Autodesk Joins WEF Water-AI Nexus Council to Curb AI’s Water Drain

Autodesk has joined the Water-AI Nexus Center of Excellence Advisory Council, the body convened by the Water Environment Federation to put artificial intelligence to work on water systems while curbing AI’s own water footprint. The seat places the design-software firm alongside Amazon, the University of Pennsylvania’s Water Center, and the Leading Utilities of the World, partners who launched the council on September 25, 2025.

The timing is sharp. United States data centers pulled an estimated 17.4 billion gallons of water in 2023 and could draw between 38 and 73 billion gallons annually by 2028. Utilities, meanwhile, lose treated water through cracks the size of pencil leads, a problem machine learning is starting to crack open.

Autodesk’s appointment sits at the exact spot where those two pressures meet. The company sells the modeling software engineers use to plan, build, and run water networks. It also runs cloud services hosted in the same kind of facilities fueling the demand surge.

Inside The Coalition Autodesk Just Joined

The Water-AI Nexus Center of Excellence formed in a September 2025 founding announcement from WEF, Amazon, and Penn’s Water Center as a knowledge hub bridging utilities, technology providers, academia, and finance. Its remit is unusually wide. The council must accelerate AI adoption inside water utilities while pushing the AI industry to disclose and reduce the water it consumes for cooling and chip fabrication.

Autodesk now sits at the table with hyperscalers, university researchers, and senior utility operators from the Leading Utilities of the World network. The council operates as an advisory body rather than a regulator. Its outputs are roadmaps, technical guidance, and pilot case studies that members agree to road-test in their own operations.

Ralph Exton, executive director of the Water Environment Federation, framed the council’s logic when the Autodesk appointment was confirmed.

AI has the potential to transform how we manage water, from improving system performance to informing long-term resilience. At the same time, we must ensure that the growth of AI is aligned with responsible water use. The Water-AI Nexus is focused on advancing both, and supporting innovation while safeguarding one of our most critical resources.

AI’s Thirsty Side Of The Bargain

The reason a council like this exists is straightforward. AI workloads pull serious volumes of water through evaporative cooling towers and through the supply chains that fab chips. A UK government technical report on water use in AI and data centres compiles the consumption picture across regions and shows draw concentrating where reservoirs are already running hot and dry.

Five numbers anchor the scale of the problem.

• 17.4 billion gallons were directly consumed by US data centers in 2023, according to consumption inventories cited in industry reports.
• 38 to 73 billion gallons is the projected annual range for US data centers by 2028 if AI demand keeps climbing.
• 6.1 billion gallons went through Google’s data centers in 2024, up from 4.3 billion in 2021.
• 519 milliliters is the water cost UC Riverside researchers attached to a 100-word AI prompt, roughly one plastic bottle.
• 500,000 gallons per day is the draw at Meta’s Newton County, Georgia data center, about 10 percent of the county’s entire water use.

What Machine Learning Already Does For Pipes

Run the same calculator the other way and AI looks like a gift to utilities. American water systems lose roughly one in six gallons before it reaches a tap. Replacing every aging main is impossible. Finding the worst pipes before they fail is the next-best move, and machine learning has gotten very good at it.

Generative adversarial networks now compare live pressure readings against synthetic normal patterns to flag leaks in real time. Peer-reviewed work in the journal Water on early leak and burst detection reports detection accuracy near 70 percent on operational data, enough to dispatch a crew rather than a hunch.

Predictive models pair sensor streams with pipe age, material, soil chemistry, and break history. They rank thousands of segments by likelihood of failure. Utility planners then turn that ranking into capital budgets, dropping crews on the riskiest stretches before the next freeze cycle.

The InfoDrainage Machine Learning Deluge tool is one example already in the field. Project Centre, a UK engineering consultancy, uses it to identify flooding hotspots at the design stage and steer resilience work toward the right streets. Itron’s 2026 Water Utility Trends outlook calls real-time data and AI the defining shift of the next budget cycle, with utilities moving from reactive repair to condition-based maintenance.

The Software Stack Autodesk Brings To The Table

Autodesk’s water portfolio is what makes the council seat meaningful. The 2026 release year, detailed on Autodesk’s One Water blog overview of the 2026 lineup, threads machine learning into the same products engineers already use to design networks.

The lineup overlaps with the council’s stated priorities at several points.

Connected data is the throughline. Autodesk’s pitch to the council is that planning, design, construction, and operations should share one model so machine learning has clean inputs to learn from. Disconnected systems force AI to guess; integrated ones let it predict.

How Outside Researchers See It

Independent voices on AI’s water cost have grown sharper in the past two years. Computer scientist Shaolei Ren and colleagues at the University of California, Riverside, drew the early line with a paper titled Making AI Less Thirsty, calling out what they described as “the secret water footprint of AI models” and pricing a short ChatGPT exchange at roughly half a liter of evaporated water.

That framing now echoes across policy papers and corporate disclosures, including Amazon’s Water-AI Nexus roadmap report on water sustainability, which lays out hyperscaler commitments to be water positive by 2030 in the regions where they operate. The Autodesk seat reads, in that context, as a useful counterweight, a software vendor whose direct customers are the utilities most exposed to data-center build-out next door.

Frequently Asked Questions

Does AI Help Or Hurt Water Supplies On Net?

Both, and the math depends on geography. Inside a utility, machine learning can cut non-revenue water losses by double-digit percentages and stretch existing supply. Outside the fence, a hyperscaler data center can pull 500,000 gallons a day in a stressed county. The Water-AI Nexus exists because the same technology does both at once, and decisions on siting, cooling design, and reuse credits decide which side wins locally.

Is Autodesk Paying To Sit On The Council?

WEF has not disclosed a fee structure, and council membership is described as an advisory role rather than a sponsorship. Members commit time, technical contributions, and pilot deployments rather than dues. The composition mirrors WEF’s other technical councils, where founding partners like Amazon contribute funding and providers like Autodesk contribute software, training data, and engineering hours. Utilities looking to engage can reach WEF directly through the wef.org membership portal.

When Will The Council Publish Recommendations?

The Center of Excellence has signaled an ongoing publication cadence rather than one big deliverable, with research briefs and pilot case studies rolling out through 2026. Public-facing material is being aggregated at water-ai-nexus.org. Utilities looking to apply findings can subscribe to WEF technical updates, and council member firms typically circulate draft recommendations to their utility customers ahead of public release.

Can A Small Utility Use AI Without A Big Budget?

Yes, and most do it through vendor platforms rather than in-house data science. Cloud-hosted analytics tools, including the operational layer in Autodesk’s Info360 product family, charge per asset or per signal rather than requiring a custom build. A starter implementation typically maps SCADA data into the platform, runs anomaly detection on existing sensors, and earns its keep on avoided emergency repairs within the first budget year.

Council watchers will want to track two things over the next 12 months. First, whether hyperscaler members publish water-use figures for AI training runs at the same granularity they publish for total operations. Second, whether utility members start citing measurable non-revenue water reductions tied to council-shared tooling.

Autodesk is talking less about transformation and more about plumbing, the digital kind. That’s the right register for an industry whose biggest wins still come from finding the leak before it floods the basement.