swamped by volume.

The Electric Power Research Institute's February 2026 update put the range even wider. EPRI now projects data centers could consume 9% to 17% of US electricity generation by 2030 - estimates that are roughly 60% higher than what EPRI itself anticipated just two years ago. The revision isn't a modeling error; it reflects how rapidly announced projects have accumulated since early 2025.

The geographic concentration makes the national average misleading as a planning tool. In Virginia, data centers could consume between 41% and 59% of the state's electricity by 2030, up from 25% today, according to EPRI. Seven other states - Arizona, Indiana, Iowa, Nebraska, Nevada, Oregon, and Wyoming - could see data centers consuming more than 20% of their power by decade's end. Planners working from national averages will miss the stress points entirely.

Water: The Underreported Half of the Story

Electricity is the number that gets the headlines. Water is the number that should be keeping local officials up at night.

US data centers consumed an estimated 17 billion gallons of water directly in 2023 for cooling, and that figure could double or quadruple by 2028, based on Lawrence Berkeley National Laboratory scenarios. The indirect water footprint - the water consumed by the power plants generating the electricity - is far larger still, estimated at around 211 billion gallons annually.

The location problem compounds this. About two-thirds of US data centers built since 2022 are in high water-stress areas, according to Bloomberg News analysis. The KHI report notes that even five years ago, 20% of US data centers were already sited in water-stressed regions. The industry has not corrected course; it has accelerated into it.

Cooling is the mechanism. As chip density increases, air cooling becomes insufficient, and operators shift to water-based chillers and cooling towers. Indirect water use - from electricity generation - often makes up 80% or more of a data center's total water footprint. That means the energy source matters as much as the cooling technology: a data center powered by gas or coal carries a much larger water burden than one powered by wind or solar.

The Grid-Planning Problem Nobody Wants to Own

Here is where the KHI report is most useful, and most honest. It identifies a structural mismatch between the speed of data center development and the speed of grid infrastructure.

Grid interconnection queue-to-commercial-operation timelines have grown roughly 60% since 2017, now averaging over 2,100 days for projects with a first power year in 2025. Getting a new data center connected to the grid in a constrained market now routinely takes five or more years. That timeline is commercially untenable for hyperscalers with capital already committed and compute demand already live.

The response from developers has been predictable: bypass the grid entirely. Forty-six planned data centers, with a combined capacity of 56 GW, will avoid connecting to the grid altogether, according to analysis by Cleanview. Behind-the-meter gas generation - turbines and reciprocating engines installed at the data center site - can be deployed in as little as 18 months. xAI's Colossus facilities outside Memphis are the most visible example, running on mobile gas turbines while communities downwind absorb the air quality consequences.

Bank of America analysts estimate that data centers alone could add roughly 125 GW of US electric load through 2031. The planned natural gas capacity queued for interconnection between 2026 and 2030 has grown dramatically in response: by the first quarter of 2026, the total planned natural gas project pipeline stood at 64 GW, up from just 23.8 GW at the start of 2025. That's a near-tripling in 15 months. Under current policy trajectories, natural gas dominates near-term incremental supply regardless of what tech companies say in their sustainability reports.

The Policy Gap Is the Story

The KHI report is careful here, and the carefulness is itself informative. It notes that policy responses "reflect a tension between furthering state-level economic development interests and balancing local community health and environmental concerns." That's a polite way of saying that states have been competing to attract data centers with tax incentives, and are now discovering they didn't read the fine print on the energy and water bills.

New York moved first and hardest. On July 14, Governor Kathy Hochul signed an executive order imposing the nation's first statewide moratorium on new hyperscale data centers - defined as facilities capable of drawing at least 50 MW - while the state develops environmental standards. The Department of Public Service will spend up to a year preparing an environmental impact statement covering energy demand, water use and quality, air pollution, noise, and effects on disadvantaged communities.

The moratorium is a genuine policy experiment. It trades near-term investment for the time to build a regulatory framework that didn't exist before. Tech companies have pushed back, arguing it cedes ground to China in the AI race. That argument has real political weight. Whether New York holds the line will tell us a lot about whether any state can.

Photo: Taylor Vick / Unsplash

The KHI report also notes that local governments have tended to act more directly - on land use, zoning, and permitting - driven by community opposition and the immediacy of local resource impacts. That's where the real friction is happening: not in federal rulemaking, but in county commission meetings where residents are asking why their water pressure dropped and their electricity bills went up.

What Grid Planners Should Take From This

The KHI report is addressed to policymakers and public health professionals. But its implications for grid operators and utility planners are just as direct.

The load forecasting problem is real and underappreciated. EPRI's electricity-demand estimates for data centers are about 60% higher than the group anticipated two years ago. If the most sophisticated energy research organization in the country is revising its forecasts by 60% in two years, integrated resource plans built on older projections are already stale. The question isn't whether to update them - it's whether utilities have the regulatory authority and capital access to move fast enough once they do.

The behind-the-meter trend complicates this further. Load that goes behind the meter disappears from utility forecasts. It still consumes gas. It still affects air quality. It still competes for water. But it doesn't show up in the interconnection queue, and it doesn't trigger the transmission planning reviews that would otherwise force a reckoning with cumulative impact. Planners are flying partially blind.

The water-electricity nexus deserves dedicated modeling. Most grid planning treats electricity and water as separate systems. Data centers make that separation untenable. The fuel mix that powers a data center determines its water footprint as much as its cooling technology does. Integrated resource plans that don't account for water stress in siting decisions are missing a material risk.

None of this is a reason to stop building data centers. The compute infrastructure being built right now will matter. But the KHI report is a useful corrective to the idea that the planning can happen after the fact. The grid doesn't work that way. Neither does a watershed.