Sam Altman almonds became a major talking point after the OpenAI CEO said about 38,000 ChatGPT queries use roughly the same amount of water involved in producing one California almond. Altman made the comparison during the premiere of the Sources podcast with Alex Heath. He also acknowledged that he was recalling the number from memory and that the precise figure could be different.
The unusual comparison immediately drew attention because it connects two very different activities: growing an agricultural product and operating an artificial intelligence service. Altman used the example while challenging claims that ChatGPT and the data centers supporting AI consume extraordinary amounts of water.
The statement has since prompted renewed discussion about how much water individual AI interactions actually require. It has also raised questions about the assumptions behind the calculation and how the comparison fits with figures Altman previously shared.
For readers trying to understand the issue, the most important point is that the 38,000 figure is an approximate claim from Altman, not a universally established scientific benchmark.
What Sam Altman Said About One Almond
Altman’s comparison came while he was discussing criticism surrounding AI infrastructure and water consumption.
He said that roughly 38,000 ChatGPT queries use the same amount of water as producing one almond in California. He stressed that he did not have the exact calculation available during the conversation and was relying on memory.
That qualification is important.
The statement does not mean that every ChatGPT request consumes exactly the same amount of water. AI workloads vary, and the infrastructure behind them can operate under very different conditions.
The comparison instead represents an estimated average intended to show the relatively small water requirement associated with an individual AI query.
Altman’s broader argument was that public conversations about AI water consumption sometimes make individual ChatGPT interactions sound much more resource-intensive than they are.
His almond comparison offered a simple way to communicate that argument.
Why California Almonds Became the Benchmark
California provides an especially relevant setting for the comparison because the state is a major center of almond production.
Almond trees require water throughout their growing cycle. Commercial production in California depends heavily on irrigation because of the state’s climate and agricultural conditions.
Water-footprint estimates for almonds have been widely discussed for years. Researchers and analysts can calculate those requirements in different ways, depending on what forms of water use they include.
That creates an immediate challenge when comparing an almond with an AI query.
The amount of water associated with producing one almond depends on the methodology used. The amount attributed to an AI query also depends on methodology.
A direct comparison can therefore be useful as an illustration, but it should not be treated as a fixed conversion rate.
The 38,000 Figure Has an Important Qualification
Altman’s own wording provides the biggest reason to avoid treating the figure as exact.
He indicated that he was working from memory and did not have the calculation directly in front of him.
That means the claim should be described as an approximate estimate.
It would be inaccurate to present 38,000 as a precise scientific measurement applying to every ChatGPT request.
The distinction is especially important because AI technology continues to change rapidly. Different models require different amounts of computing power. A short question can require far less processing than a complex request involving extended reasoning or multiple computational steps.
Data centers also differ in their cooling systems and operating environments.
Consequently, there is no single water-use number that applies identically to every ChatGPT interaction.
Altman Previously Gave a Different Water Estimate
The latest comparison has attracted additional attention because Altman previously published another estimate for ChatGPT water consumption.
In 2025, he said an average ChatGPT query used approximately 0.000085 gallons of water, which is about 0.32 milliliters.
That earlier estimate has become an important reference point in discussions about the new almond comparison.
Using the previously cited 0.32-milliliter estimate alongside a commonly cited figure of about 3.56 liters for the water associated with producing one California almond produces a result of roughly 11,000 queries per almond.
That is significantly lower than the approximately 38,000 queries Altman cited in the newer discussion.
The difference does not automatically establish that the newer statement is false.
Instead, it demonstrates how strongly the result can depend on the assumptions behind each calculation.
Different water accounting methods can include different parts of the overall resource footprint.
Why AI Water Estimates Vary
Water consumption associated with artificial intelligence is more complicated than simply measuring water flowing through a data-center cooling system.
A data center needs to remove heat generated by computing equipment. The method used to accomplish that can determine how much water the facility consumes directly.
Some facilities use evaporative cooling. Others use systems designed to minimize or eliminate direct water consumption.
Electricity also matters.
Power generation can have its own water footprint. If an analysis includes the water associated with generating the electricity used by a data center, the resulting figure can be much different from an analysis that considers only direct on-site consumption.
This is one reason experts can produce very different estimates without necessarily measuring the exact same thing.
The location of a data center also matters. Climate, electricity sources, cooling design and local water availability can all influence the final calculation.
Not Every ChatGPT Query Uses the Same Resources
The phrase “one ChatGPT query” can create the impression that every request places an identical load on computing infrastructure.
That is not how AI systems operate.
A short factual question can require a relatively modest amount of processing. A longer request can demand considerably more computation.
More advanced AI systems can also perform additional processing before producing an answer.
The workload can change depending on the model and task.
This matters when interpreting the 38,000 figure because an average estimate cannot describe every individual interaction.
It is better understood as a broad comparison rather than a promise that any specific user’s request will consume a particular quantity of water.
What the Almond Comparison Does and Does Not Prove
The comparison does provide a useful perspective on the scale of an individual interaction.
If Altman’s estimate is close to the actual average for the workload he had in mind, the direct water requirement associated with a single request would be extremely small.
However, that does not mean AI infrastructure has no water footprint.
The number of queries matters.
A tiny amount of resource consumption multiplied across a massive user base can become significant. The same principle applies to many agricultural and consumer products.
