ChatGPT AI Developed in Iowa with Abundant Water Supply

Measuring the cost of developing an artificial intelligence product such as ChatGPT can pose challenges.

But Microsoft-backed OpenAI needed plenty of water for its technology, drawn from the Raccoon and Des Moines rivers’ watershed in central Iowa, to cool the powerful supercomputer as it helped its AI systems mimic human typing.

As they race to capitalize on the generative AI craze, leading technology developers such as Microsoft, OpenAI and Google have acknowledged that the growing demand for their AI tools is creating huge costs, from expensive semiconductors to increased water consumption.

But they often hide the details. Few people in Iowa knew of its status as the birthplace of OpenAI’s most advanced major language model, GPT-4, until a top Microsoft executive said in a speech that it was “literally made next to a cornfield west of Des Moines.”

Building a large language model requires analyzing patterns from a huge set of human-written text. All this data processing consumes a lot of electricity and generates a lot of heat. To keep it cool on hot days, data centers must pump water—often to a cooling tower outside warehouse-sized buildings.

In its latest environmental report, Microsoft revealed that its global water consumption rose 34 percent 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. its AI research.

“It’s fair to say that most of the growth is due to AI,” including “its heavy investment in generative AI and its partnership with OpenAI,” said Shaolei Ren, a researcher at the University of California, Riverside, who has tried to quantify generative AI products like ChatGPT. environmental effects.

In a paper to be published later this year, Ren’s team estimates that ChatGPT drinks 500 milliliters of water (about the amount of a 16-ounce water bottle) every time you ask it a series of 5 to 50 prompts or questions. The selection varies depending on the location of its servers and the time of year. The estimate includes indirect water use that companies do not measure – for example, for cooling power plants that supply electricity to data centers.

“Most people are not aware of the resource usage behind ChatGPT,” said Ren. “If you are not aware of the use of resources, we cannot help save resources.”

Google reported a 20 percent increase in water use over the same period, which Ren also largely attributes to his AI work. Google’s spike wasn’t flat — it was flat in Oregon, where its water use has drawn public attention, while outside Las Vegas it doubled. It was also thirsty in Iowa, drawing more potable water to its Council Bluffs data centers than anywhere else.

In response to questions from The Associated Press, Microsoft said in a statement this week that it is investing in research to measure the energy and carbon footprint of artificial intelligence “while working on ways to make large systems more efficient in both training and applications.”

“We will continue to monitor our emissions, accelerate progress, and increase the use of clean energy to power data centers, purchase renewable energy, and other efforts to achieve our sustainable development goals of being carbon negative, water positive, and zero waste by 2030,” said the company’s announcement.

OpenAI echoed those comments in its own statement on Friday, saying it was “significantly” thinking about the best use of computing power.

“We understand that training large models can be energy and water intensive” and are working to improve efficiency, it said.

Microsoft made its first billion-dollar investment in San Francisco-based OpenAI in 2019, more than two years before the startup introduced ChatGPT and sparked a global fascination with AI development. As part of the agreement, the software giant would supply the computing power needed to train artificial intelligence models.

To do at least some of that work, the companies headed to West Des Moines, Iowa, a town of 68,000 where Microsoft has for more than a decade assembled data centers to run its cloud services. Its fourth and fifth data centers are scheduled to open there later this year.

“They’re building them as fast as they can,” said Steve Gaer, who was the city’s mayor when Microsoft came to town. Gaer said the company attracted the city’s commitment to building public infrastructure and donated a “substantial” amount of money in tax payments to support that investment.

“But you know, they were quite secretive about what they were doing there,” he added.

Microsoft first said it would develop one of the world’s most powerful supercomputers for OpenAI in 2020, but declined to disclose its location to the AP at the time, describing it as a “single system” with more than 285,000 traditional semiconductor cores and 10,000 graphics cards. – a type of chip that has become central to AI workloads.

Experts have said that it may make sense to “pre-train” an AI model in one place because of the large amounts of data that need to be transferred between computing cores.

It wasn’t until late May that Microsoft president Brad Smith revealed that it had built its “advanced AI supercomputer data center” in Iowa so that OpenAI could train its fourth-generation model, GPT-4. The model now uses premium versions of ChatGPT and some of Microsoft’s own products, and has fueled the debate about curbing AI’s societal risks.

“These extraordinary engineers did it in California, but it was really done in Iowa,” Smith said.

In some ways, West Des Moines is a relatively efficient place to train a powerful AI system, especially compared to Microsoft’s Arizona data centers, which use much more water for the same computing needs.

“So if you’re developing AI models at Microsoft, you should schedule your internship in Iowa instead of Arizona,” Ren said. “There is no difference when it comes to education. There is a big difference in water or energy consumption.”

For most of the year, the weather in Iowa is cool enough that Microsoft can use the outside air to keep the supercomputer running properly and let the heat out of the building. Only when the temperature exceeds 29.3 degrees Celsius (about 85 degrees Fahrenheit) does it remove water, the company has said publicly.

It can still be a lot of water, especially in the summer. In July 2022, a month before OpenAI announced it had completed its GPT-4 training, Microsoft pumped about 11.5 million gallons of water into its Iowa data center cluster, according to West Des Moines Water Works. It was about 6% of all water used in the area, which also supplies drinking water to the city’s residents.

In 2022, a document from the West Des Moines Water Authority said it and the city government would consider future Microsoft data center projects only if those projects could “demonstrate and deploy technology that significantly reduces peak water use from current levels” to preserve water supply for residential and other commercial needs.

Microsoft said Thursday that it is working directly with water utilities to respond to their feedback. In a written statement, the water utility said the company has been a good partner and has worked with local authorities to reduce its water footprint while still meeting its needs.