Speaking at the annual meeting of the World Economic Forum in Davos, Sam Altman delivered an unusually direct warning: the next constraint on artificial intelligence will not be ideas, talent or even chips — it will be electricity.
According to Reuters, Mr. Altman told attendees that the rapid expansion of advanced A.I. systems will require “staggering amounts of energy,” adding that breakthroughs in clean and abundant power are essential to sustain progress.
Compute is the limiting factor for A.I.,
he said, noting that the industry must prepare for energy demand on a scale few policymakers currently anticipate.
Mr. Altman, who leads OpenAI, has increasingly linked the development of frontier models to physical infrastructure. At Davos, he argued that while model capabilities continue to advance, the ability to deploy them widely will depend on reliable, scalable electricity generation.
Why It Matters Now
Mr. Altman’s remarks come amid a global race to secure semiconductor supply chains and build hyperscale data centers. Utilities in parts of the United States and Europe have reported record requests for new grid connections from technology companies constructing A.I.-optimized facilities. In some regions, regulators have raised concerns about transmission bottlenecks and grid stability.
The challenge is not only total generation but reliability. A.I. training clusters require uninterrupted, high-density power. While renewable sources such as wind and solar are expanding rapidly, their variability creates integration challenges unless paired with storage or firm generation.
At Davos, Mr. Altman suggested that long-term solutions could include advanced nuclear technologies and other forms of high-output clean energy. He has previously invested in energy ventures, arguing that abundant power is foundational to unlocking A.I.’s economic potential.
How Analysts Interpreted the Message
Energy and technology analysts see Mr. Altman’s comments as part caution, part strategic positioning. By highlighting energy as the next bottleneck, A.I. leaders may be signaling to governments that infrastructure policy is now inseparable from digital competitiveness.
Efficiency improvements are underway. New generations of A.I. accelerators deliver greater performance per watt, and researchers are developing smaller, more specialized models that reduce computational load. Techniques such as quantization and sparsity can lower inference costs significantly.
Yet many experts warn of a rebound effect: as computing becomes more efficient and less expensive, overall usage tends to increase. In that scenario, total energy consumption may continue rising even as individual tasks become more efficient.
Pushing Back on Water and Per-Query Claims
Days after Davos – last week, Mr. Altman addressed a related set of concerns while speaking at an event hosted by The Indian Express during his visit to India for a major A.I. summit.
There, he pushed back forcefully against viral claims about A.I.’s environmental footprint. Concerns about the water usage of systems like ChatGPT are “totally fake,” he said, acknowledging that water consumption had once been higher “when we used to do evaporative cooling in data centers.”
Now that we don’t do that, you see these things on the internet where, ‘Don’t use ChatGPT, it’s 17 gallons of water for each query’ or whatever,
Mr. Altman said.
This is completely untrue, totally insane, no connection to reality.
He also rejected comparisons suggesting that a single ChatGPT query consumes the equivalent of 1.5 iPhone battery charges.
There’s no way it’s anything close to that much,
he said when asked about the figure.
At the same time, Mr. Altman acknowledged that broader concerns about total energy use are legitimate. It is “fair” to worry about
the energy consumption — not per query, but in total, because the world is now using so much AI,
he said. In his view, that reality strengthens the case for accelerating investment in nuclear, wind and solar power.
The world needs to move towards nuclear or wind and solar very quickly,
he added.
The Data and the Debate
There is currently no global legal requirement for technology companies to disclose detailed energy and water consumption figures for specific A.I. workloads, leaving researchers to estimate impacts independently. Some academic studies have linked large data center clusters to localized increases in electricity demand and, in certain cases, higher wholesale power prices.
The International Energy Agency estimates that global data center electricity consumption could exceed 1,000 terawatt-hours annually by the end of the decade, roughly equivalent to Japan’s total electricity use today. A.I.-specific workloads are projected to account for a rising share of that total.
Mr. Altman has argued that public discussions often frame the issue unfairly, particularly when they compare the energy required to train a large A.I. model with the energy needed for a single human task.
Many discussions about ChatGPT’s energy usage are unfair,
he said in India, especially when they focus on
how much energy it takes to train an AI model, relative to how much it costs a human to do one inference query.
But it also takes a lot of energy to train a human,
he added.
It takes like 20 years of life and all of the food you eat during that time before you get smart. And not only that, it took the very widespread evolution of the 100 billion people that have ever lived and learned not to get eaten by predators and learned how to figure out science and whatever, to produce you.
In his telling, the more appropriate comparison is between a trained A.I. system answering a question and a human doing the same task.
If you ask ChatGPT a question, how much energy does it take once its model is trained to answer that question versus a human?
he said.
And probably, A.I. has already caught up on an energy efficiency basis, measured that way.
What Can Be Done
Policy experts point to several responses. Accelerating grid modernization, streamlining permitting for new transmission lines and investing in next-generation clean energy projects could help absorb the surge in demand. Co-locating data centers with dedicated renewable or nuclear facilities is another approach already being explored by major technology firms.
There is also a growing push for “energy-aware A.I.” — designing models and systems that optimize not only for accuracy and speed, but also for power efficiency.
At Davos, Mr. Altman framed the challenge as an opportunity rather than a deterrent. If artificial intelligence can dramatically increase productivity and scientific discovery, he suggested, then investing in abundant clean energy may be one of the most consequential economic decisions of the decade.
In tying A.I.’s trajectory to the electrical grid, Mr. Altman underscored a broader reality: the digital revolution now rests squarely on physical infrastructure. And the next frontier in artificial intelligence may depend less on algorithms than on kilowatts.
Material by Iva Abadjievа
Image: The Indian Express Youtube Channel, “Sam Altman Unfiltered: ChatGPT, AI Risks & What’s Coming Next, 40 Questions in 60 Minutes”
Image: Sam Altman at World Economic Forum from Benedikt von Loebell at Flickr
