California Tightens Rules on AI Data Center Energy and Water Use

California AI data center rules have made the state the first in the United States to impose direct environmental reporting requirements on large compute facilities — a move that arrives just as the state's own electricity grid is straining under the weight of new compute capacity.

The California Energy Commission adopted new rules on September 22 that require large AI data centers to report their energy consumption and water use, and to submit to efficiency targets that the commission will set over the next 18 months. The rules apply to facilities consuming more than 10 megawatts of electricity — a threshold that captures the large-scale AI compute campuses that have been driving a significant portion of the state's recent load growth.

The move is the first of its kind in the United States. Other jurisdictions, including the European Union, have been moving on data center environmental reporting for years, but no US state had previously imposed direct efficiency requirements on compute facilities of any kind. Virginia, which hosts the largest concentration of data centers in the world, has been debating similar measures but has not yet acted.

What the rules actually require

The new regulations require covered data centers to report, on a quarterly basis, their total electricity consumption, their power usage effectiveness (PUE), and their water consumption for cooling. Facilities that fail to meet the efficiency benchmarks the commission sets will face penalties — the specifics of which are to be determined through the rulemaking process over the next year and a half.

The commission has also been directed to study the indirect environmental effects of AI compute, including the carbon intensity of the electricity the facilities draw and the upstream water costs of the semiconductor manufacturing supply chain that builds the GPUs and accelerators those facilities depend on. That study is due within two years.

The rules do not cap how much energy a data center can consume. They do not mandate renewable sourcing. What they do is create a disclosure and benchmarking framework that the commission can use to set enforceable efficiency standards — a softer tool than a cap, but one that creates a paper trail and a compliance duty that did not exist before.

Why California is moving now

The context is a state grid under pressure. California's electricity demand has been rising after years of flatness, driven in large part by data center load, electrified transport, and industrial reshoring. The California Independent System Operator has warned of growing reliability risks during heat waves, and the state has had to import more power than in previous years to meet summer peaks.

AI data centers are a particular concern because their load profiles are both large and concentrated. A single 100-megawatt AI training campus can draw as much power as a small town, and the trend toward larger clusters — driven by the competitive dynamics of the frontier model race — means the per-facility load is growing faster than the overall data center footprint.

The water question is separate but related. AI data centers use significant amounts of water for cooling, and in California — where drought conditions have been recurring and where the state's water rights system is already contested — that use is politically sensitive. The new rules make water consumption a reportable metric for the first time, which means the state will, for the first time, have a systematic picture of how much water AI compute is actually using.

The industry response

The data center industry's response has been predictable. Trade groups have argued that the rules duplicate existing federal reporting requirements, that the efficiency benchmarks are premature before the commission has done its study, and that California risks driving data center investment to other states if the compliance burden becomes too heavy.

There is truth to some of these arguments. Data centers already report energy data to utilities and, in some cases, to federal agencies. The question is whether those reporting streams are granular enough or public enough to support the kind of policy making the commission is attempting. The new rules argue that they are not.

The risk of investment flight is real but overstated in the short term. California remains a major market for cloud services, and many of the large AI compute facilities in the state are tied to specific infrastructure constraints — fibre connectivity, proximity to population centers, existing power contracts — that are not easily replicated elsewhere. The build-out can move, but it moves slowly, and the most recent wave of construction is already underway.

The national picture

California's move is likely to influence the federal conversation. The Federal Energy Regulatory Commission has been studying data center interconnections and grid impacts, and the Department of Energy has been funding research into data center efficiency. But neither has moved toward direct regulation of data center energy use.

Virginia's legislature has been considering data center siting and energy rules, but its approach has focused more on land use and transmission than on operational efficiency. Other states with significant data center footprints — Texas, Arizona, Oregon — have been watching California's process without committing to their own.

The European Union's Energy Efficiency Directive already requires data centers with an IT power demand of at least 500 kilowatts to report on energy reuse, PUE, and water use. California's 10-megawatt threshold is much higher, which means the state is regulating fewer, larger facilities than the EU — but it is doing so in a way that creates enforceable efficiency benchmarks, which the EU framework does not.

What this means for AI infrastructure

The AI infrastructure build-out has been treated, until now, as a market phenomenon — a story about demand, capital, and supply chains. California's rules reintroduce the regulatory dimension. The state is not trying to stop AI compute. It is trying to make it legible — to know how much power it uses, how much water it draws, and whether that use is efficient by some measurable standard.

For the companies building large AI compute facilities in California, the practical effect is a new compliance burden and a new source of public scrutiny. For the companies building them elsewhere, the practical effect is a signal that the regulatory environment is shifting — and that the states and countries that want this investment will need to offer more than just cheap power and permissive zoning.

The longer-term question is whether efficiency standards actually reduce AI's environmental footprint, or whether they simply push the compute to places with less scrutiny. The commission's study of indirect effects — including the semiconductor supply chain — is an attempt to address the second question. But that study is two years away, and the rules are effective now.

Sources

• California Energy Commission, "Decision Adopting Data Center Energy and Water Reporting Requirements," September 22, 2026
• The Verge, "California tightens rules on AI data center energy and water use," September 22, 2026, by Emma Roth
• California Independent System Operator, summer reliability assessment, 2026
• European Union Energy Efficiency Directive, Article 53, data center reporting requirements