UN Report Projects AI Could Use 3% Electricity
A United Nations University report estimates that by 2030 the world's data centers could consume nearly 3% of global electricity, rising to 935 trillion watt-hours from about 448 trillion watt-hours in the last reported year, as summarized by The Associated Press and The Conversation. AP reports the study attributes about 208 million tons of CO2 to data-center electricity use last year, comparable to Argentina, and projects nearly 440 million tons by 2030. It also estimates the AI share of data-center energy could rise from about 20% today to 40% by 2030, and that cooling consumed roughly 1.2 trillion gallons of water last year. The report invokes the "Jevons paradox" to argue efficiency gains may increase, not reduce, total resource use. UN University adds that AI-specialized data-center capacity is highly concentrated, with over 90% in two countries.
What happened
A United Nations University report quantifies the environmental footprint of data centers and projects sharp growth this decade, as reported by The Associated Press and summarized by The Conversation. Per AP, global data centers used about 448 trillion watt-hours of electricity in the last reported year and produced roughly 208 million tons of CO2, comparable to Argentina. The report projects that by 2030 data centers could use 935 trillion watt-hours, nearly 3% of projected global electricity, producing close to 440 million tons of CO2. AP reports the AI share of data-center energy could rise from about 20% today to 40% by 2030.
Technical details
Per AP, data-center cooling consumed about 1.2 trillion gallons (roughly 4.5 trillion liters) of water in the last year. The Conversation reports the study frames efficiency against the "Jevons paradox," the idea that cheaper, more efficient AI encourages more total use and can erase efficiency savings. UN University also notes AI-specialized capacity is geographically concentrated, with over 90% in two countries.
Editorial analysis
Class B analysis: when compute demand scales quickly, bottlenecks tend to shift from server capacity to supporting infrastructure such as grid capacity, transformer supply, cooling water, and local permitting. Data-center siting then interacts with regional water stress and grid planning, especially where water-intensive cooling is used.
What to watch
- •Regional water-availability assessments where large campuses are planned.
- •Facility-level disclosures of energy and water intensity by operators.
- •Procurement shifts toward less water-intensive cooling or on-site renewable power, and more granular splits between general cloud and AI workloads.
Key Points
- 1A UN University report projects data centers could use nearly 3% of global electricity by 2030 (935 trillion watt-hours), increasingly driven by AI.
- 2It invokes the "Jevons paradox," warning efficiency gains may be offset by higher overall AI use, water, and emissions.
- 3AI-specialized data-center capacity is highly concentrated, raising grid, water, and siting pressures that shift focus to infrastructure planning.
Scoring Rationale
The report quantifies infrastructure-scale energy, emissions, and water impacts that directly affect AI cost, siting, and regulatory exposure, making it broadly relevant to practitioners and operations teams. It is an authoritative synthesis with mainstream policy weight rather than a research breakthrough, placing it in the upper-notable tier.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems