Research note

Research Notes: AI's Energy Budget

Electricity demand

The International Energy Agency's Energy and AI report estimated global data-center electricity consumption at approximately 415 TWh in 2024, around 1.5% of global electricity use.

Its Base Case projected approximately 945 TWh by 2030, just under 3% of projected global electricity consumption. The report's 2030 scenarios ranged from about 670 TWh in the Headwinds Case to 1,260 TWh in the Lift-Off Case, with a High Efficiency Case around 800 TWh.

These figures describe data centers as a whole, not AI alone. The report identifies AI as the most important driver of growth alongside other digital services. Accelerated-server electricity use, driven mainly by AI adoption, was projected to grow faster than conventional-server use.

Infrastructure constraints

The IEA described grid congestion, connection queues, and stretched supply chains for equipment including transformers and gas turbines. It estimated that around 20% of projected data-center additions by 2030 in the Base Case could be at risk of delay.

This makes the constraint local and temporal as well as global. Total annual electricity statistics do not establish that generation, grids, equipment, and permits can reach a particular cluster when required.

Rebound and absolute impact

Price and environmental impact are different quantities. If impact per service falls tenfold while service use rises twentyfold, total impact doubles. If the energy source and material intensity change, a different result is possible; the mechanism has to be measured in physical units.

For rival goods, expanded access also expands consumption. Access to surgery, housing, transport, and staffed education consumes resources. Information and other low-throughput services have different scaling properties, but they are not impact-free.

A serious accounting therefore needs several measures rather than one efficiency ratio: absolute ecological ceilings, energy and material throughput, access to capabilities, distribution, and labor or time requirements.

What these figures do not support

They do not show that AI alone consumed 415 TWh in 2024, that a 10x economy must use ten times as much electricity, or that data-center growth makes expanded digital capability impossible. They establish a physical infrastructure requirement and a scenario range, not a verdict on the ambition.

Sources

  • International Energy Agency, Energy and AI, April 2025, especially the Executive Summary and Chapter 2, “Energy for AI.” Figures checked against the report PDF.

↓ Markdown