The most consequential environmental debate in the technology industry has quietly shifted from carbon to water. As artificial intelligence workloads continue to expand inside the world’s data centres, regulators in the United States and Europe are moving in the same direction simultaneously: demanding that operators account for every litre they consume, or face legally enforceable penalties. The political window for voluntary reporting is closing.
The Regulatory Turn: Two Jurisdictions Move This Month
In Utah — the second-driest state in the continental United States and home to more than 20 operational data centre facilities — the state legislature passed House Bill 76, sponsored by Rep. Jill Koford (R-Ogden), during its 2026 session. Under HB 76, data centres would be required to notify local water systems before construction of how much water they expect to use, and then report actual annual usage thereafter. The bill passed the Utah Legislature and is among those awaiting signature by Governor Spencer Cox. MarketScreener
The legislation would impose fines of $10,000 per day for failure to report. Koford framed the bill as a matter of planning rather than prohibition, saying: “What we want to know is what are you diverting? What are you consuming? What are you discharging? So that we can get a handle on it and we can plan for the future.” Utah State Engineer Teresa Wilhelmsen confirmed that the state currently lacks reliable data on how much water data centres consume.
The scale of what has gone untracked is significant. The NSA’s data centre in Bluffdale consumed more than 23 million gallons in a single month, according to an Aspen Institute analysis cited in the legislative process. Some individual facilities can use up to 5 million gallons of water per day — more than 12,000 times the average daily household consumption.
At the European level, the regulatory architecture is further along but still developing. The European Commission is preparing a Data Centre Energy Efficiency Package planned for adoption in April 2026, which will introduce a rating scheme for data centres in Europe and launch work on minimum performance standards — with the goal of achieving carbon-neutral European data centres by 2030. CNBC
The reporting obligation already in force under EU Delegated Regulation 2024/1364 requires data centre operators with IT power demand of at least 500 kW to report annually by May 15 to a central EU database, covering key sustainability metrics including energy consumption, waste heat, and crucially, freshwater use. The forthcoming package is expected to formalise performance thresholds around these disclosures. CNBC
The Science Behind the Mandate
The regulatory pressure is grounded in a body of research that, until recently, remained largely confined to academic journals. A peer-reviewed study by Shaolei Ren, Pengfei Li, Jianyi Yang, and Mohammad A. Islam, published in Communications of the ACM, established that training GPT-3 in Microsoft’s US data centres consumed a total of 5.4 million litres of water. For inference, the study found that GPT-3 needs to consume approximately 500ml of water for every 10 to 50 medium-length responses, depending on where and when the model is deployed. Republic World
The caveat in that figure matters for how it is understood. The 500ml estimate applies specifically to the GPT-3 model and varies materially by data centre location, cooling system design, and the time of day at which queries are processed. Ren, an associate professor of electrical and computer engineering at UC Riverside, has noted that data centres consume water in two distinct ways: indirectly, through the electricity they draw from power plants that use water-cooled generation systems, and directly, through on-site cooling towers that remove server heat by evaporating water into the atmosphere. Autoguideindia
Ren’s research projects that global AI demand could account for between 4.2 and 6.6 billion cubic metres of water withdrawal in 2027 — more than the total annual water withdrawal of four to six Denmarks, or approximately half that of the United Kingdom.
What the Technology Industry Is Doing
The largest technology companies have made public commitments to become “water-positive” — meaning they return more water to local watersheds than they withdraw — by 2030. The gap between pledge and execution remains a subject of ongoing scrutiny.
Microsoft, Google, and Meta have each implemented distinct approaches to closing that gap. Ren’s research has noted that most industry efforts have focused on improving on-site water efficiency — through better cooling tower design, recycled water use, and scheduling AI training for cooler hours when evaporative losses are lower. Ren proposed the analogy: “We don’t water our lawns at noon because it’s inefficient. Similarly, we shouldn’t train AI models when it’s hottest outside. Scheduling AI workloads for cooler parts of the day could significantly reduce water waste.”
The engineering response to the water problem is also driving a structural shift in how data centres are built. Direct-to-chip liquid cooling and immersion cooling systems — which circulate coolant at the server level rather than evaporating water through towers — reduce facility water needs substantially. Data centre electricity demand is projected to more than double from its current level to approximately 945 terawatt-hours by 2030, driven primarily by the growth of accelerated computing for AI. That scale of expansion, combined with water-intensive conventional cooling, is what is forcing both the engineering and regulatory responses simultaneously.
The Transparency Problem
oth the legislative and voluntary corporate responses are constrained by a fundamental obstacle: the absence of reliable baseline data. In Utah, the data centre owned by Meta in Eagle Mountain entered a confidentiality agreement with the city when it was built, requiring the city to alert Meta whenever anyone requested its water usage records. Rep. Koford’s bill would override such arrangements for the purposes of state reporting. Seatrade Maritime
Ren has argued that the absence of transparency makes measurement, benchmarking, and improvement nearly impossible. He has proposed that AI model developers disclose their water footprints alongside carbon footprints in published research and corporate sustainability filings — a standard that does not yet exist in most jurisdictions.
The Utah bill, if signed, would be among the first state-level laws in the United States to make that transparency a legal rather than voluntary requirement. The EU’s forthcoming package would create the first binding rating scheme in Europe. Together, they represent the beginning of a governance framework for a resource that, until recently, no one in the data centre industry was required to count.
EU regulatory details are sourced from the European Commission’s official data centre energy page and Delegated Regulation EU/2024/1364, as analysed by White & Case LLP and Osborne Clarke. The EU Data Centre Energy Efficiency Package is confirmed for planned adoption in April 2026 and had not been formally adopted as of the date of this report. Utah HB 76 details are sourced from the Utah Legislature’s official bill text, the Salt Lake Tribune, Utah News Dispatch, and Axios Salt Lake City. The Governor’s signature has not been confirmed as of March 10, 2026. The 500ml water consumption figure is sourced from the peer-reviewed paper “Making AI Less Thirsty” by Li, Yang, Islam, and Ren, published in Communications of the ACM; the figure applies specifically to GPT-3 and varies by facility and deployment conditions.
