[AI Paradox] Already facing water shortages... 'Groans' heard worldwide due to AI

Increase in Data Centers Due to AI Boom Raises Cooling Water Demand
Big Tech-Local Community Conflicts in US, Uruguay, Spain
Sustainable Solutions Proposed but No Resolution Yet

The artificial intelligence (AI) boom is triggering a 'water war' around the world. Due to the impact of climate change, water is already scarce, and the increased use of data centers for AI has caused a surge in demand for cooling water. As AI demand is expected to grow, conflicts between big tech companies and local communities over water are inevitable worldwide, prompting urgent calls for solutions.

◆ No Drinking Water Amid the 'Worst Drought,' Yet Cooling Water?

According to Bloomberg and The Guardian on the 23rd (local time), conflicts between big tech companies operating data centers and local communities are intensifying in the western United States, where Silicon Valley is located, as well as in Spain and the Netherlands in Europe, and Chile and Uruguay in Latin America. While there is a movement to attract these companies because building major facilities creates jobs and revitalizes the local economy, conflicts arise due to the significant side effect of water scarcity caused by the nature of data centers.

Last month in Uruguay, Google's plan to establish a data center in the southern Canelones region sparked controversy. Uruguay, suffering its worst drought in over 70 years, had difficulty securing drinking water since April, yet news emerged that Google's data center would consume 7.69 million liters (ℓ) of cooling water daily. As soon as this news spread, public anger grew in Uruguay, and the local government announced that Google was revising its plan to reduce the size of the data center.

A similar debate occurred in Talavera de la Reina, a small town in central-west Spain. Meta Platforms plans to invest 1 billion euros (approximately 1.45 trillion KRW) to build a data center expected to use 665 million ℓ of cooling water annually amid a drought lasting over a year. The local government has even urged residents not to water their gardens and to tightly close faucets at night due to the water shortage. Consequently, a movement opposing Meta's data center establishment has arisen, led mainly by local farmers.

A data center of a startup located in Santa Clara, California, USA [Image source=Reuters Yonhap News]

Because data centers consume such large amounts of water, some regions like Ireland and the Netherlands have taken measures to prevent new data centers from being established, Bloomberg reported.

◆ Cooling Water Used When Talking to AI... Annual Consumption of 300 Million Gallons

Water is essential for operating AI. Cooling water is required during the operation of data centers, which are indispensable for AI development. Data center servers generate a lot of heat during operation. Current data centers lack separate cooling systems, so if they are not continuously cooled with water, servers can overheat, causing malfunctions or fires. Large-scale data centers consume between 1 million and 5 million gallons of water daily.

Cooling water is used throughout the AI usage process. Professor Xiaolei Ren of UC Riverside analyzed that Microsoft’s (MS) U.S. data center used about 700,000 ℓ of cooling water per month just to train GPT-3. He also estimated that generative AI like ChatGPT consumes about 500 milliliters (ml) of water during 20 to 50 short conversations with users.

As ChatGPT and various AI technologies enter industrial sites and daily life, water issues have become unavoidable. Data centers are increasing to meet this demand. According to global real estate advisory firm CBRE, the scale of data center development plans in major markets in the second quarter of this year reached 7,242 megawatts (MW), a 158% increase compared to the end of 2020.

Water consumption is inevitable. Global data center water consumption was recorded at 292.2 million gallons last year. Market research firm Bluefield Research forecasts that water usage for these data centers will exceed 300 million gallons this year and continue to rise rapidly, surpassing 400 million gallons by 2028.

The Facebook data center under construction in Utah, USA, in 2019 [Image source=Reuters News Agency]

The situation can be seen even in Google, a leading big tech company directly involved in the generative AI boom. According to Google's '2023 Environmental Report' released last month, the company's water usage surged 20% year-on-year to 5.6 billion gallons last year. Of this, 93% was used as cooling water for data centers. Meta also used 1.33 billion gallons of cooling water in its data centers in 2021.

◆ Big Tech "Considering Solutions"... Cautious About Disclosing Water Consumption

Big tech companies emphasize that they understand the global issues of water scarcity and climate change and are striving to find sustainable solutions.

Google has declared that by 2030 it will replenish 120% of the water used in its data centers. However, as of last year, only 6% of the water used in data centers was replenished. Microsoft (MS) also stated that it is investing in research to develop sustainable and efficient systems. In fact, MS introduced a 'zero water' cooling method using outside air instead of water at its Arizona data center in 2021, but it has been pointed out that this method has limitations as it can only be used when the ambient temperature is below a certain level.

However, some criticize big tech companies for being reluctant to disclose their water consumption. Concerns include the risk of revealing internal secrets to competitors through cooling water usage data and sensitivity to criticism from local communities. According to consulting firm Uptime Institute, only 39% of data centers tracked water usage last year, a 12% decrease from the previous year.

Google, for example, owns a data center in The Dalles, Oregon, USA, and has faced legal disputes over disclosing water usage information there. A local media outlet requested The Dalles city to release Google's water usage data, but the city refused, leading to a 13-month lawsuit. Eventually, the city abandoned its stance and disclosed the information, resolving the issue.