The AI boom has sparked a race for chips, electricity and talent. Another constraint has appeared: water.
In 2023, data centers directly used 66 billion liters of water to cool their facilities and indirectly used 12 times that amount — 800 billion liters — for electricity generation. While 800 billion liters is plenty to raise communities’ concerns, both together account for only a small share of U.S. water consumption.
This growing demand does not mean policymakers should panic and ban data centers or require they use no water. The U.S. economy runs on data centers, regardless of AI, and policymakers should not hold back AI adoption because of fears (reasonable or overblown) about water use. There are good solutions, if policymakers adopt them.
Water is fundamentally a local issue. A gallon of water differs drastically in scarcity and value in Phoenix, Minneapolis, Columbus and Seattle — it may even vary significantly within a state. Local officials will have to decide where a proposed water use is acceptable and what obligations should be attached to its use.
Making these decisions quickly and efficiently is challenging, given the enormous scale of proposed data center construction. That, in part, explains the push for data center moratoriums. However, good decisions can be made faster if local decision-makers adopt the following five principles.
■ Improve transparency. Too often, communities cannot learn much about a data center’s proposed water use, which means stakeholders debate access to information rather than impact assessments. States should avoid nondisclosure clauses and instead require every large water user to report key metrics: water withdrawals, water consumption, source of supply, use of potable versus reclaimed water, peak-day demand, cooling technology employed and projected water demand at full build-out. Reporting should be quarterly for large facilities, annual for smaller ones, with very small facilities exempt. Some reports may be for regulator use only; other metrics should be fully public.
■ Effect review focused on local watersheds. States should review projects at the watershed level. Every watershed is different. Minnesota, for example, now requires watershed-level review, and imposes requirements based on the stress level of individual watersheds. Ohio has different rules for the Ohio River Basin and the Lake Erie region. This makes sense. Like in Minnesota, pre-application review should help developers before expensive decisions are set in stone.
■ Focus on performance. Regulators should set performance standards, not technology mandates. Cooling technology is changing fast. Older evaporative systems used more water, and older chips had to be kept much cooler. Today, chips can run hotter, racks can be denser and closed-loop liquid cooling virtually eliminates direct water use. There are still obstacles to deploying the newest technology at scale, but any rule that mandates today’s preferred technology may become tomorrow’s obstacle. States should instead set water-consumption limits per unit of computing load, adjusted by facility size, climate and watershed stress. During droughts, large data centers should be curtailed in line with other large users.
■ Use integrated water-energy reviews. Water used directly for data center cooling and water used indirectly for related electricity generation use the same watershed, so regulators should combine their assessments. Indirect use is at least 90 percent of the total, so it makes no sense to focus only on direct use. State utility commissions, water agencies and energy offices should jointly review large data center projects. They do not need to merge authority, but they do need to share a common record: projected electricity demand, peak load, likely marginal generation, direct water use, indirect water intensity, drought sensitivity and transmission assumptions among other variables.
■ Standardize metrics. Because water capacity and use are inherently local, the federal government should not pre-empt state water law or impose national cooling standards. It should, however, fund the development of standardized metrics and data collection, use federal cloud procurement as a lever to improve disclosure and encourage low-water technologies for facility cooling and electricity generation, and support research into technologies to reduce water needs for cooling and power generation. Federal grid reliability analysis should include the water implications of large new loads in interconnection rulings.
These principles would give local communities access to the full range of information needed for effective local review and management of the local water effects from new (and existing) data centers. These guidelines would also provide a path to increased certainty for data center developers and clearer demarcation of issues and possible solutions.
The alternative is much worse: secrecy, local suspicion, bans and litigation. Water has become a convenient hook for broader opposition to data centers, Big Tech and AI. Some of that opposition is opportunistic; some of it is justified. Either way, ignoring the water challenge will not make it disappear.
An improved regulatory framework would separate serious problems from political theater. It would let reasonable projects proceed in suitable places with certainty and speed. It would block or condition projects in unsuitable places. It would give communities facts before a decision becomes a done deal.
And it would provide greater speed and certainty, partly by eliminating arguments over data and information and by providing a factual foundation for better decisions.
Robin Gaster is a research director at the Information Technology and Innovation Foundation’s Center for Clean Energy Innovation and a visiting scholar at George Washington University. He wrote this commentary for InsideSources.com.
