AI models require a tremendous amount of computing power, which in turn demands a significant cooling capacity. This has led to a substantial increase in water consumption. In response to the escalating water usage, Microsoft has announced plans to achieve net water positivity by 2030.
According to media reports, the OpenAI technology supported by Microsoft requires a large amount of water for cooling supercomputers and facilitating AI learning to mimic human writing. In addition to constructing cooling towers, the operation of servers consumes a substantial amount of electricity, and thermal, natural gas, and nuclear power generation also consume significant amounts of water.
In its latest environmental report, Microsoft revealed that its water consumption in 2022 increased by 34% compared to the previous year, representing a significant surge in recent years. This is equivalent to more than 2,500 Olympic-sized swimming pools. External researchers believe this increase is related to Microsoft's artificial intelligence research.
As a hub for Microsoft's cloud computing data centers, West Des Moines, Iowa, serves as a major source of water consumption. It is reported that Microsoft's fourth and fifth data centers will also be operational in West Des Moines this year. Additionally, the supercomputing data center used by OpenAI to train GPT-4 is also located there. Therefore, compared to Microsoft's data center in Arizona, the data center in Iowa consumes much more water under equivalent computing demands.
Although the cool weather in Iowa ensures the normal operation of supercomputers, the data centers need to draw water for cooling when the surrounding temperature exceeds 29.3 degrees Celsius. In July 2022, one month before OpenAI announced the completion of GPT-4 training, Microsoft pumped 11.5 million gallons (approximately 43.53 million liters) of water into the West Des Moines data center cluster, accounting for 6% of the local water consumption.
Shaolei Ren, a researcher at the University of California, Riverside, stated, &#34;It can be said that most of the water consumption growth is caused by artificial intelligence, including the significant investment in generative AI and collaboration with OpenAI.&#34;
In an upcoming paper to be published this year, Ren's team estimated that each time you provide 5-50 prompts or questions to ChatGPT, it &#34;consumes&#34; approximately 500 milliliters of water. This estimate includes indirect water usage that Microsoft has not calculated, such as the cooling water for power plants supplying electricity to data centers. The specific water consumption varies depending on the location of the servers and the season.
Google's 2023 environmental report also reveals that it consumed a staggering 21.2 billion liters of water in 2022, with approximately 19.6 billion liters used for data centers.
In a statement released this week, Microsoft announced its investment in research related to measuring the energy and carbon footprint of artificial intelligence and exploring ways to make large-scale systems more efficient in training and application. The statement further emphasizes:
<blockquote>
&#34;We will continue to monitor our emissions, accelerate progress, and use clean energy sources more for powering our data centers. We will also purchase renewable energy and make other efforts to achieve our sustainable development goals, which include achieving carbon negativity, positive water emissions, and zero waste by 2030.&#34;
</blockquote>
OpenAI also conveyed the same viewpoint in its statement last Friday, stating that it is &#34;seriously considering&#34; how to better utilize computing power.

MSFT

How to deal with the "thirsty" AI?

The large model consumes a lot of water? Microsoft's water usage soars by 34%!