A product like ChatGPT can be difficult to measure in terms of its cost. To teach its artificial intelligence system how to mimic human writing, Microsoft’s artificial intelligence-backed OpenAI needed plenty of water from the Raccoon and Des Moines rivers in central Iowa to cool a powerful supercomputer.
Microsoft’s artificial intelligence, OpenAI, and Google have all acknowledged that growing demand for their AI tools carries hefty costs, from expensive semiconductors to increased water usage.
Few people in Iowa knew that GPT-4, OpenAI’s most advanced large language model, was developed near cornfields west of Des Moines, until a top Microsoft executive mentioned it in a speech.
In order to build an extensive language model, a huge trove of human-written text must be analyzed. As a result of all that computing, data centers need to pump in water — often to a cooling tower outside their warehouse-sized buildings — to keep them cool on hot days.
In its latest environmental report, Microsoft disclosed that its global water consumption spiked 34% from 2021 to 2022 (to nearly 1.7 billion gallons, or more than 2,500 Olympic-sized swimming pools), a sharp increase compared to previous years that outside researchers attribute to its artificial intelligence research.
In addition to its heavy investment in generative AI and partnership with OpenAI, Shaolei Ren, a researcher at the University of California, Riverside, who has been trying to calculate the environmental impact of generative AI products like ChatGPT, says the majority of its growth is due to AI.
In a paper due to be published later this year, Ren’s team estimates ChatGPT gulps up 500 milliliters of water (similar to the amount in a 16-ounce water bottle) every time you ask it a series of between 5 and 50 prompts or questions. The estimate includes indirect water consumption that the companies do not measure — such as the cooling of power plants that supply the data centers. The range varies depending on where the server is located and the season.
There isn’t much awareness of ChatGPT’s resource usage,” Ren said. “If you don’t know what’s going on, we can’t help.”
In the same period, Google reported a 20% increase in water use, which Ren also attributed largely to its artificial intelligence efforts. While Google’s water use spiked in Oregon, where it has drawn public attention, it doubled outside Las Vegas, where it has been steady. In Iowa, it also used more potable water in Council Bluffs data centers than anywhere else, indicating it was thirsty.
Microsoft responded to the Associated Press this week with a statement that it is investing in research to measure artificial intelligence’s energy and carbon footprint “while working on ways to make large systems more efficient in both training and application.”
In order to reach our sustainability goals of being carbon negative, water positive and zero waste by 2030, the company will continue monitoring its emissions, accelerating progress, using more clean energy to power data centers, purchasing renewable energy, and other measures and Microsoft’s artificial intelligence tool will also be helpful for the water shortage issues.
In its own statement Friday, OpenAI said it is considering the best use of computing power.
Training large models can be energy- and water-intensive, and we strive to improve efficiency, the company said.
More than two years before OpenAI launched ChatGPT and sparked worldwide fascination with artificial intelligence, Microsoft invested $1 billion in San Francisco-based OpenAI in 2019. Software giant Microsoft will provide the computing power needed to train the AI models as part of the deal.
Both companies looked to West Des Moines, Iowa, a city of 68,000 people where Microsoft has been building data centers to power its cloud computing services for more than a decade, to do at least part of that work. Later this year, it will open its fourth and fifth data centers.
As mayor of the city when Microsoft came here, Steve Gaer said, “They’re building them as fast as they can.” The company was attracted by the city’s commitment to building public infrastructure, and it has contributed a “staggering” amount through tax payments.
“But they were pretty secretive about what they were doing out there,” he explained.
In 2020, Microsoft announced the development of one of the most powerful supercomputers in the world for OpenAI. At the time, it refused to reveal its location, but described it as a “single system” with more than 285,000 conventional semiconductor cores and 10,000 graphics processors, a kind of chip vital to AI.
Considering the large amounts of data that need to be transferred between computing cores, experts say “pretraining” an AI model in one place can make sense.
Microsoft’s president, Brad Smith, did not reveal to the public until late May that the company had built its “advanced AI supercomputing data center” in Iowa, solely for OpenAI to train the GPT-4 model. As a result, Microsoft’s own products and ChatGPT are powered by the model, which has prompted a debate about containing AI’s societal risks to rage.
Smith said, “These extraordinary engineers made it in California, but it was really made in Iowa.”
West Des Moines may be a relatively efficient location for training a powerful AI system, especially when compared to Microsoft’s Arizona data centers that consume more water. Microsoft artificial intelligence tool will definitely helpful for reduce water shortages.
In terms of training, there is no difference between Iowa and Arizona. In terms of water or energy consumption, there is a big difference.
During most of the year, Iowa’s weather is cool enough for Microsoft to vent heat out of the building and keep the supercomputer running properly. The company said in a public disclosure that it only withdraws water when the temperature exceeds 29.3 degrees Celsius (about 85 degrees Fahrenheit).
A lot of water, especially in the summer. According to the West Des Moines Water Works, Microsoft pumped in about 11.5 million gallons of water to its cluster of Iowa data centers in July 2022, the month before OpenAI claims it completed its GPT-4 training. About 6% of all water used in the district, which also supplies drinking water to the city’s residents, was used for this purpose.
As of 2022, the West Des Moines Water Works and the city government “will only consider future data center projects” from Microsoft’s artificial intelligence if those projects can “demonstrate and implement technology that significantly reduces peak water usage from current levels” so that residential and commercial water supplies can be protected.
According to a written statement from the water works, Microsoft has been a good partner and has been working with local officials to reduce its water footprint while meeting its needs.