Sam Altman has dropped a seriously unexpected comparison into the already heated debate over AI’s environmental impact: around 38,000 ChatGPT queries use the same amount of water as producing one almond in California. The OpenAI CEO made the claim during Sources with Alex Heath, released September 1, 2026, while pushing back against what he described as exaggerated claims about how much water AI systems and their data centers consume.
The comparison sounds simple, but there is a major catch. While California almonds have a heavily studied water footprint, researchers say there is not enough publicly available information about individual AI data centers, their cooling systems, locations and water sources to independently verify Altman’s 38,000-query calculation. So, for now, the viral number is Altman’s estimate — not an established scientific measurement.
More AI News
Gemini’s Big Moment: Google just dodged a major legal breakup and dropped its newest model in the same week. For more, read our story on Google’s Gemini 3.8 Flash launch and the AdX lawsuit win.
Sam Altman’s 38,000 ChatGPT Queries vs One Almond Claim Explained
During the podcast, Altman addressed claims circulating online that a single ChatGPT query consumes an enormous amount of water.
“You know, I saw this thing going around about the water usage of ChatGPT. And it was like every time you run a single ChatGPT query, it’s like, you know, you run your shower for like 6 hours and the water never comes back and it’s just done. I don’t have the exact calculation in front of me, but I think the real number is something like, doing this from memory, it might be wrong, but it’s close — for every 38,000 ChatGPT queries, that is the same amount of water that is used in the production of a single almond in California, which is like, really, and this is like the full-on, you know, total true water accounting, not just what’s running in one data center.”
Altman then questioned why almonds do not attract the same level of public concern despite their water requirements.
“There’s a question of like where this came from, because the people that are scarfing down 12 almonds at a time don’t feel like they’re doing something horrible from a water perspective, for the most part.”
He also argued that some descriptions of data-center water consumption are based on older cooling technologies.
“It is true that data centers at one point used evaporative cooling. But they have not done that in a long time. Like, if you look at a modern very large data center, it uses the equivalent amount of water as an office building in terms of you know people like running the sinks and the toilets and whatever.”
Altman ultimately dismissed the broader narrative as something that does not stand up to scrutiny.
“So, that has been a robust meme and difficult to disprove, but I don’t think it holds up to any scrutiny.”
There is one detail in those comments that matters a lot: Altman himself said he was recalling the 38,000 figure from memory and did not have the exact calculation in front of him. That qualification becomes increasingly important once researchers begin looking at what can actually be verified.
More Breaking News
Emergency in the Skies: A routine departure turned into a mid-air scare when pilot incapacitation forced an unexpected U-turn. For more, read our report on the Delta Flight 1208 emergency return to Minneapolis.
Why Researchers Cannot Independently Verify the 38,000-Query Water Estimate
The problem is not that researchers have zero information about AI water consumption. It is that water use can change dramatically depending on how, where and when computing takes place.
Shaolei Ren, a professor of electrical and computer engineering at the University of California, Riverside, told CalMatters:
“The information we have is so limited.”
Ren explained that a number of factors can influence the water associated with an AI request, including the location of a data center, outdoor temperature, cooling technology, prompt length, amount of reasoning required and length of the resulting output.
That means there is no obvious universal water number that can automatically be attached to every ChatGPT query. A short request handled in one facility could have a different water footprint from a computationally intensive prompt processed somewhere else.
Michael Kiparsky, director of the Wheeler Water Institute at the UC Berkeley Center for Law, Energy, & the Environment, pointed to the same transparency problem.
“This is the whole reason that these bills need the governor’s signature,”
Kiparsky said.
“There’s no published data — certainly not for California — and what is public seems all over the place.”
And that is where the almond comparison starts becoming much more complicated.
More Environmental News
Melting Under Pressure: It’s not just data centers straining natural resources — even the world’s tallest peak is heating up from human traffic. For more, read our piece on Everest Base Camp’s new glacier heat warning.
How Much Water Does One California Almond Actually Use?
The almond side of Altman’s comparison is based on a real and extensively researched environmental issue.
A peer-reviewed 2019 study published in Ecological Indicators estimated that the water footprint of California almonds averaged 10,240 liters per kilogram of almond kernels, or approximately 12 liters (3.2 gallons) per almond kernel, during the period studied.
The research also found significant variation depending on location and year.
So, the claim that almonds require substantial amounts of water is not fabricated. But there is an important distinction: water-footprint calculations depend on methodology and exactly what types of water use are included.
That matters when an agricultural product is compared with artificial intelligence.
Agricultural water-footprint calculations and data-center water-use calculations do not necessarily measure identical categories of consumption. A comparison can therefore sound extremely precise while still depending on different accounting boundaries.
The almond study itself emphasized substantial spatial and temporal variation in water footprints across California.
More Trending Stories
A Fight That Won’t Fade: Public disputes don’t always stay contained to tech — legal battles over reputation and truth are heating up elsewhere too. For more, read our coverage of Al Sharpton’s defamation standoff over the Nolan Wells case.
Sam Altman’s Earlier ChatGPT Water Estimate Raises More Questions
There is another number that makes the new comparison even more interesting.
