Show the site for
Oh My AIJob helper Beta
Choose a botAsk with

PrivacyNothing you type here is sent to us or saved by us. When you press Go, your question goes to the bot you picked, in a new tab, and that bot’s company handles it under its own privacy policy. Please don’t type personal details.For some bots we also copy your question to your clipboard so you can paste it. How this site handles data

Home · Handbook · Energy & compute

Build & rules · AI Policy Handbook

Energy & compute

Technology, law, and practice. As of November 3, 2026.

Facts checked November 3, 2026

As of November 3, 2026

Figures reported by laboratories, agencies, and operators. Facility design varies; totals and per-task efficiency are separate measurements.

Why data centers appear on local agendas

Training and inference for large models concentrate electricity demand in data centers. Cooling historically used substantial water in many designs. Proposals for new facilities therefore intersect zoning, ratepayer, water, and emergency-curtailment policy.

Published demand figures

  • U.S. loadLawrence Berkeley National Laboratory estimated U.S. data centers used about 4.4% of U.S. electricity in 2023 and could use about 6.7% to 12% by 2028 (report as of Dec 2024). The same report estimated about 66 billion liters (about 17 billion gallons) of direct water use in 2023, largely for cooling.[4]
  • Global loadThe International Energy Agency estimated data centers used about 415 TWh in 2024 (about 1.5% of global electricity) and projected about 945 TWh by 2030 (as of Apr 2025).[8]
  • PUEUptime Institute’s Global Data Center Survey 2025 reported an average power usage effectiveness (PUE) of 1.54 industry-wide, with newer sites averaging 1.48 (as of 2025).[6]

Design changes operators describe

Liquid coolingCoolant delivered near chips rather than only room-scale air conditioning.
Closed-loop waterMicrosoft stated that its datacenter designs since August 2024 evaporate no water for cooling (company blog, Dec 9, 2024).[5]
Heat reuseMeta’s Odense fact sheet (Feb 2025) describes heat supply serving about 9,000 homes.[7]

Per-query efficiency can improve while aggregate electricity use still rises if total inference and training volume grow faster. Published totals and published per-task figures answer different questions.

Questions recorded when a large load is proposed

  • PowerSource mix, on-site generation, interconnection costs, and ratepayer allocation.
  • WaterAnnual withdrawal, closed-loop share, and drought contingency.
  • PUE / designDesign PUE, cooling method, and heat-reuse plans.
  • Texas large loadsSB 6 (2025) addresses interconnection, cost-sharing, and emergency curtailment for large loads of 75 MW or more; the Public Utility Commission adopted 16 TAC §25.370 on Feb 6, 2026.[9,10]
  • Backup / land useGenerator fuel, test schedules, noise, setbacks, lighting, and traffic.

Related background on cooling history: The data center comes full circle.

← Limits & evaluationFederal & EU rules →