Sam Altman disputes viral claims about AI’s water use

Your chatbot has a footprint

AI feels invisible when it shows up as a quick answer on your phone. But behind that fast reply is a row of servers, cooling equipment, and a lot of electricity. That is why questions about AI’s water and power use are getting louder across America.

Sam Altman stepped right into that debate during the India AI Impact Summit in February 2026. He pushed back hard on viral claims about water use per prompt.

He also said energy demand is a legitimate concern and argued that the world needs more nuclear, wind, and solar power quickly.

Altman called some claims fake

Altman did not deny that AI uses resources. What he challenged was the dramatic idea that every single prompt gulps down huge amounts of water. He called those viral claims “completely untrue” and “totally insane.”

That is an important distinction for readers from Texas to Ohio. A system can still use a lot of water overall, even if the “gallons per query” headlines are overstated. The real picture depends on how the data center is cooled, powered, and supplied.

His human comparison stood out

Altman also made a line that got laughs from the crowd. He said it takes a lot of energy to “train a human” too, pointing to years of food and life before a person becomes knowledgeable. That comparison quickly became one of the most talked-about parts of the event.

His bigger point was about efficiency after training. He argued that once an AI model is built, the energy required to answer a single query may compare favorably with that of a human performing the same task. That is a provocative claim, but it differs from saying that AI has no environmental cost.

OpenAI shared its own number

In June 2025, Altman wrote that the average ChatGPT query uses about 0.34 watt-hours of electricity. He compared that to what an oven uses in a little over one second. He also said the average query uses about 0.000085 gallons of water.

Those figures are useful, but they are also narrow. They came before newer model releases and do not mean every task uses the same amount of power. A simple question and a complex image request are not the same kind of workload.

The bigger issue is scale

One chatbot request may look small on paper. The challenge arises when millions of people use AI tools every day, and companies expand data center capacity to keep up.

In Texas, a major AI data-center project is being built outside Abilene as part of “Stargate,” with multiple buildings planned.

At that scale, small per-query averages can turn into great total demands—so many analyses focus on overall infrastructure growth rather than a single prompt.

Cooling technology changes the story

How a data center is cooled can dramatically change its water footprint. Some facilities use evaporative cooling, which relies on water to remove heat.

Other designs use closed-loop, non-evaporative liquid cooling that recirculates coolant and can reduce ongoing water consumption—though systems may still need makeup water for maintenance over time.

Oracle says the AI data centers it is building use direct-to-chip, closed-loop, non-evaporative cooling.

Water use still adds up

Even with more efficient cooling, overall AI-driven water demand could still rise as infrastructure scales.

New research from Xylem and Global Water Intelligence (published January 14, 2026) estimates that AI could add 30 trillion liters to annual water demand by 2050, nearly a 130% increase across the AI value chain.

The estimate is intended to cover impacts beyond a single server room, including direct data-center needs and broader upstream effects tied to powering and supporting AI infrastructure.

Power and water are linked

A lot of AI’s water footprint does not happen at the data center itself. It can also happen at power plants that generate the electricity those centers need. So when electricity demand rises, water demand can rise with it.

That is one reason Altman treated electricity as the fairer criticism. If AI keeps scaling, the country will need more power and cleaner power simultaneously. That helps explain why he pointed to nuclear, wind, and solar rather than pretending energy is a side issue.

Texas is part of the picture

Texas has become a major focal point in the data-center water debate. The Texas Tribune reported that a large AI data-center project outside Abilene would initially use 8 million gallons of city water to fill its cooling system.

Abilene water officials told the Tribune that this initial fill is small compared with the city’s average daily water use of about 22 million gallons, but they also said the system may need additional water over time and could require draining and refilling for maintenance.

Communities are watching closely

This is no longer just a boardroom issue. Local officials and residents in several U.S. communities have begun asking harder questions about water access, grid strain, and land-use tied to data center growth. AI infrastructure is becoming a neighborhood issue as much as a tech issue.

That makes the public conversation more practical and less abstract. People want to know whether a project will create jobs, strain utilities, or increase local pressure on water systems. Those concerns are likely to keep growing as more sites are proposed.

Tech companies see the pressure

Big tech companies know this scrutiny is not going away. Microsoft has maintained its goal to become water-positive by 2030. That means it aims to replenish more water than it uses.

That kind of target matters because AI is making the issue harder, not easier. Companies now have to prove they can expand computing without looking careless about local resources. For readers in drought-prone states, that balance feels especially important.

Per prompt is the wrong lens

The viral “one prompt equals huge water waste” framing is catchy, but it can miss the bigger truth. AI’s impact depends on total demand, hardware design, cooling systems, and power generation. A single number rarely captures all of that.

That is why this debate gets messy online. One side can cherry-pick the smallest average, while the other can pick the scariest total. The more honest view is that some claims are exaggerated, but the overall resource challenge is still real.

Opposition is growing over data center and gas power plans near Dinosaur Valley. Check out why this fight is raising bigger questions about growth, energy, and Texas communities.

America will need better answers

The next phase of this story is not really about one quote from Altman. It is about whether the U.S. can build an AI infrastructure that is smarter about power, water, and local trust. Faster chips and bigger campuses will keep raising the stakes.

That means better cooling systems, cleaner grids, and more transparency from companies. It also means local communities will want clearer answers before projects break ground. AI may be digital, but its footprint is physical.

Trump has blacklisted an AI company he blasted as a “left-wing nutjob” after it drew a line on weapons and surveillance. Check out what this clash could mean for the future of AI policy in America.

Do you think AI companies are being clear enough about what their growth will require? Share your thoughts in the comments.

This slideshow was made with AI assistance and human editing.

Read More From This Brand:

• Texas Wants More Patriotism, Minimal World History In Sixth-Grade Social Studies
• The massacre site Texas tried to erase from history is now marked on Highway 90
• California’s new push could make extreme highway speeding a lot riskier