AI Energy Consumption

Definition

The total electrical power consumed by AI systems across training, inference, and supporting infrastructure — including data centers, cooling, and networking. AI energy consumption is growing exponentially, with the International Energy Agency projecting AI could consume 3-4% of global electricity by 2030.

Why it matters

AI's energy footprint is becoming a board-level ESG risk and a physical constraint on growth. Data center power is already the bottleneck delaying new GPU clusters — not chip supply, not capital, but literal electricity availability. Investors are pricing energy access into AI company valuations, and companies with secured power purchase agreements trade at premiums. Regulators in the EU and California are beginning to require energy consumption disclosure for AI systems. For any enterprise scaling AI workloads, energy cost is becoming a line item that rivals compute cost. If your AI strategy does not account for energy constraints, it is incomplete.

In practice

The International Energy Agency estimated AI data centers consumed 415 TWh globally in 2024, projected to double by 2030 — equivalent to Japan's total electricity consumption. Microsoft's carbon emissions rose 29% year-over-year in 2024, driven almost entirely by AI data center expansion. Amazon, Google, and Meta all signed nuclear power agreements: Amazon contracted with Talen Energy for 960 MW from the Susquehanna nuclear plant, and Google signed a first-of-its-kind deal with Kairos Power for small modular reactors. In Texas, xAI's Colossus 2 supercomputer buildout faced grid capacity delays. Training a single frontier model like GPT-4 is estimated to consume 50-100 GWh — equivalent to powering 5,000-10,000 US homes for a year.