In his 2025 essay The Gentle Singularity, Altman previously wrote that an average ChatGPT query used approximately 0.000085 gallons of water, equivalent to roughly 0.32 milliliters.
That earlier estimate does not automatically produce the new 38,000-query ratio when combined with the 2019 academic estimate of approximately 3.2 gallons per almond.
That does not, by itself, prove that either number is false.
Different calculations can measure different things and use different system boundaries. But the difference does highlight a key unanswered question: what exactly did Altman include in his latest “full-on” water accounting?
To independently validate the 38,000 figure, researchers would need to know which facilities were included, what cooling systems were considered, what water sources were counted and how the almond comparison was constructed.
Until that methodology is available, the figure remains difficult to independently confirm.
More Political News
A Fight Over Balance of Power: Just as data-center oversight is being debated state by state, Washington is fighting its own institutional battle. For more, read our article on the House rejecting the bid to lock the Supreme Court at nine justices.
AI Data Center Water Consumption Is Still Growing
Here is where the story gets more complicated — because questioning the 38,000-query comparison does not mean AI has no water footprint.
Research cited by CalMatters estimates that direct U.S. data-center water consumption reached approximately 17 billion gallons in 2023, up from about 5.6 billion gallons in 2014.
The Congressional Research Service has also noted that data centers are using more water than they did previously.
Another estimate cited in the reporting projects that hyperscale data centers could use approximately 150 billion gallons of water between 2025 and 2030.
These numbers describe the data-center sector as infrastructure rather than the water footprint of one individual ChatGPT prompt.
That distinction is crucial.
A query-level calculation can help explain the marginal resources associated with computing. But it cannot tell the whole story of what happens when enormous facilities are constructed and operated at massive scale.
And there is another factor that researchers say deserves attention: when the water is needed.
More Travel News
Crackdowns and Capacity Limits: Managing demand isn’t just a data-center problem — even ancient landmarks are tightening control over foot traffic. For more, read our post on Rome’s Colosseum becoming a security ‘red zone’ for 2026 tourists.
AI Data Center Peak Water Demand Could Be the Bigger Problem
Annual consumption figures do not necessarily reveal the full pressure that data centers can place on local water systems.
According to research by Ren, the hottest periods of the year can create particularly large bursts of cooling demand.
That matters because water utilities cannot simply plan around an annual average. They must maintain enough capacity to handle periods of peak demand.
Ren explained:
“For community water systems, the peak capacity is more important because they need to be designed and built to accommodate the peak demand at all times, similar to our power grid,”
A UC Riverside report on the research estimated that data-center cooling systems could require between 697 million and 1.45 billion additional gallons of peak water capacity per day in coming years without greater water efficiencies.
Researchers compared that potential demand with New York City’s typical daily water supply.
There is also a significant difference in the type of water involved.
Data centers may use treated, potable water from public water systems, while agricultural irrigation generally relies on different water sources and infrastructure.
So while an almond and an AI query can both be described in terms of water, the consequences for communities are not necessarily comparable.
More Viral News
Caught on Camera: Not every shocking video involves technology — sometimes it’s what shows up right at the front door. For more, read our story on the naked DoorDash driver caught on Ring camera in Houston.
New Cooling Technology Could Reduce AI’s Direct Water Use
Altman’s argument that cooling technology has changed is not without evidence.
Data-center cooling systems have evolved, and newer approaches can substantially reduce direct water consumption. Ren’s research indicates that switching to less water-intensive systems, including dry cooling, could reduce water use by as much as 50% in some circumstances.
But there is a trade-off.
Less water-intensive cooling systems can require more energy. Electricity generation can itself have a water footprint, meaning that reducing water use inside a data center does not necessarily eliminate the wider water implications of the computing infrastructure.
The emerging picture is therefore much more complicated than either extreme.
It would be misleading to claim that every AI query consumes an enormous amount of water. But it would also be misleading to dismiss concerns about AI infrastructure simply because modern cooling systems can reduce direct water consumption.
More Sports News
When Paperwork Becomes the Real Opponent: Even elite athletes aren’t immune to logistical disasters derailing their biggest moments. For more, read our report on Li Yueru’s passport nightmare ahead of China’s FIBA World Cup clash with the USA.
California Data Center Water Transparency Laws Could Change the Debate
The argument over Altman’s calculation arrives at a particularly relevant moment in California, where lawmakers are considering stronger disclosure requirements for data centers.
Assemblymember Diane Papan introduced AB 2469 and AB 2619, two measures addressing data-center water use.
| Bill | Main focus | Status by end of August |
| AB 2469 | Restrictions and water-related requirements for new or expanded data centers that increase peak water use | Passed both legislative chambers; Senate amendments concurred in; moving toward final engrossing and enrolling |
| AB 2619 | Disclosure of data-center water use through existing business licensing procedures | Passed the Senate; Assembly concurred in Senate amendments; sent for engrossing and enrolling |
AB 2469 would restrict local governments from approving new or expanded data centers that increase peak water use unless specified water-related requirements are met.
Those requirements include information about water supply, water-use assessments and water-scarcity planning. The legislation also addresses the costs of water infrastructure needed to serve a project.
