Altman: 0.32 ml per ChatGPT request, 3.56 L for an almond

Key Takeaways

1Sam Altman claims that a ChatGPT query consumes 0.32 ml of water, compared to 3.56 L for an almond according to a 2019 study

271% of Americans oppose a data center near them, with 18% citing water as a concern

3The industry mentions energy-efficient cooling systems, but surveys highlight the impact of evaporative cooling

Why it matters : The water consumption of AI data centers crystallizes environmental concerns, despite figures put forward by industry players.

Sam Altman claims that AI consumes little water per request and compares ChatGPT to water-intensive crops like almonds. He cites 0.32 ml per request, a figure that Google complements for Gemini. Despite these orders of magnitude and a report from Virginia mentioning uses similar to those of office buildings, surveys show massive opposition to data centers, while investigations highlight very water-intensive cooling techniques and the industry promotes more efficient systems.

Milliliters per request versus liters per nut

Sam Altman indicates that an average ChatGPT request requires about 0.32 milliliters of water. Google reports that the average text request for Gemini uses 0.26 milliliters. For comparison, a 2019 study conducted by researchers affiliated with the US Geological Survey estimates that it takes an average of 3.56 liters of water to grow an almond in California. Altman asserts that a single almond requires more water than thousands of ChatGPT requests. It has not been possible to independently verify this comparison, but it is generally considered to align with the available orders of magnitude.

Marked public skepticism towards data centers

According to a Gallup poll published in May, 71% of Americans oppose the construction of a data center in their area. In this survey, 70% express concerns about the environmental impact, with half mentioning pressure on resources and 18% specifically citing water. Water consumption thus ranks among the top concerns related to these infrastructures.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Altman contests the scale of the problem and invokes office usage

Sam Altman believes that fears regarding water consumption linked to AI are exaggerated and refers to it as an unfounded "meme." According to him, the amount of water used by a modern data center is similar to that consumed by a large office building, even though they are often seen as major consumers. He made these statements during the first episode of a podcast aired on a Tuesday, where he was questioned about resistance to AI and water-related issues. In 2024, a government commission in the state of Virginia indicated that while some facilities show significantly higher consumption, the majority use an amount of water equivalent to or less than that of a large office.

Cooling: criticism over evaporation, defense of closed-loop systems

According to a survey published last year by Business Insider, part of the water crisis in the West is attributed to data centers, particularly due to evaporative cooling, which is very water-intensive. For its part, the Data Center Coalition claims that developers also use closed-loop or waterless systems and prioritize practices suited to local availability. While he considers the concerns exaggerated, Sam Altman emphasizes that it is up to the industry to change public opinion by improving the usability of products and providing value.