A new framework is emerging to measure how much water artificial intelligence systems consume, addressing a gap that energy efficiency alone cannot solve. As AI data centers grow larger and more powerful, water consumption has become a critical environmental concern that organizations are only beginning to measure and manage systematically. Unlike carbon emissions, which have global impacts, water stress is deeply local, meaning a liter of water consumed in a drought-prone region carries far greater environmental weight than the same amount in a water-abundant area.

Why Is Water Becoming as Important as Energy in AI Sustainability?

For years, the computing industry focused almost exclusively on energy efficiency and carbon emissions when evaluating environmental impact. Yi Ding, an Assistant Professor at Purdue University and Project Lead for the Green Software Foundation's Software Water Intensity (SWI) project, explains that this narrow focus missed a crucial piece of the sustainability puzzle. "Energy has always been a central sustainability metric in computing, so naturally that's where most of my work began," Ding noted. "But after collaborating with environmental sustainability experts, I realized sustainability is much broader than just carbon or energy".

The realization came when Ding and her team developed SCARF, one of the first frameworks for evaluating water impacts across computing systems, AI services, data centers, and semiconductor manufacturing. The research revealed something surprising: water's environmental impact depends heavily on local water stress conditions. This insight exposed a critical blind spot in the industry. Organizations could report their water consumption accurately, but without understanding the local context, those numbers meant very little for actual environmental decision-making.

Water is essential for multiple stages of AI infrastructure. It cools data centers, supports electricity generation, and is used extensively in semiconductor manufacturing. As demand for AI continues to grow, organizations are building larger data centers and deploying increasingly powerful hardware, naturally raising questions about water consumption and withdrawal. The computing community has made significant progress measuring carbon emissions over the past decade, but almost no one was asking how to evaluate AI's water footprint systematically.

What Makes Water Different From Carbon in Sustainability Metrics?

The fundamental difference between water and carbon creates unique measurement challenges. Carbon emissions disperse globally and have uniform impacts regardless of where they're released. Water, by contrast, is hyper-local. A data center consuming water in California faces a completely different environmental reality than one in the Pacific Northwest, where water is abundant. This geographic specificity means organizations need metrics that go beyond simply reporting water volumes and instead reflect local water stress conditions.

The SWI project is addressing this challenge by developing a practical standard that organizations can use to measure and report their software's water footprint. Ding emphasized that the biggest barrier to improving sustainability isn't necessarily technology; it's measurement. "Engineers and scientists are very good at optimizing what they can measure. If we don't have the right metrics, it's extremely difficult to make informed engineering decisions or compare different solutions fairly," she explained.

Ding

How Is the SWI Standard Being Developed?

Creating an industry-wide water measurement standard requires careful consideration of competing priorities and technical tradeoffs. The SWI project is navigating several critical decisions that will shape how organizations measure and report water impact:

• Consumption vs. Withdrawal: Water consumption represents water that is no longer available to the local watershed, while water withdrawal reflects the total amount of water taken from the environment, even if much of it is later returned. Some cooling technologies have high withdrawal but relatively low consumption, while others show the opposite pattern. The challenge is that these two metrics answer different questions, so the project must carefully evaluate whether SWI should report one, both, or guide on when each is appropriate.
• Water Stress Indicators: Several well-established water stress indicators exist, such as AWARE and the WRI Baseline Water Stress Index, but they were developed for different purposes and represent different aspects of water scarcity. The SWI project is carefully evaluating which approach best aligns with its goals and how water stress should be incorporated into the metric.
• Scope and Practicality: The project aims to make SWI scientifically robust while keeping it practical enough to support real-world adoption. Balancing rigor with usability is essential for encouraging organizations to actually implement the standard rather than treating it as an academic exercise.

Ding noted that developing an industry standard requires collaboration across disciplines. "We are working closely with researchers, sustainability experts, and industry practitioners to ensure that SWI is transparent and practical for real-world use. The goal isn't simply to choose a methodology, but to build a standard that organizations can trust and consistently apply," she stated.

Ding

What Impact Could SWI Have on AI Sustainability?

Looking ahead, Ding envisions SWI playing a role similar to what the Software Carbon Intensity (SCI) specification has played for carbon emissions. Just as SCI helped organizations measure, report, and reduce emissions, she hopes SWI will enable the same kind of progress for water. Within five years, she expects software teams to think about water intensity as naturally as they currently think about energy efficiency or carbon impact.

The broader ecosystem is already recognizing water as a critical sustainability metric. Over the past few years, water has moved from being a niche research topic to an issue that sustainability leaders, infrastructure teams, policymakers, and the general public are actively discussing. Companies have become more sophisticated in how they approach sustainability, and many organizations are now asking what other environmental impacts they should measure beyond carbon. Water has become a natural next step because of its essential role in electricity generation, semiconductor manufacturing, and data center cooling.

The development of SWI represents a significant shift in how the AI industry approaches environmental responsibility. By creating a shared methodology for measuring water footprint, the standard will enable organizations to benchmark progress, make informed engineering decisions, and drive meaningful improvements in water efficiency. As Ding emphasized, "If you're interested in our work, we'd love to have you involved. Whether it's reviewing the specification, sharing feedback, contributing data or use cases, or simply participating in the conversation, collaboration across the industry is essential to building a standard that everyone can rely on".

As Ding