Altman Says AI Efficiency Has Risen as Data Center Power Use Is Forecast to Rise

OpenAI CEO Sam Altman estimated that AI has become thousands of times more efficient, but the claim leaves overall electricity demand unresolved as data center power use is forecast to rise.

At ADIPEC’s opening ceremony in Abu Dhabi on Nov. 2, Altman said, “We’re thousands or hundreds of thousands of times maybe more efficient than we were three or four years ago.” His remarks concerned energy used per AI query, but he offered no standardized measure for the estimate.

Altman did not specify a baseline, model, hardware configuration or workload, and he did not show a calculation. His qualified estimate therefore cannot be treated as a precise reduction in electricity use for a particular query. It also does not indicate whether total AI electricity use has fallen.

Altman cited another measure of change: use by the most active AI users. He said the heaviest user six years ago used about 100,000 tokens per month, compared with roughly 500 billion tokens a month for the leading user now. Those figures concern individual users and do not measure total electricity consumption.

Global data center electricity use is projected to increase from 485 terawatt-hours (TWh) in 2025 to 950 TWh in 2030. Electricity use at AI-focused data centers is projected to triple over the same period. The global figure includes workloads beyond AI and is a forecast, not a measure of the queries Altman described.

The energy required for a query also depends on the task. Video generation, reasoning and agentic tasks can consume hundreds or thousands of times more energy per query than simple text generation. Changes in the mix of tasks therefore matter when assessing electricity demand alongside efficiency gains.

For 2030, the U.S. reference case estimates data-center demand at 649 TWh, or 11.8% of projected national electricity consumption. The estimate includes non-AI workloads, while other scenarios put the share between 9.5% and 15.3%.

No comprehensive global figures show how often AI is used or how demanding those uses are. Without a measure of query volumes and task complexity, the change in energy required per task cannot resolve the broader question of AI’s total power use.