Overview
- Speaking at India’s AI Impact Summit, the OpenAI CEO called per‑query water claims such as “17 gallons” for a ChatGPT answer “completely untrue,” noting many newer facilities no longer use evaporative cooling.
- He disputed comparisons that peg a single query to an iPhone battery charge and said concern should focus on total power demand, urging rapid expansion of nuclear, wind and solar.
- Altman reframed efficiency by comparing a trained model answering a question to a trained human doing the same task, asserting AI may already match human energy efficiency on that basis.
- He dismissed space‑based data centers as unrealistic at scale this decade, citing launch costs and maintenance hurdles like repairing GPUs in orbit.
- The comments drew public pushback, including from Zoho’s Sridhar Vembu, as coverage pointed to rising data‑center electricity and water needs, a San Marcos, Texas, data‑center rejection, and the lack of mandatory resource‑use disclosure that fuels conflicting estimates.