Microsoft-backed OpenAI’s AI Training Comes at a Cost: Soaring Water Consumption Raises Concerns
The demand for artificial intelligence (AI) tools is on the rise, with tech giants like Microsoft, OpenAI, and Google racing to capitalize on the craze. However, the development and training of AI models come at a significant cost, including soaring water consumption. While the specifics of these costs are often shrouded in secrecy, a recent environmental report from Microsoft revealed a 34% spike in global water consumption from 2021 to 2022, largely attributed to its AI research and partnership with OpenAI.
Building large language models, such as OpenAI’s advanced GPT-4, requires immense computing power and generates substantial heat. To keep the systems cool, data centers rely on water, pumping it into cooling towers outside their warehouse-sized buildings. However, this process consumes a staggering amount of water, which has led to concerns about its environmental impact.
Shaolei Ren, a researcher at the University of California, Riverside, has been studying the environmental impact of generative AI products like ChatGPT. In a forthcoming paper, Ren’s team estimates that ChatGPT consumes approximately 500 milliliters of water every time it processes a series of 5 to 50 prompts or questions. This estimate also takes into account indirect water usage, such as for cooling power plants supplying electricity to the data centers. Ren emphasizes that many people are unaware of the resource consumption underlying these AI systems, preventing proactive conservation efforts.
Water consumption by tech giants has attracted attention in recent years, particularly with Google’s data centers doubling their water usage near Las Vegas. Notably, Microsoft’s data centers in Iowa, where it partnered with OpenAI, also saw a significant surge in water usage. The undisclosed location in West Des Moines served as the birthplace of OpenAI’s GPT-4, powering premium versions of ChatGPT and other Microsoft products.
Compared to Microsoft’s water-thirsty data centers in Arizona, training AI models in Iowa proves relatively efficient. The cool climate allows Microsoft to utilize outside air to maintain the supercomputers’ optimal temperature and vent excess heat. The withdrawal of water for cooling purposes only occurs when the temperature exceeds 29.3 degrees Celsius (85 degrees Fahrenheit). Nonetheless, the water consumption during hot summer months can still be substantial, amounting to about 6% of the water used by the West Des Moines district in July 2022.
In response to concerns raised by the West Des Moines Water Works, Microsoft is working closely with the organization to address feedback and reduce its water footprint while meeting its operational needs. The company is committed to monitoring its emissions, increasing its use of clean energy, and pursuing sustainability goals, aiming to be carbon negative, water positive, and zero waste by 2030.
OpenAI, which received a $1 billion investment from Microsoft in 2019, echoes Microsoft’s commitment to improving the efficiency of large AI systems. The companies acknowledge the energy and water-intensive nature of training these models and are actively researching ways to mitigate their environmental impact.
While AI advancements have undoubtedly revolutionized various industries, including language processing, the associated costs and resource consumption raise valid concerns. As the demand for AI continues to grow, it is crucial for researchers, developers, and consumers to be more aware of the environmental consequences and work towards sustainable solutions.
