OpenAI CEO Sam Altman mentioned it’s truthful to match the power required for ChatGPT to reply a question with the power a 20-year-old human has consumed over a lifetime to answer the identical question, suggesting that the AI chatbot could possibly be thought of extra power environment friendly in that manner.
Talking as a visitor at Specific Adda held on February 20, on the sidelines of the India-AI Influence Summit 2026, Altman mentioned that many discussions round ChatGPT’s power utilization have been unfair as a result of they have an inclination to deal with “how a lot power it takes to coach an AI mannequin, relative to how a lot it prices a human to do one inference question.”
“Nevertheless it additionally takes lots of power to coach a human. It takes like 20 years of life and all the meals you eat throughout that point earlier than you get sensible. And never solely that, it took the very widespread evolution of the 100 billion those who have ever lived and realized to not get eaten by predators and realized how to determine science and no matter, to provide you,” he mentioned.
Within the view of the OpenAI chief, a good comparability can be: “When you ask ChatGPT a query, how a lot power does it take as soon as its mannequin is educated to reply that query versus a human? And possibly, AI has already caught up on an power effectivity foundation, measured that manner.”
Altman was responding to a query posed by Anant Goenka, Govt Director of The Indian Specific Group, who cited a earlier interview with Invoice Gates and requested whether or not it’s correct to say a single ChatGPT question at present makes use of the equal of 1.5 iPhone battery fees.
“There’s no manner it’s something near that a lot,” Altman replied. He additionally mentioned that synthetic normal intelligence (AGI) — a hypothetical state of technological maturity wherein an AI system turns into able to performing each activity {that a} human being can, with better precision, effectivity and velocity — feels “fairly shut at this level.” “Given what I do know to be a quicker takeoff, I count on [artificial] superintelligence isn’t that far off,” Altman mentioned.
You’ll be able to watch Anant Goenka’s full interview with Sam Altman at Specific Adda by clicking on the hyperlink under.
Story continues under this advert
The large build-out of AI infrastructure, together with giant knowledge centres, has come below scrutiny due to the huge assets they eat, notably power which has additionally been related to rising electrical energy costs. In January this month, the Trump administration and several other US state governors signed a pact that requires tech firms to pay for brand new energy vegetation constructed on the PJM electrical energy grid within the nation and is used to energy knowledge centres which can be, in flip, used to coach and energy AI fashions.
Altman’s remarks additionally come at a time when India is positioning itself to be the information centre hub of the world, with funding commitments exceeding US $200 billion made within the nation, a majority of which is devoted to constructing out AI infrastructure right here over the subsequent decade or so.
AI’s water utilization ‘completely pretend’: Altman
In response to a query concerning the quantity of water going into knowledge centres housing GPU server racks that energy AI fashions, Altman prompt that such issues have been “completely pretend” as a result of “we used to do evaporative cooling in knowledge centres.”
Story continues under this advert
“Now that we don’t try this, you see these items on the web the place, ‘Don’t use ChatGPT, it’s 17 gallons of water for every question’ or no matter. That is utterly unfaithful, completely insane, no connection to actuality,” he additional mentioned.
Nonetheless, he acknowledged that it was truthful to be involved about “the power consumption — not per question, however in whole, as a result of the world is now utilizing a lot AI is actual, and we have to transfer in direction of nuclear or wind and photo voltaic [energy] in a short time.”
Apparently, Altman rejected the idea of space-based knowledge centres. “Placing knowledge centres in area with the present panorama is ridiculous. Orbital knowledge centres should not going to matter at scale this decade as a result of tough math of launch prices and the way onerous it’s to repair a damaged GPU in area,” Altman advised Goenka.
A rising variety of tech firms, together with SpaceX, owned by Altman’s archrival Elon Musk, desires to indicate how outer area could be a hospitable setting for knowledge centres compared to monumental multi-gigawatt terrestrial amenities that eat hundreds of thousands of litres of water every day and produce substantial quantity of greenhouse fuel emissions.
Story continues under this advert
Altman’s clip on the power utilization of AI additionally went viral on social media and triggered numerous reactions on-line:
I construct AI for a dwelling. I imagine in what we’re constructing. However this type of rhetoric makes my work tougher and extra harmful.@sama, evaluating human improvement to mannequin coaching is tone-deaf, strategically reckless.
Persons are shedding jobs. They’re getting indignant. They’re… https://t.co/DgiaPEABm1
— Muratcan Koylan (@koylanai) February 21, 2026
I don’t get it: aren’t you constructing a device to make human lives simpler?
What kind of an argument equates an ai mannequin to a human being, evaluating a device to an finish person? Or are you imagining a world the place it’s the opposite manner round?
For years, entrepreneurs have solid… https://t.co/Xm0RHeFgN6
— Raunaq Mangottil (@RaunaqMangottil) February 22, 2026
Sam Altman simply mentioned coaching AI is like coaching a human as a result of “it takes 20 years of life and all of the meals you eat earlier than you get sensible.”
Let’s run the precise numbers he’s hoping you’ll not run.
Coaching a human for 20 years prices roughly 17 megawatt-hours of meals power.… https://t.co/scaKGSS9bq pic.twitter.com/JCk9viDseI
— Shanaka Anslem Perera ⚡ (@shanaka86) February 21, 2026
A human consumes about 2,000 energy per day. Over 20 years, that’s roughly 17,000 kWh of whole meals power. Coaching GPT-4 consumed an estimated 50 GWh of electrical energy. That’s 3,000 people value of “coaching power” for a single mannequin run.
And GPT-4 is already useless. OpenAI… https://t.co/C7KojpAIqA
— Aakash Gupta (@aakashgupta) February 21, 2026