The sources reviewed here do not establish a measured or verified total for ChatGPT’s water use in calendar year 2025. A rough calculation is possible: combining Sam Altman’s estimate of 0.000085 gallons per query with OpenAI’s report of more than 2.5 billion ChatGPT messages per day in July 2025 gives about 78 million gallons (294 million liters) if that daily volume and the per-query estimate held steady for all 365 days. That is a hypothetical annualization—not a reported 2025 total. (blog.samaltman.com)
The figures behind that estimate
In The Gentle Singularity, Sam Altman wrote that an average ChatGPT query uses about 0.000085 gallons of water, describing it as roughly one-fifteenth of a teaspoon. The post provides a per-query estimate, not an annual total. (blog.samaltman.com)
Separately, OpenAI reported on July 22, 2025 that users sent more than 2.5 billion messages per day to the platform. OpenAI’s later usage paper says ChatGPT was sending 18 billion messages weekly by July 2025. These are snapshots of activity—not counts of every query across the whole calendar year. (openai.com)
A reproducible, conditional calculation
For an illustration, treat OpenAI’s reported messages as if they were queries, assume the July daily rate applied all year, and assume Altman’s per-query estimate applied to each one:
- Per day: 0.000085 gallons × 2.5 billion messages = 212,500 gallons
- For 365 days: 212,500 × 365 = 77,562,500 gallons, or about 78 million gallons
- Metric equivalent: about 294 million liters
This calculation is useful for understanding how a per-query estimate scales. It does not show how much water ChatGPT actually used in 2025. Message volume may have changed during the year; a message is only being used here as a proxy for a query; and the published per-query figure does not, in the cited statement, define a measurement method or water-accounting boundary. So the result should not be treated as a measured total or as a complete water footprint.
Does the estimate apply to production model APIs?
Not automatically. The per-query figure concerns an average ChatGPT query; the sources reviewed do not provide a matching water-use factor for production API requests. An API workload may differ, and the cited figure does not establish that its model, workload, infrastructure, or accounting scope matches yours.
For an API-specific estimate, you would need a water-use factor and usage data that refer to comparable models and workloads, the same time period, and a clearly stated accounting boundary. Without that match, multiplying API request volume by the ChatGPT figure would produce an assumption—not a reliable API water estimate.
Bottom line: The most defensible answer is that the reviewed sources do not establish ChatGPT’s actual 2025 annual water use. The roughly 78-million-gallon figure is a transparent scenario calculation, not a measured or official calendar-year total.
Sources
The Gentle Singularity — Sam Altman