In the wake of viral posts about chatgpt water usage, OpenAI CEO Sam Altman claimed that one almond uses as much water to grow as handling 38,000 user queries in ChatGPT. Altman’s comparison sparked intense discussion across newsrooms and social platforms, revealing deeper issues in how technology leaders communicate artificial intelligence’s actual environmental impact. This article verifies the arithmetic, reviews documented water usage for both almonds and data centers, and unpacks the policy debate over AI sustainability.
What Sam Altman actually said about ChatGPT water usage
On September 3, 2026, Sam Altman discussed environmental concerns surrounding AI in an episode of the Sources podcast with The Verge’s Alex Heath. Altman stated: “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.”
This claim was quickly amplified on X (Twitter), then picked up by outlets including Tom’s Hardware, CalMatters, and Business Insider. Altman’s intention was to push back on the prevalent narrative that AI data centers are uniquely water-intensive and that each user prompt extracts a significant ecological cost.
Does the almond vs. ChatGPT comparison hold up?
Water researchers generally agree that growing a single California almond requires about 3.2 gallons (about 12 liters). Widely cited sources such as UC Davis and the Water Footprint Network back this figure. OpenAI’s own estimates—and independent environmental research—find that a typical ChatGPT query consumes roughly 0.32 milliliters of water per prompt, counting both direct data center cooling and indirect water tied to electricity generation (Explainx.ai).
Dividing the almond water figure by ChatGPT’s per-query usage, the arithmetic for Altman’s headline claim (1 almond ≈ 38,000 queries) checks out, at least on a narrow per-query basis. In other words, the numbers themselves are not far off—if you’re only counting a simple text prompt and using recent data center technology.
What about more complex AI tasks?
However, several investigative reports (see Spokesman-Review), alongside Microsoft’s internal findings, show that much longer, more complex requests—even by individual users—can inflate energy and water consumption by factors up to 10,000 times compared to basic queries. Extended AI “agent” and research tasks increasingly dominate usage patterns and carry far greater environmental costs.
Why the comparison faces expert criticism
- Different scales and contexts. The almond figure reflects the cumulative, indirect water embedded in agriculture, often spread regionally over months. By contrast, AI water demand draws on local utilities or aquifers, sometimes straining water-stressed communities near data centers.
- Opaque, aggregated disclosure. Experts and journalists note that public data on AI data center water use remains patchy. Companies like OpenAI, Microsoft, and Google rarely break out water usage by specific service, model, or task. Ongoing efforts to require transparent reporting are receiving industry pushback (see more from our category).
- Resource substitution fallacy. Framing the discussion as “almonds vs. ChatGPT” ignores the vastly different social and economic functions of food crops and advanced computing, as summarized in CalMatters and Tom’s Guide.
What do we know about total ChatGPT water usage?
Recent peer-reviewed analysis (Patterns, Dec 2025; Moduledge, Sep 2026) concluded that total AI water consumption worldwide reached 312–764 billion liters in 2025, a figure rivalling global bottled-water output. The biggest contributors are model training cycles (like GPT-3/4/5), which can each require hundreds of thousands of liters to cool vast server arrays. “At global scale, these small numbers become enormous,” notes Alex de Vries (Patterns, 2025).
AI’s indirect water draw—including all upstream power generation—means that actual volume per ChatGPT query depends heavily on data center location, local cooling tech, and energy mix. Newer installations increasingly use waterless cooling or recapture to minimize direct withdrawals, shifting some impact onto electricity grids.
OpenAI’s stance and industry response
Altman insists, “modern very large data centers use the equivalent amount of water as an office building in terms of, you know, people running the sinks and the toilets.” OpenAI has stated its intention to improve sustainability, but as CalMatters and CyberProfi coverage note, the industry has lobbied against legislation that would require precise disclosure of facility-level water usage. This hinders researchers’ and the public’s ability to verify corporate claims and compare AI’s true footprint with other resource-heavy industries.
The policy debate: Sustainability versus access
Public criticism is not just about arithmetic, but about policy and ethical priorities. Comparing the water needs of basic food crops with a global AI tool, some experts argue, obscures the local and cumulative impact of large-scale digital infrastructure concentrated in water-sensitive regions. California, home to vast almond groves and many AI data centers, faces ongoing debates about shared water allocation between agriculture, utilities, and tech campuses. Some states and countries are now working to require transparency on data center water and power use.
Practical context and next steps
Ordinary users’ ChatGPT activity does not, individually, account for significant water stress. Yet, as AI adoption accelerates, policymakers and the public are scrutinizing not just per-query statistics but how the industry allocates resources, reports impacts, and invests in sustainability for the long term.
For more on AI energy and water issues, see CyberProfi’s AI sustainability channel.
Frequently Asked Questions
- How much water does one ChatGPT query actually use?
- Recent peer-reviewed research places usage at around 0.32 milliliters of water per basic prompt, though longer tasks may require significantly more.
- Did Sam Altman exaggerate ChatGPT’s relative water efficiency?
- The math checks out for simple queries, but the framing is criticized for ignoring local impact, longer tasks, and transparency gaps.
- How does ChatGPT’s water consumption compare to Google Gemini?
- Google reports their Gemini AI requires 0.26 milliliters per typical text prompt—comparable to ChatGPT. Both rely mainly on efficient, modern data centers that try to minimize water-wastage.
- Is any regulation coming for AI data center water disclosure?
- Proposals are being debated in California and other states to require more detailed, facility-level water reporting from data center operators, but industry resistance remains strong.
- Are OpenAI and others moving toward water-free cooling?
- Many new data centers operate with waterless cooling systems, such as mechanical heat pumps and high-efficiency air cooling, but overall impact shifts to electricity demand and indirect water use for power generation.