For that reason, examining only one query does not provide a complete picture of AI’s overall water demand.
The total consumption of data centers, the number of facilities, their cooling technologies and their locations all remain relevant.
Altman’s Broader Argument About Data Centers
The almond comparison formed part of a larger argument from Altman about how people perceive AI data centers.
He has repeatedly pushed back against claims that modern data centers universally consume enormous quantities of water.
His position is that some criticism reflects older cooling technologies and does not accurately describe newer facilities.
Altman has argued that modern data centers can use substantially less water than the public may assume.
The reality, however, depends on the facility.
Data centers are not identical. Cooling systems vary, and water consumption can differ substantially between locations and designs.
That means Altman’s argument should not be interpreted as saying every data center in the United States has the same water profile.
Instead, the important issue is the specific infrastructure being examined.
Why the Debate Continues Despite the Small Per-Query Figure
The discussion has continued because AI is being deployed at a scale far beyond a few individual questions.
Millions of people use AI services for writing, research, coding, brainstorming and other tasks.
As usage grows, data-center operators need more computing infrastructure.
That creates questions about the resources needed to support the overall system.
A small water requirement per request can therefore coexist with a large aggregate resource requirement.
This distinction explains why both sides of the debate can focus on different numbers while discussing the same technology.
One side may focus on the tiny amount associated with an individual request.
Another may focus on the total resources required by large data centers operating continuously.
Both measurements can matter.
Why the Exact Water Accounting Method Matters
The phrase “water used” can mean different things in environmental calculations.
Direct consumption is one measurement.
A broader calculation may also account for water associated with electricity generation and other parts of the infrastructure.
Agriculture presents a similar challenge.
A calculation for an almond can focus on irrigation requirements or incorporate a broader water-footprint methodology.
The resulting figures may differ substantially.
Therefore, comparing the two industries requires consistent accounting.
If one side includes indirect water consumption while the other uses only direct consumption, the comparison can become misleading.
This is one of the biggest reasons readers should be cautious about interpreting viral numbers without understanding the underlying methodology.
The Difference Between an Almond and a ChatGPT Query
An almond is a physical agricultural product. Its water requirement comes primarily from the resources needed to grow and harvest the crop.
A ChatGPT query is a digital interaction supported by physical computing infrastructure.
The comparison is therefore not a direct comparison between identical activities.
Instead, it is a way of expressing resource consumption in familiar terms.
Altman used the almond as a benchmark because the water requirements associated with California almond production have received considerable public attention.
The comparison makes a technical discussion easier to understand.
But it should not obscure the differences between agriculture and computing.
The two activities have different production systems, environmental footprints and measurement methods.
What the Latest Information Actually Establishes
Several facts can be separated from the speculation surrounding the viral claim.
First, Altman did make the comparison publicly.
Second, he put the figure at approximately 38,000 ChatGPT queries for the water associated with producing one California almond.
Third, he acknowledged that he was recalling the number rather than reading from a calculation in front of him.
Fourth, Altman previously estimated average ChatGPT water use at about 0.32 milliliters per query.
Fifth, applying that earlier figure to commonly cited almond-water estimates produces a materially different ratio.
These facts explain why the latest claim has attracted scrutiny without requiring either side to dismiss the entire comparison.
What U.S. Readers Should Take Away
For American readers, the most useful takeaway is that the environmental cost of AI cannot be reduced to one viral number.
The amount of water associated with an AI interaction depends on several variables.
These include the model being used, the computing workload, the data center, its cooling system and the method used to account for water consumption.
The almond comparison highlights one important point: an individual AI query does not necessarily consume the enormous quantity of water sometimes claimed in viral discussions.
At the same time, the existence of a small per-query estimate does not eliminate legitimate questions about the total resource demand created by rapidly expanding AI infrastructure.
Those two ideas can exist together.
Why the 38,000 Number Is Still Important
The exact number may remain debated, but the statement has significance because it puts the discussion into a form that millions of people can understand.
Instead of discussing data-center cooling systems, electricity generation and water accounting alone, the comparison asks readers to think about something familiar.
How much water does it take to produce one almond?
How much water does an AI interaction require?
The resulting comparison provides a starting point for a much broader conversation.
It also highlights why transparent methodology matters.
If technology companies publish detailed information about water use, researchers and the public can better understand the actual environmental footprint of AI.
Greater transparency would make it easier to compare different models, facilities and cooling systems using consistent standards.
The Bigger Issue Behind the Almond Debate
The discussion ultimately goes beyond one California almond.
AI infrastructure is becoming an increasingly important part of the U.S. technology landscape. As computing demand expands, questions about electricity, cooling and resource consumption will remain important.
Altman’s latest statement challenges the idea that every ChatGPT interaction carries a massive water burden.
But the 38,000 figure should remain in context.
It is an approximate comparison, and Altman himself indicated that he was recalling the figure rather than presenting a verified calculation during the podcast.
The most responsible interpretation is therefore neither to dismiss AI’s resource footprint nor to exaggerate it.
Instead, readers should look at how the numbers were calculated, what they include and whether the same methodology applies to both sides of a comparison.
The viral almond comparison has made that complicated issue much easier to discuss. It has also shown why precise definitions matter when measuring the environmental footprint of emerging technology.
The conversation around Sam Altman and almonds is far from settled, so share your thoughts on the comparison and stay informed as new verified figures emerge.