Sam Altman’s ChatGPT Water Claim Sparks AI Debate

Artificial intelligence is often discussed in terms of computing power, electricity consumption and carbon emissions. But as AI services become part of everyday life, another environmental concern is attracting growing attention: water consumption.

OpenAI CEO Sam Altman recently offered a striking comparison while discussing the environmental impact of ChatGPT. According to Altman, approximately 38,000 ChatGPT queries use about the same amount of water as producing one California almond.

The comparison was presented as a way to challenge viral claims suggesting that a single ChatGPT query could consume as much water as a long shower. However, the figure has since attracted scrutiny because estimates of AI water consumption depend heavily on how researchers define and measure water use.

The debate highlights a larger issue surrounding AI data center water consumption: there is no single number that accurately represents the water footprint of every AI query.

Sam Altman Says 38,000 ChatGPT Queries Equal One Almond

Altman made the comparison during an appearance on the Sources podcast with journalist Alex Heath in early September 2026.

The OpenAI chief was responding to social media claims about the amount of water required to operate ChatGPT. Some viral posts had suggested that a single AI query could consume an amount of water comparable to a six-hour shower.

Altman argued that such claims significantly exaggerate the water requirements of an individual ChatGPT interaction.

He cited an estimate of approximately 0.000085 gallons of water per ChatGPT query, equivalent to around 0.32 milliliters. Using a water-footprint estimate of approximately 3.2 gallons for an almond, the calculation produces a figure of roughly 38,000 queries.

The arithmetic itself is straightforward. The bigger question is whether the two numbers being compared accurately represent the underlying environmental costs.

Altman also acknowledged that he was recalling the figure from memory, adding another reason for researchers and observers to examine the methodology behind the claim.

Why the ChatGPT Water Estimate Is Being Questioned

The main difficulty with measuring ChatGPT’s water consumption is that different studies can count different categories of water use.

The approximately 0.32-milliliter figure cited by Altman focuses on water associated with data-center operations, particularly direct cooling requirements. That approach can produce a relatively small number for an individual query.

However, AI systems also depend on electricity, and generating that electricity can require water.

Power plants may use water for cooling and other processes. If that indirect water consumption is included, the overall water footprint associated with an AI query can become considerably larger.

This distinction is important because two researchers can produce very different estimates while both using technically valid calculations.

In other words, asking “How much water does one ChatGPT query use?” does not have a simple answer unless the scope of the calculation is clearly defined.

The Almond Comparison Is More Complicated Than It Sounds

Altman’s comparison also depends on the estimated water footprint of an almond.

The figure of approximately 3.2 gallons per almond has been questioned because other commonly cited estimates put the water footprint of an almond closer to one gallon.

If the one-gallon estimate is used alongside Altman’s 0.32-milliliter-per-query figure, the resulting comparison changes significantly. Instead of approximately 38,000 queries, the calculation would produce roughly 11,000 to 12,700 ChatGPT queries per almond.

That difference demonstrates why environmental comparisons can be misleading when the assumptions behind them are not explained.

Agricultural water footprints can also vary depending on location, farming practices, climate conditions and the methodology used to calculate water consumption.

Consequently, comparing AI queries with agricultural products can be useful for creating an understandable reference point, but it should not be treated as a universal measurement of AI’s environmental impact.

AI Water Usage Depends on What You Count

One of the most important distinctions in the AI water debate is between direct and indirect water consumption.

Direct water use generally refers to water consumed at or associated with a data center. Modern data centers can use sophisticated cooling technologies to control the temperature of servers and other equipment.

Indirect water consumption can occur through the electricity supply that powers those facilities.

An AI data center may therefore have a relatively small direct water requirement while still contributing to water consumption elsewhere through electricity generation.

Some independent estimates that include electricity-related water use have placed AI query consumption in the range of several milliliters per query, while earlier research has produced estimates reaching considerably higher levels.

These numbers cannot simply be compared without considering the underlying assumptions.

For consumers, this means that claims such as “one AI query uses X milliliters of water” should always be viewed in context. The number may describe only cooling water, or it may attempt to account for a wider water footprint.

Why AI Data Centers Still Matter at a Local Level

Looking only at the water required for one ChatGPT query can make the environmental impact appear insignificant.

A fraction of a milliliter is extremely small. But AI platforms operate at enormous scale.

OpenAI has said ChatGPT handles approximately 2.5 billion user prompts per day. When billions of queries are processed, even a small amount of water associated with each request can add up to a substantial total.

Using Altman’s estimate of roughly 0.32 milliliters per query, 2.5 billion daily prompts would correspond to approximately 800,000 liters of direct water consumption per day.

