Microsoft used 1.7 billion gallons of water for AI in 2022

An April of 2023 report revealed that ChatGPT-3's training alone guzzled up a whopping 85,000 gallons (700,000 liters) of water. The research further estimated that ChatGPT, which followed GPT-3, would be required to consume the equivalent of a 500-milliliter (60-ounce) water bottle to execute a discussion with a user of around 25 to 50 questions.

Large-scale cooling solutions are needed in data centers, where AI computations are run, to avoid overheating the servers and other gear that power the technology. HVAC (heating, ventilation, and air conditioning) systems and cooling towers use a significant amount of water.

In addition, AI computations can be energy-intensive. In some places, producing the electricity required for these operations can lead to a lot of water use, especially when using power generation technologies like thermoelectric plants, which rely heavily on water for cooling.

APreached out to Microsoft, who responded with a statement saying the firm was “working on ways to make large systems more efficient, in both training and application.”

In order to reach its sustainability targets of becoming carbon negative, water positive, and waste-free by 2030, the firm stated it would continue to analyze its emissions, boost the usage of clean energy to power data centers, purchase renewable energy, and invest in other eco-friendly initiatives.

Meanwhile, Open AI told the news outlet that the firm “recognize[s] training large models can be energy and water-intensive” and is actively working to improve efficiencies.

Creative solutions explored

One solution being explored is to lower the energy requirements of AI computations. These efforts can make AI more energy-efficient by building better algorithms and hardware, which can indirectly reduce water use.

There is also a more creative option. Back in 2018, Microsoft sunk one of its data centers off the coast of Orkney and studied its performance for two years. The center was found to have operated with increased efficiency during that time period.

Part of the reason could have been that the water surrounding it kept it cool, Microsoft said at the time of its retrieval in 2020 that the process had environmental benefits. "The retrieval launched the final phase of a years-long effort that proved the concept of underwater data centers is feasible, as well as logistically, environmentally and economically practical," wrote Microsoft's John Roach in a blog.