Particle.news
Download on the App Store

Sam Altman Defends AI Energy Use With Human-Training Analogy, Calls Per-Query Water Claims ‘Totally Fake’

Critics say the comparison sidesteps limited disclosure and fast-rising data-center demand.

Overview

  • At India’s AI Impact Summit, Altman argued a fair benchmark is the energy for a trained model to answer a question versus the decades of energy and food invested in “training” a human, citing roughly 20 years before a person becomes smart.
  • He dismissed viral per‑query resource figures — including claims of 17 gallons of water and 1.5 iPhone battery charges per ChatGPT response — as inaccurate, pointing to shifts away from evaporative cooling in newer data centers.
  • Altman acknowledged overall electricity use from widespread AI as a valid concern and urged a rapid build‑out of low‑carbon power, naming nuclear, wind and solar.
  • The remarks prompted backlash, with Zoho co‑founder Sridhar Vembu and online critics condemning any framing that equates a piece of technology with a human being.
  • Context from researchers and analysts highlights scarce mandatory disclosure on energy and water use and mounting infrastructure strain, with estimates placing data‑center power consumption on the scale of major countries and local resistance such as San Marcos, Texas, rejecting a $1.5 billion facility.