SAN FRANCISCO — OpenAI Chief Executive Officer Sam Altman has pushed back against rising public anxiety over artificial intelligence’s environmental footprint, dismissing concerns surrounding data center water consumption as an overblown, persistent “meme.”

Speaking on the premiere episode of the Sources podcast hosted by Alex Heath, Altman insisted that criticism regarding AI’s resource demands does not stand up to scrutiny. He contended that modern hyper-scale facilities have largely phased out older evaporative cooling methods in favor of closed-loop systems, consuming water levels comparable to a standard commercial office building through everyday facility usage like sinks and restrooms.

To highlight his point, Altman offered a striking agricultural comparison, asserting from memory that it takes approximately 38,000 ChatGPT queries to consume the equivalent amount of water required to cultivate a single California almond.

“That has been a robust meme and difficult to disprove, but I don’t think holds up to any scrutiny,” Altman told Heath. “People that are scarfing down 12 almonds at a time don’t feel like they’re doing something horrible from a water perspective, for the most part.”

Fact-Checking the Math: Estimates and Variables

Independent analysis indicates that while Altman’s broader point highlights the massive water demands of agricultural production, his specific ratios depend heavily on unverified variables.

Previous estimates, including figures referenced by Altman, put the direct cooling footprint of a single standard ChatGPT query at roughly 0.32 milliliters of water. By contrast, a 2019 study affiliated with the U.S. Geological Survey (USGS) calculated that growing a single California almond consumes an average of 3.56 liters (about 1.1 gallons) of water—a ratio closer to 11,000 queries per almond. Older estimates, which factored in model training and high-powered reasoning tasks, put query consumption higher, lowering the ratio to a few thousand queries per almond.

Academic Experts Point to Lack of Data and Infrastructure Strain

Researchers and environmental policy experts stress that verifying any specific claim remains nearly impossible due to the secretive nature of data center operations.

Shaolei Ren, an associate professor of electrical and computer engineering at UC Riverside, emphasized that calculating a query’s water footprint is far from straightforward. Factors such as geographical location, local weather conditions, cooling technologies, prompt length, and backend processing power continuously alter real-world consumption.

“The information we have is so limited,” Ren noted, pointing out a critical distinction between agricultural and technological water use: data centers predominantly rely on treated, municipal potable water rather than raw irrigation supply.

Ren’s research shows that on extreme heat days—when cooling systems work hardest—data centers can create massive spikes in peak water demand. Nationally, these demand surges can require an extra 697 million to 1.45 billion gallons of peak water capacity per day, placing immense strain on aging, municipal water utilities in dry regions.

Michael Kiparsky, director of the Wheeler Water Institute at UC Berkeley’s Center for Law, Energy, & the Environment, echoed those transparency concerns. “There’s no published data — certainly not for California — and what is public seems all over the place,” Kiparsky said.

California Legislation and Growing Public Backlash

The debate comes as tech leaders face mounting regulatory pressure and public pushback. A recent nationwide Gallup poll revealed that 71% of Americans oppose data center developments in their local communities, with 70% citing environmental impact as a chief concern.

In Sacramento, California lawmakers recently passed a two-bill oversight package authored by Assemblymember Diane Papan (D-San Mateo):

• Assembly Bill 2619: Would require data center operators to disclose expected water usage totals and water sources under penalty of perjury when applying for business licenses.
• Assembly Bill 2469: Bars local governments from approving new or expanded data centers unless developers submit detailed water supply assessments and agree to pay for any necessary municipal infrastructure upgrades.

Both measures now sit on Governor Gavin Newsom’s desk. Newsom vetoed a similar transparency bill the previous year, citing concerns over rigid regulatory burdens on tech innovation. However, with local municipalities across California enacting temporary moratoriums on new data center construction, the governor faces renewed pressure from environmental groups and local leaders demanding mandatory disclosures.

As tech companies rapidly scale infrastructure to power next-generation AI agents, public policy experts maintain that transparent, audited utility reporting remains the only way to resolve the debate between executive claims and local resource realities.