The water consumption of the emerging AI industry has come under scrutiny in a recent environmental report released by Microsoft. The report showed a substantial increase in water usage between 2021 and 2022, directly linked to the acceleration of the company’s AI operations. It was found that Microsoft consumed approximately 6,399,415 cubic meters of water in a 12-month period, representing a 30% increase from the previous year.
This finding aligns with a study conducted by the University of California Riverside, which estimated that engaging with AI chatbots like ChatGPT can take up to half a liter of water per interaction. Additionally, the study projected that Microsoft used approximately 700,000 liters of water to train its GPT-3 model over a two-week period. The authors of the study expressed concern over the sheer amount of freshwater being used, emphasizing the pressing challenge of global freshwater scarcity.
According to Shaolei Ren, one of the study’s authors, AI is significantly more energy-intensive than other forms of computing. AI servers, equipped with GPUs, consume two to three kilowatts of power per server, compared to the typical consumption of below 500 watts for regular servers. This higher energy density results in increased cooling needs, further contributing to water consumption.
While there are potential methods for tech companies to reduce water usage in AI training, the lack of publicly released data from AI vendors makes it challenging for oversight and evaluation. Transparency regarding water conservation efforts is crucial in addressing the environmental impact of the AI industry.
FAQ:
Q: How much water did Microsoft consume in a 12-month period?
A: Microsoft consumed approximately 6,399,415 cubic meters of water in a 12-month period.
Q: How much water does it take to engage with AI chatbots like ChatGPT?
A: Engaging with AI chatbots like ChatGPT can take up to half a liter of water per interaction.
Q: Why is AI more water-intensive than other forms of computing?
A: AI servers, with their high GPU density, consume two to three kilowatts of power per server, resulting in higher cooling needs and increased water consumption.
Q: What is the challenge associated with the water consumption of the AI industry?
A: The increased water consumption of the AI industry exacerbates global freshwater scarcity, which is one of the most pressing challenges of our time.