
Fact check: Is Sam Altman right that almonds use more water than ChatGPT queries?
California Governor Gavin Newsom is considering two bills that would require data centers to disclose their water usage and plans. OpenAI CEO Sam Altman claims AI queries use less water than almond production, a statement experts challenge due to a lack of transparent industry data. The legislation aims to increase transparency amidst growing concerns over data center water consumption across the state.
California Governor Gavin Newsom is currently reviewing two transparency bills, introduced by Assemblymember Diane Papan, that would mandate data centers to disclose their water sources and usage when seeking business permits, and require developers to provide water plans and cover upgrade costs for new facilities. This legislative push comes as concerns escalate over the environmental impact of data centers, particularly their significant water consumption.
OpenAI CEO Sam Altman has downplayed these concerns, asserting that 38,000 ChatGPT queries use less water than producing a single almond in California, and comparing a large data center's water use to that of a typical office building. However, experts like Shaolei Ren from UC Riverside and Michael Kiparsky from UC Berkeley's Wheeler Water Institute challenge such claims, citing the extreme lack of public data from the secretive data center industry, making it impossible to verify these figures. They emphasize the numerous variables affecting data center water consumption, from location and temperature to cooling systems and query complexity.
Despite Governor Newsom having vetoed similar legislation last year due to concerns about business impact, the renewed effort for transparency is gaining urgency amid increasing data center developments and local government bans across California. Studies indicate a substantial rise in data center water usage, with one report showing facilities directly used 17 billion gallons in 2023, up from 5.6 billion in 2014. Researchers also highlight that peak water demand on hot days, critical for cooling, can strain aging and fragmented community water systems, which often supply treated potable water to data centers, unlike the irrigation water typically used by agriculture.