That calculation does not include additional water associated with electricity generation.

This is where the conversation shifts from individual users to infrastructure.

A person sending a handful of ChatGPT prompts is unlikely to have a meaningful direct impact on local water supplies. Large AI data centers, however, can place significant demands on regional electricity and water infrastructure, particularly when several facilities operate in the same area.

Local Water Stress Can Be Different From Global Averages

Another problem with focusing exclusively on averages is that water is a local resource.

A data center consuming millions of gallons of water in an area experiencing abundant water supplies presents a different challenge from the same facility operating in a drought-affected region.

Reports about data-center water consumption in places such as Georgia have demonstrated why local conditions matter. Large facilities can consume substantial amounts of water, and their impact can become more noticeable when communities are already dealing with limited supplies.

This means that AI water usage should not be evaluated solely on a per-query basis.

The more relevant questions include where a data center is located, what cooling technology it uses, how much water it consumes, where its electricity comes from and whether the region is experiencing water stress.

Billions of AI Queries Change the Environmental Equation

The rapid growth of generative AI makes scale one of the most important factors in the discussion.

ChatGPT is no longer a niche technology used by a relatively small group of researchers. AI assistants are being integrated into search, education, software development, business operations, customer service and everyday productivity.

At the same time, AI workloads are becoming more complex.

A simple text request may require considerably less computing power than an AI agent performing a series of actions, analyzing large files or generating complex outputs.

As AI agents become more capable, they may execute multiple model calls to complete a single user objective. This makes the idea of assigning one fixed environmental cost to “an AI query” increasingly difficult.

The water footprint of AI could therefore vary depending on the model, workload, data-center infrastructure, cooling system, location and electricity source.

AI Companies Face Growing Pressure to Improve Transparency

The debate surrounding Altman’s statement points to a broader challenge for the technology industry: transparency around AI’s environmental footprint.

Companies developing increasingly powerful AI systems are expected to provide information about their energy and resource consumption, but standardized reporting remains limited.

Without consistent measurement standards, companies can highlight particularly favorable statistics while researchers may use broader accounting methods.

A standardized approach could make it easier to compare different AI models and data centers.

Such reporting could potentially include electricity consumption, direct cooling water, indirect water associated with power generation, data-center location and the type of cooling technology being used.

That information would give users and policymakers a clearer understanding of the environmental cost of AI.

The Six-Hour Shower Comparison Is Also Misleading

Altman’s comments were partly intended to counter the viral claim that a single ChatGPT query consumes as much water as a six-hour shower.

That comparison is problematic because it presents an extremely large water footprint for an individual AI request without clearly explaining the methodology behind it.

At the same time, replacing one dramatic comparison with another does not completely resolve the underlying issue.

The most accurate conclusion is that the water consumption of AI varies significantly depending on what is being measured.

Neither “one query equals a six-hour shower” nor “38,000 queries equal one almond” should be treated as a universal environmental measurement.

Both comparisons simplify a much more complicated infrastructure problem.

The Real AI Water Question Is About Scale and Infrastructure

The debate over Sam Altman’s ChatGPT water claim ultimately illustrates a broader lesson about technology’s environmental impact.

Individual AI interactions may have a relatively small resource footprint. But when billions of interactions are processed every day through large data centers, those small amounts can accumulate.

The location and design of AI infrastructure also matter enormously.

Efficient cooling systems, renewable electricity, water recycling, alternative cooling technologies and careful data-center placement could all influence the environmental footprint of future AI services.

For consumers, the important takeaway is not that every ChatGPT query is consuming a large amount of water. Nor is it that AI has no meaningful water footprint.

Instead, AI’s water consumption needs to be measured using transparent and consistent methods.

Until researchers and technology companies agree on common accounting standards, headline-friendly comparisons will continue to produce confusion.

What Sam Altman’s Claim Really Tells Us About AI Water Consumption

Sam Altman’s 38,000 ChatGPT queries per almond comparison has generated attention because it turns an invisible infrastructure cost into something people can easily visualize.

But the controversy surrounding the number is arguably more important than the comparison itself.

The figure depends on assumptions about both ChatGPT’s direct water use and the water footprint of almond production. Different methodologies can produce dramatically different results.

The bigger issue is the rapid expansion of AI infrastructure. As billions of prompts are processed and increasingly demanding AI models require more computing power, the industry’s total consumption of electricity, water and other resources will remain an important environmental question.

The future of AI will not only depend on how intelligent these systems become, but also on how efficiently and sustainably the infrastructure behind them can operate.

For now, the safest conclusion is simple: one AI query does not have a single universal water footprint. The number depends on what you count, where the computation happens and how the calculation is performed.