OpenAI CEO Sam Altman says AI energy efficiency has improved so sharply that a query now uses between 1,000 and 100,000 times less energy than three to four years ago. During his remarks at the ADIPEC energy conference held in Abu Dhabi on November 2, 2026, he combined that figure with a contrasting point: as AI usage continues to climb, energy supply still poses the industry’s foremost obstacle.
Key takeaways
Altman’s per-query energy figures are estimates, not a measured benchmark.
According to a research summary of his remarks, top AI users consume hundreds of billions of tokens monthly.
OpenAI, ADNOC and G42 agreed to a multi-year AI deployment partnership.
Altman identified energy—not chips, talent or regulation—as AI’s primary constraint.
AI energy efficiency gains come with a broad estimate
Altman attributed the reduction in AI query energy use to more efficient models. His remarks described a large improvement, but the wording was less precise than the numerical range suggests.
Queries are now “thousands or hundreds of thousands of times maybe more efficient” than a few years back, he said.
The qualification matters. The range represents Altman’s own estimate, and no calculation breakdown accompanied his remarks. It is best understood as a directional claim about improving efficiency rather than a benchmark for measuring the energy consumed by an individual query.
Surging token use offsets per-query savings
Improved AI energy efficiency has not removed the pressure from rising demand: people are running vastly more queries. Lower consumption per query and higher total consumption can happen at the same time.
According to a research summary of his remarks, top AI users now process hundreds of billions of tokens monthly , compared with roughly 100,000 tokens a month six years ago. That growth far exceeds the efficiency improvements described in the remarks.
Tokens are the small pieces of text AI models process—roughly a word or part of one. Processing them requires computing power, which consumes energy.
Tokens and queries are different units, so the usage figures cannot be directly matched against Altman’s per-query savings estimate. They nevertheless illustrate the scale of demand growth behind the industry’s continuing need for power. Total energy consumption depends on how much AI is used, not just how efficiently each task runs.
OpenAI, ADNOC and G42 target energy operations
A multi-year partnership was unveiled by OpenAI, Abu Dhabi National Oil Company ( ADNOC ), and Abu Dhabi-based AI firm G42 , aimed at rolling out advanced AI models throughout ADNOC’s operations. The agreement aims to optimize how the energy company runs its business.
The partnership puts AI to work within energy production and operations—the same sector supplying the power needed for AI’s expansion. Altman expressed hope that AI-driven operational and production efficiencies would eventually improve global energy availability by freeing up supply.
The stated deployment goal is to use OpenAI’s frontier technology to improve ADNOC’s operations.
Energy remains Altman’s primary industry challenge
Altman identified energy as AI’s primary challenge , ahead of chips, talent or regulation. His assessment places power availability alongside efficiency gains, rather than treating more efficient models as a solution on their own.
The distinction runs through his remarks: AI energy efficiency can improve substantially while rapidly expanding usage keeps increasing overall power needs. He has also called for a more informed discussion of AI’s energy requirements , including comparisons with energy consumption by humans more generally.
Altman has repeatedly advocated expanding electricity generation capacity, pointing to renewables and nuclear power as paths to supporting AI’s growth.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.
