Sam Altman of OpenAI: AI data centers consume even less water than almond farming.

OpenAI CEO Sam Altman responded to the criticism over water consumption of AI data centers, stating that the water consumption of modern data centers is comparable to that of ordinary large office buildings, and he also cited the 2024 government committee report of Virginia as supporting evidence.

In a podcast program released on Tuesday, when the host asked Sam Altman, Chief Executive Officer of OpenAI, to respond to people's resistance to artificial intelligence (AI) data centers, including concerns about water resource consumption, he refuted this claim.

Altman pointed out that although data centers have long been reputed as large water consumers, the water consumption of modern data centers is roughly equivalent to that of office buildings.

"This used to be a widely circulated claim that was difficult to falsify, but I think it cannot stand any scrutiny," he added.

He mentioned a 2024 report from the state government of Virginia in the United States. Virginia is a hot area for data center construction. A committee of the local government found that although some data centers consume significantly more water than other facilities, most data centers consume the same amount of water or less than ordinary large office buildings.

But in fact, water resource consumption has always been one of the core issues of opposition to data centers. A report released by Gallup in May shows that 71% of Americans say they oppose the construction of data centers in their local areas, and 70% of them say they are particularly worried about the environmental impact of data centers. Half of the respondents mentioned the impact on resources, of whom 18% specifically referred to water resources.

A survey released by Business Insider last year found that data centers have exacerbated the water crisis in the West, partly due to the extremely water-intensive evaporative cooling technology.

However, according to the Data Center Coalition, an industry trade organization, data center developers are adopting a variety of cooling methods, including some closed-loop or water-free systems, and prioritizing efficient water use practices to adapt to local water resource supply conditions.

Altman also compared the water consumption of ChatGPT with the water consumption of growing almonds in California, which is a crop that consumes huge amounts of water. He said, The amount of water required to grow one almond in California is more than the amount of water required for thousands of ChatGPT queries, adding: "Most people who eat 12 almonds at one time do not feel that they have done something terrible in terms of water resources."

Although Altman's claim cannot be verified, relevant media pointed out that the statement that "the water consumption for growing one almond is several orders of magnitude higher than the water consumption of an AI chatbot prompt" is basically consistent with the existing data.

Altman said that an average ChatGPT query consumes about 0.32 milliliters of water, while Google stated that the average text prompt of the Gemini application consumes 0.26 milliliters of water. A 2019 study by researchers from the United States Geological Survey (USGS) found that growing an almond in California requires an average of 3.56 liters of water, which is slightly less than one gallon of water per almond.

He further stated that although people's water-related concerns about artificial intelligence have been exaggerated, he believes that it is the responsibility of the industry to change public perceptions.

"I think the right way to make people like something is to provide value for them," he added: "The industry still has a lot of work to do in terms of how to make these products easy to use and how to let people get great value from them."

This article comes from the WeChat official account "Sci-Tech Innovation Board Daily", author: Huang Junzhi, published with authorization from 36Kr.