OpenAI chief executive Sam Altman said on 2 November that the world’s heaviest AI user now consumes about 500 billion tokens a month, against 100,000 a month for the heaviest user six years ago, who he said worked at OpenAI. He spoke at the opening of ADIPEC, the Abu Dhabi energy conference, the same morning OpenAI, ADNOC and G42 signed an agreement on AI across ADNOC’s operations.
That is a rise of about 5 million times in six years, by our arithmetic (500 billion divided by 100,000). Altman also said that “the global average per capita is about 100,000 tokens a month and the token leader uses about 500 billion tokens a month,” according to YouTube’s automatic captions of the session video, at 11:17 to 11:21.
Watch · Sam Altman at the ADIPEC 2026 opening
The ADIPEC 2026 opening session with Sam Altman’s keynote and Q&A, as posted by DRM News on YouTube on 2 November. Timestamps in this piece refer to this video.
Business Pill · Tokens per Watt
A short explainer on tokens per watt as a measure of AI efficiency. It teaches the general idea only and says nothing about Altman’s remarks or the ADNOC agreement.
The key insight: Sam Altman put the heaviest AI user at about 500 billion tokens a month, against 100,000 six years ago, a rise of about 5 million times by our arithmetic, and on the same stage named energy as OpenAI’s one challenge, saying the physical world will be “the very dominant factor of the cost.”
Six years ago, Altman said, that top user “was using a 100,000 tokens a month and it seemed ridiculous to us.” The global average then “was like zero,” he said. He called the change “an unbelievable growth rate of how much people want to use AI.”
He set that demand against energy per query. “The energy per query is falling incredibly dramatically,” Altman said, adding: “We’re thousands or hundreds of thousands of times maybe more efficient than we were three or four years ago.”
Then he described the other side. “As we make it more efficient, people just want to use it much more,” Altman said. He gave no figure for total energy use.
Tokens a month as Sam Altman described them at the ADIPEC 2026 opening on 2 November: the heaviest user six years ago, the global average per person now, and the heaviest user now. Bar length is the power of ten. The 5 million times ratio is our arithmetic; the figures are Altman’s approximations.
The 10,000-Agent Proof
In his keynote, Altman referred to the moment “when we proved the first Millennium Prize problem,” and said: “It was 10,000 agents of the model that is going to be released soon, the next GPT working” for about four days, “non-stop collaborating with each other perfectly.”
“There was no hierarchy. There were no bosses. It was fully decentralized,” he said, at 5:58 to 6:02 in the captions. He did not name the problem or the model on stage.
We reported on 26 October that OpenAI says its AI refuted the Exponential Time Hypothesis, in a post that also announced an existence proof for quantum chromodynamics that OpenAI says includes pure Yang–Mills theory. The Advisory Group on Mathematics and Artificial Intelligence, an independent group that OpenAI says it is working with, notes that “the existence of quantum Yang-Mills theory is part of a Millennium Prize problem,” and asked OpenAI to release its outputs.
Those results are OpenAI’s own descriptions and had not been independently confirmed when we reported them. Altman did not say on stage whether the problem he referred to is the one in that announcement.
The Energy Deal on Stage
The Abu Dhabi Media Office said ADNOC, G42 through its subsidiary Inception42, and OpenAI signed an agreement making ADNOC the first company in the energy sector to deploy OpenAI’s frontier technology, using in-country inference infrastructure in the UAE. It gave no dollar value.
Altman said the three companies had discussed the arrangement for a couple of years. “And now the models have caught up,” he said, describing a plan to take “the hardest most valuable problems” ADNOC faces to the frontier of AI.
Opening his keynote, Altman said a point made by ADNOC chief executive Sultan Al Jaber had resonated with him: that the AI industry has long talked about what it needs from the energy industry. “And now I think we can repay the favor,” Altman said. Later he added that OpenAI would like to “more than return the favor of the energy demands that we’ll have by making, you know, a net gain for energy availability in the world.”
Asked in a closing round to name one challenge OpenAI faces today, Altman answered with one word: “Energy.” He said models, robots and chip fabs would get cheaper, but “atoms are still atoms,” and the physical world that powers AI is “going to be I think the very dominant factor of the cost and the abundance and how much we can create.”
What Else He Said
On capability, Altman said that at the start of this year “AI was mostly not as smart as our smartest scientist and now it’s mostly smarter than them,” calling the comparison very approximate. On the UAE, he said he thought 80% of young people there are weekly ChatGPT users.
On control, he said OpenAI thinks “people remain in control of the future” must not be up for debate, and that “there should not be a handoff to AI.”
Connecting the Coverage
This morning we reported that Abu Dhabi signed 4 OpenAI deals in an AI-native government push, with ADNOC as the first energy company on OpenAI’s frontier technology. What’s new here is Altman’s own account on stage: the token figures, the 10,000-agent description and energy named as OpenAI’s one challenge.
As we read Altman’s remarks together, he described two curves moving in opposite directions: energy per query falling by what he called thousands of times or more, and demand for the heaviest user rising about 5 million times, by our arithmetic from his figures.
As we read it, the ADNOC agreement gives that argument a counterpart: an energy company buying frontier AI from the lab whose chief executive, on the same morning, named energy as the constraint on how much AI can be produced.
Sam Altman, at the ADIPEC 2026 opening in Abu Dhabi (YouTube automatic captions)
“atoms are still atoms”
Three Implications
FOR ENERGY COMPANIES Altman said OpenAI would like to “more than return the favor” of AI’s energy demands, and the Abu Dhabi Media Office said ADNOC is the first company in the energy sector to deploy OpenAI’s frontier technology.
FOR BUYERS OF AI Altman said energy per query has fallen by thousands of times or more in three or four years, and that “people just want to use it much more” as it gets more efficient.
FOR ANYONE TRACKING OPENAI’S MODELS Altman said 10,000 agents of “the model that is going to be released soon, the next GPT” worked for about four days on the result he described as a Millennium Prize proof; he did not name the model.
The framework describes the bottom layer of the AI stack, Energy & Physical, as “The constraint that defines the ceiling on how much intelligence can be produced in any given year.”
As we read it, Altman’s one-word answer, energy, puts OpenAI’s stated constraint in that bottom layer, while his token figures describe demand at the top. The framework is the owner’s argument and does not discuss Altman’s remarks.
The quotes come from YouTube’s automatic captions of a 21-minute video of the session posted by DRM News on 2 November, not from an official transcript; we corrected only the spelling of names. Altman’s token and efficiency figures are his approximations, and he did not say how they were measured.
We did not establish which problem and which model the 10,000-agent run involved, or whether OpenAI has released the outputs the advisory group asked for. Neither OpenAI nor ADNOC has given a value for the ADNOC agreement in anything we found.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Altman used an energy conference to put numbers on AI demand: the heaviest user up about 5 million times in six years, by our arithmetic from his figures, while energy per query has fallen by what he called thousands of times or more. He named energy as OpenAI’s one challenge.
A note on sourcing. This piece relies on YouTube’s automatic captions of the ADIPEC 2026 opening session video (posted by DRM News on 2 November 2026), read in full, and the Abu Dhabi Media Office statement of 2 November. Timestamps refer to that video. Captions are machine-generated; we corrected only the spelling of names. The 5 million times ratio is our arithmetic from Altman’s figures.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
