In an era marked by the rapid advancement of artificial intelligence (AI), the demand for AI technology is surging. Major players like Microsoft, OpenAI, and Google are at the forefront of this AI revolution. However, behind the scenes, the development of AI, particularly large language models like ChatGPT and GPT-4, comes with a substantial environmental cost, including a surge in water consumption. The heart of this story lies in central Iowa, where OpenAI’s cutting-edge AI technology has a surprising birthplace.
The role of water in AI
AI development, especially training large language models, relies heavily on computing power, generating significant amounts of heat. To keep these systems cool, data centers require massive amounts of water. The recent disclosure by Microsoft reveals that its global water consumption spiked by 34% from 2021 to 2022, primarily attributed to AI research, including its partnership with OpenAI.
Water consumption of AI models
A forthcoming research paper led by Shaolei Ren at the University of California, Riverside, sheds light on the water consumption of AI models like ChatGPT. The study estimates that every interaction with ChatGPT consumes roughly 500 milliliters of water. This estimate encompasses not only direct water usage but also indirect water usage for cooling power plants supplying electricity to data centers.
Google’s water usage
Google, another tech giant invested in AI, reported a 20% increase in water use during the same period. Much of this increase can be attributed to AI work, with variations in water consumption depending on the location of their data centers. Iowa, in particular, saw a substantial increase in potable water usage by Google.
When questioned about their environmental impact, both Microsoft and OpenAI emphasized their commitment to sustainability. Microsoft stated its dedication to research measuring AI’s energy and carbon footprint while striving for efficiency improvements. They pledged to use more clean energy and purchase renewable energy to meet sustainability goals. OpenAI echoed these sentiments, acknowledging the resource-intensive nature of training large models and vowing to enhance efficiencies.
Microsoft’s collaboration with OpenAI brought the AI development process to West Des Moines, Iowa, where they have been amassing data centers for over a decade. This discreet location proved efficient for training AI systems, notably GPT-4. The city’s climate allows for natural cooling through most of the year, reducing the need for water-based cooling systems.
Water usage and sustainability
Despite the advantages of West Des Moines, the water consumption by Microsoft’s data centers during peak periods raised concerns. In July 2022, for instance, Microsoft used approximately 6% of the district’s water supply, prompting local authorities to call for a reduction in water usage. Microsoft is currently collaborating with local officials to address this concern while maintaining its operational needs.
This situation highlights the intricate relationship between AI development, water consumption, and sustainability. As the demand for AI technology continues to grow, tech giants are facing pressure to minimize their environmental footprint. The choice of data center locations and cooling methods plays a crucial role in achieving this balance.
A broader debate
The development of GPT-4 and similar AI models raises broader questions about the societal and environmental risks of advanced AI technology. As AI systems become increasingly powerful and prevalent, it is imperative to consider the environmental and ethical implications of their development and usage.
The race to capitalize on generative AI technology is ongoing, with tech developers striving to strike a balance between innovation and sustainability. As AI technology continues to evolve, it is crucial for companies to transparently address the environmental impact of their AI research and development activities.
While the AI revolution brings exciting possibilities, it also raises concerns about its environmental footprint. The convergence of AI, water consumption, and sustainability is a pressing issue that tech giants like Microsoft, OpenAI, and Google must address as they shape the future of AI technology.
