OpenAI CEO Sam Altman spoke out on AI water use and energy demands Friday at an event in India. He called worries about water consumption totally fake. Altman pushed back hard against online claims. This came amid rising talk about data centers' environmental toll. The discussion happened at the Express Addda event with The Indian Express.
Key Takeaways
- Sam Altman labeled AI water use claims as "totally fake" since data centers no longer rely on evaporative cooling.
- He agreed total energy use for AI is a real issue but called for quick shifts to nuclear, wind, and solar power.
- Altman compared AI's energy per query to a human's, saying AI has already caught up or done better.
- The remarks drew quick backlash online, with critics calling them out of touch.
Background
AI data centers have faced heat lately. People worry they guzzle water and power. Reports show big tech firms like Google and Microsoft ramped up water pulls by double digits last year. One study from California researchers pegged AI chat queries at up to 2 liters of water for just 10 to 50 asks. That's way more than old guesses. In places like the UK, new centers might drink as much as a whole city. Ireland's data hubs already eat 21% of the nation's electricity.
But Altman says the story's changed. Data centers ditched old cooling methods that evaporated water. Now they use other ways to stay cool. Online posts still spread old stats. Things like "17 gallons per ChatGPT query." Altman shot that down flat. He made these points while in India for an AI summit. The country eyes big AI growth. Altman has pitched huge data center builds to the White House too. Five giant ones, with massive power needs.
And energy? That's real. Training AI models burns a ton. But once trained, answering a question takes little juice. Altman noted humans use energy too. Think about what your brain does for one thought. Or how much power goes into making jeans – 7,500 liters of water. Hundreds of thousands of AI prompts use less than that pair. It's not visible. Most water footprints hide off-site.
Critics push back. They say local strain matters. Data centers cluster near sensitive spots. Almost 70% sit by protected areas that need clean water. Demand could outstrip supply by 40% this decade. Tech firms promise offsets by 2030. But in dry summers, peaks hit public systems hard. One professor noted AI could help here too. It spots leaks, cuts waste in pipes. A UK firm already slashed leaks with it.
Key Details
Altman laid it out plain.
"Water is totally fake. It used to be true, we used to do evaporative cooling in data centers, but now that we don’t do that… This is completely untrue, totally insane, no connection to reality."
He turned to power next. Total use matters, not per ask. The world runs tons of AI now. Solution? Nuclear. Wind. Solar. Fast. Bill Gates floated an idea once. AI learns from human evolution to get leaner on energy. Altman agreed partly. But the real compare? Trained AI query versus human brain tick. AI wins or ties there already.
Online Reaction
X lit up quick. One user called it dystopian. Researcher Matt Stoller tweeted: a spreadsheet versus a baby? Morally equal? Memes flew. Jabs at Altman's own "training" energy. Some said one human second beats all AI. Backlash paints tech leaders as anti-human. Altman stands firm. OpenAI shared stats before. Average query: 0.34 watt-hours. Tiny.
But numbers stack up elsewhere. Google pulled 6.1 billion gallons fresh water in 2024. Power? 680,000 kWh a day for prompts alone, some reports claim. Microsoft saw 22.5% water jump. Meta and Google around 17%. OpenAI stays quiet on totals. Altman wants transparency on prompts. Firms need to share model needs.
India fits in. Altman's visit ties to local AI push. Quantonation Closes €220M Fund, Doubles Down on Quantum Tech shows funding flows into related tech. Water stress hits here too. Monsoon fails amplify it. Data centers could strain more if they boom.
What This Means
AI water use debates won't fade. Altman's words shift focus to energy fixes. Nuclear plants take years. Wind and solar scale uneven. Data centers keep growing. AI touches everything now. Phones. Cars. Factories. Flow App Boosts Android Voice Typing With AI Tools proves daily creep. Efficiency gains help. But total load rises with use.
Communities watch close. Water bans in dry spells? Possible. Thames Water eyes limits. Regulators push disclosures. UK groups call for real footprint data. AI might save water elsewhere. Leak hunts. Smart grids. But addictive models drive constant queries. Rivals build too. More power chase.
Investors note. AI stocks ride high. But green pushback grows. Trump tariffs didn't shake markets, per Trump’s New Tariffs Fail to Shake Steady Markets. Energy costs could. Altman bets on clean breakthroughs. World weighs if it's enough. Data hubs plan expansions. White House listens. Local voices demand say.
Short term? Scrutiny ramps. Reports multiply. Firms report more. Long term? Tech adapts. Or faces caps. Balance hangs on facts versus claims.
Frequently Asked Questions
How much water does one ChatGPT query really use?
Altman says less than 1/15 teaspoon on average. But studies vary. 10-50 queries hit up to 2 liters total, due to cooling.
Why focus on total energy, not per query?
Per query, AI matches human brain use post-training. Total matters as billions query daily. That's the real strain.
Can AI cut its own water use?
Yes. It already spots leaks, optimizes distribution. Full rollout could trim 15% in some systems.
Frequently Asked Questions
How much water does one ChatGPT query really use?
Altman says less than 1/15 teaspoon on average. But studies vary. 10-50 queries hit up to 2 liters total, due to cooling.
Why focus on total energy, not per query?
Per query, AI matches human brain use post-training. Total matters as billions query daily. That’s the real strain.
Can AI cut its own water use?
Yes. It already spots leaks, optimizes distribution. Full rollout could trim 15% in some systems.
