Sam Altman ChatGPT Water Usage Claims Draw Scrutiny

Cable network (Photo by Taylor Vick on Unsplash )

Cable network (Photo by Taylor Vick on Unsplash)

Summary
  • Altman said 38,000 ChatGPT queries equal one California almond in water use
  • Altman blamed viral memes and defended modern data center cooling practices
  • Researchers say one almond uses about 1.1 gallons of water
  • Reporting cited older model training and local US data center water incidents

OpenAI chief executive Sam Altman defended the water footprint of ChatGPT and said public concern had been exaggerated. He made the remarks during an interview on the Sources Podcast with Alex Heath. Altman said a viral claim about a single ChatGPT query using as much water as six hours of showering did not hold up.

Altman offered a different comparison and said, from memory, that 38,000 ChatGPT queries use roughly the same amount of water as producing one almond in California. He said that assertion represented what he called full water accounting rather than a narrow data center measure. Altman also noted that data centers once relied on evaporative cooling but that most large modern facilities no longer do.

He argued modern data centers use about the same amount of water as an office building for sinks and toilets. Altman said that persistent memes about AI water use were hard to disprove but did not reflect current cooling technology.

Context Evidence And Industry Responses

Researchers cited in reporting say it takes about 1.1 gallons of water to grow a single almond in California. Altman has previously suggested one ChatGPT response could use between 1 and 50 milliliters of water, which equates to roughly 85 to 5,500 chat queries per almond under unspecified cooling methods.

Estimates in the reporting indicate that training older models could consume far more water, with GPT-3 cited as having used almost 185,000 gallons under older systems. The piece also noted several local concerns about data center water use, including a site in Fayette County, Georgia, that used 29 million gallons over 15 months and another development in Morgan County, Georgia, that allegedly left neighbors with muddy water.

Industry cooling approaches were highlighted as factors that change water demand. The report mentioned Nvidia’s liquid cooling system, described as running hotter than a hot tub and eliminating water use in that design, and Microsoft’s closed-loop cooling, compared to a restaurant’s water needs. Amazon was reported as claiming its data centers consume 0.075 percent of the water Americans use for lawns and gardens. Altman acknowledged older evaporative systems used significant water but said newer technologies have cut usage.