# UN Report Projects AI Could Use 3% Electricity > Source: > Published: 2026-06-04 05:21:43.889582+00:00 # UN Report Projects AI Could Use 3% Electricity A United Nations University report, as summarized by AP and The Conversation, estimates that by 2030 global data centers could consume nearly **3%** of the world's electricity, rising to **935 trillion watt-hours** from about **448 trillion watt-hours** (AP). The report attributes roughly **208 million tons** of CO2 in the last year to data-center electricity use and estimates nearly **440 million tons** of CO2 by 2030 (AP). It also says AI-driven demand will raise the share of data-center energy tied to AI from about **20%** today to **40%** by 2030 (AP). The report invokes the economic concept of the "Jevons paradox" to argue efficiency gains may increase total resource use (The Conversation). Editorial analysis: Industry observers note rapid compute growth often creates grid and cooling stress, forcing planners to factor water and power alongside capacity. ### What happened A United Nations University report, reported by AP and summarized by The Conversation, quantifies the environmental footprint of data centers and projects sharp growth over the next several years. The report states global data centers used **448 trillion watt-hours** of electricity in the last reported year and produced about **208 million tons** of CO2, per AP. It projects that by 2030 data centers could use **935 trillion watt-hours**, roughly **3%** of projected global electricity, producing nearly **440 million tons** (AP). The study also estimates current AI workloads account for about **20%** of data-center energy and could reach **40%** by 2030 (AP). ### Technical details The report links energy and water use, noting data-center cooling consumed about **1.2 trillion gallons** (approximately **4.5 trillion liters**) of water in the last year, per AP. The Conversation highlights that the report frames efficiency gains against the "Jevons paradox," arguing efficiency can lower costs and expand total consumption. The report focuses on aggregate energy and water metrics rather than specific vendor architectures or model families. ### Editorial analysis - technical context Industry observers note that when compute demand scales rapidly, the bottlenecks shift from raw server capacity to supporting infrastructure: grid capacity, transformer availability, cooling water, and local permitting. Data-center siting and operational design increasingly interact with regional water stress and grid planning, particularly where evaporative or water-intensive cooling is used. ### Context and significance The projected scale-electricity use comparable to a top-ten country and emissions on the order of a mid-size nation-elevates AI's infrastructure footprint into mainstream climate and utility planning debates. Observers following the sector will watch how cloud providers, regulators, and municipalities respond, since growth in demand can drive both capital investment and regulatory scrutiny. ### What to watch Look for regional assessments of water availability where large campuses are planned, disclosures of facility-level energy and water intensity by operators, and changes in procurement toward less water-intensive cooling or greater on-site renewable power. Also monitor whether future reports provide more granular attribution between general cloud workloads and large AI training or inference jobs. ## Scoring Rationale The report quantifies infrastructure-scale impacts that matter to ML practitioners and operations teams: energy, emissions, and water needs influence cost, deployment choices, and regulatory attention. The story is notable but not a frontier research breakthrough. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)