AB 2619 takes a different route, focusing on disclosure of data-center water use through existing business licensing procedures.
The relevance to the Altman controversy is pretty direct: researchers say the public does not currently have enough detailed information to independently evaluate many data-center water claims.
Papan has argued that local officials need reliable information before approving data-center projects. In an official statement, she said:
“These bills will allow locals to make informed decisions about data centers, ensuring they do not strain local water resources or subject ratepayers to increased costs.”
More Gaming News
Controversy Isn’t Just for Tech CEOs: Public backlash cycles are everywhere right now, including inside the world’s most anticipated video game. For more, read our piece on the GTA 6 romance system sparking online girlfriend drama.
Gavin Newsom Previously Vetoed Data Center Water Disclosure Legislation
California’s push for greater transparency did not begin in 2026.
Papan previously authored legislation addressing data-center water use, but Gov. Gavin Newsom vetoed that earlier measure in 2025.
At the time, Newsom said he was:
“reluctant to impose rigid reporting requirements … without understanding the full impact on businesses and the consumers of their technology.”
Since then, the political environment has shifted. Disputes over proposed data centers have emerged in communities around California, while concerns have grown over water supplies, infrastructure and the costs of accommodating massive facilities.
That makes AB 2469 and AB 2619 particularly significant as the state debates how much information communities should receive before AI and cloud-computing projects move forward.
More Court-Side News
When the Spotlight Becomes the Story: Public scrutiny doesn’t stop at press conferences — even athletes are dealing with backlash over how their image is managed. For more, read our story on Naomi Osaka’s US Open match and the influencer ring-light controversy.
Is Sam Altman Right About ChatGPT Water Use?
The most defensible answer right now is neither a simple yes nor a simple no.
The 38,000-query almond comparison has not been independently verified. Altman himself said he was recalling the figure from memory and acknowledged that it might be wrong. Researchers interviewed by CalMatters said publicly available information is insufficient to independently establish the calculation.
California almonds do have a substantial water footprint. The peer-reviewed 2019 study estimated an average of about 12 liters, or 3.2 gallons, per almond kernel, while also documenting significant variation across locations and years.
ChatGPT queries can consume water directly and/or indirectly depending on the accounting method. Altman previously estimated an average query at approximately 0.000085 gallons, or 0.32 milliliters, but that was an average rather than a universal measurement for every query and facility.
Data-center water consumption is a legitimate infrastructure issue. Research indicates that total data-center water consumption has increased, while peak demand can create additional challenges for local water systems.
And perhaps the biggest takeaway is that there is no single water number that accurately represents every AI prompt.
Location, weather, cooling technology, computing intensity, output length and the accounting boundary can all affect the result.
The Real AI Water Question Goes Far Beyond One Almond
The almond analogy works because it turns an invisible technological footprint into something almost everyone understands.
But the bigger question is not whether someone should feel guilty for asking ChatGPT a question or eating an almond.
It is what happens when millions or billions of AI queries are processed through rapidly expanding data-center infrastructure.
Communities need to know how much water those facilities require, when they require it, where that water comes from and who pays for the infrastructure needed to supply it.
That is where the strongest evidence currently sits.
Researchers are not saying that every ChatGPT prompt is an enormous water drain. They are pointing to a different problem: the industry’s overall water demand, especially its peak requirements, remains difficult to assess because detailed public information is often limited.
California’s proposed transparency measures could therefore become an important part of the debate.
For now, Sam Altman’s 38,000-query comparison remains striking, controversial and highly shareable — but it should not be presented as settled science.
Until more detailed data and methodology become publicly available, the most accurate description is also the simplest:
38,000 ChatGPT queries per almond is Sam Altman’s estimate, not a scientifically settled measurement.
More Crime News
When Trust Gets Exploited: Transparency failures aren’t limited to tech giants — sometimes it’s individuals footing the bill for someone else’s lack of accountability. For more, read our coverage of the fake wedding fraud case leaving Charlotte Blackwell’s victims still waiting on repayment.
Disclaimer
This article is based on a thorough review of the sources listed below and distinguishes between independently documented research, reported data and claims made by Sam Altman. Water-use estimates can vary according to methodology, location, cooling technology and which forms of direct and indirect water consumption are included. The 38,000-queries-per-almond comparison has not been independently verified from sufficient public data and is therefore presented as Altman’s estimate rather than an established scientific fact.
Sources
- CalMatters — Fact check: Is Sam Altman right that almonds use more water than ChatGPT queries?
- Sources with Alex Heath — Sam Altman on OpenAI’s next model and the AI backlash
- Sam Altman — The Gentle Singularity
- ScienceDirect — Water-indexed benefits and impacts of California almonds
- UC Riverside — Data center water spikes could cost billions
- California Assemblymember Diane Papan — Data Center Water Use Transparency and Accountability Bills
- California Assembly — AB 2469 Policy Committee Analysis
- California Water Commission — 2026 Data Center Legislation Update
- California Legislature bill information — AB 2469
- California Legislature bill information — AB 2619
Featured Image Credit: TechCrunch / Creative Commons Attribution 2.0
Leave a Reply