Blog

Behind the Data: Unveiling the Water Footprint of Artificial Intelligence

 
11 Nov 2023  |  Amber Walsh

Artificial intelligence (AI) is all the rage these days—but when consumers type a question into ChatGPT, they aren’t thinking about water. ChatGPT, a generative AI tool with 1.4 billion site visits in August 2023 alone, uses an estimated quarter of a gallon of water for every 40 to 100 queries it receives, according to a study by UC Riverside. In addition, new AI models and algorithms need to be ‘trained,’ or provided with large quantities of data to learn how to complete specific tasks. To train ChatGPT-3, for example, more than one million gallons of water were used.

But where is this water use coming from?

When you ask ChatGPT where its water use comes from, the model states, “Using ChatGPT or any text-based AI model like me does not directly lead to increased water usage. Text-based AI models run on computer servers and do not have a direct impact on water consumption. However, the operation of data centers and the infrastructure that supports AI models can have environmental impacts, including energy consumption and cooling requirements.”

So, in short, the answer is data centers. Data centers store and transfer data needed for AI models to run. Data centers already consume large amounts of water, using millions of gallons per day—largely for cooling servers. As a result of the heightened demand placed on data centers from running AI models, Bluefield projects that the water use of the global data center industry will grow at a compound annual growth rate of 5.6% through the end of the decade, making it one of the fastest-growing industrial water verticals.

A few takeaways to keep in mind when considering the water footprint of data centers:

Hosting AI models boosts tech companies’ water demands. Over the last few years, leading data center companies—Microsoft, Meta, and Google—have seen jumps in water withdrawals as their new hyperscale data centers go online. Microsoft’s water usage, for example, increased by approximately 33% between FY 2021 and FY 2022. That rapid growth has concurrently sparked a public backlash over environmental concerns—from the revelation that a Microsoft data center in the Netherlands consumed more than four times as much water as initially projected to Google’s attempt to keep its Oregon-based data center water usage private.

Technology companies implement new strategies (e.g., reuse, leak detection) to address increasing water needs. Leading data center operators like Amazon Web Services (AWS), Google, Meta, and Microsoft have announced commitments to replenish more water than they consume by 2030 to counterbalance their rising water usage. Companies have taken various approaches to achieve these targets, from investing in off-site water replenishment projects to deploying more on-site water-efficient technologies. Bluefield has identified 74 water optimization projects across leading tech sector players, with the most common projects being improving operating efficiency measures (34%), investing in restoration projects (19%), and implementing water reuse systems (18%).

AWS, for example, aims to meet its water-positive goal by investing in water reuse across the supply chain. Currently, 20 AWS data centers use reclaimed water as an intake source for cooling water, reducing demand for local potable supplies. Additionally, AWS has invested in advanced on-site water treatment to allow cooling water to be reused multiple times internally, reducing water intake needs.

Data centers are a key driver of public-private partnerships. Data center operators are increasingly partnering with municipalities to cost share and expedite mutually advantageous water projects. The approach highlights how the convergence of data centers’ ESG goals, high water demand, and capital availability can benefit municipal water providers. The private-public partnership strategy is not a new concept in the water industry, as seen in the desalination and wastewater treatment plant market of emerging countries (e.g., Mexico). Large tech companies can provide new sources of private funding to help close the public water infrastructure investment gap.

Leading data center companies have invested in community water management improvement projects:

  • Meta is currently funding a ‘Waterline Resiliency Project’ in Prineville, Oregon. The fire-protection project is set to expand the system’s water capacity and pressure, mutually benefiting Meta and the surrounding community. Meta needed increased water pressure due to high fire-safety standards, agreeing to fund the project with the city responsible for construction.
  • In April 2023, Microsoft partnered with FIDO Tech, a digital technology provider with leak detection capabilities. The project is set to identify leaks in over 350 kilometers of Thames Water’s pipe network. Microsoft will subsequently count the water saved from fixing leaks toward its corporate water replenishment targets.

Demand for data centers has skyrocketed over the past decade as the digital revolution proceeds apace. The recent surge of interest in AI—especially generative AI applications like ChatGPT—will further fuel the data center sector’s growth as well as its swelling water footprint. If big tech firms hope to mitigate negative environmental impacts and meet corporate water targets, they will need to allocate significant capital to innovation, infrastructure build-out, and new technology adoption, opening the door to new opportunities for water utilities and technology providers alike.