AI is set to consume up to 600 billion gallons of water by 2030 — rising energy consumption primarily to blame as data center power demands rise

AI's Thirst for Water: A Growing Crisis

As artificial intelligence continues its rapid expansion into every sector of the global economy, a less visible but critical side effect is emerging: the staggering water consumption required to power and cool the data centers that make AI possible. Recent reports paint a sobering picture of just how much water AI workloads could demand by the end of the decade.

According to a detailed analysis published by Tom's Hardware, drawing on data from the United Nations University and the water technology company Xylem, AI data centers are on track to consume as much as 600 billion gallons of water annually by 2030. That's enough water to supply approximately 500 million people in Sub-Saharan Africa, and represents a global electricity demand exceeding that of the entire country of Nigeria.

Beyond Cooling: The Hidden Water Footprint

The common perception is that data centers guzzle water primarily through evaporative cooling systems. However, the reality is more nuanced. Direct cooling represents only a fraction of the total water footprint. The vast majority — upwards of 80% by some estimates — comes from indirect water usage through electricity generation. Thermoelectric power plants require enormous quantities of water for steam generation and cooling, and every kilowatt-hour consumed by a data center carries this embedded water cost.

In 2023, U.S. data centers alone consumed 17.4 billion gallons of water directly, according to a report from the MostPolicyInitiative. By 2028, that direct consumption could balloon to 73 billion gallons as new gigawatt-scale facilities come online. But when indirect water usage from power generation is factored in, the UNU report estimates that global data center electricity consumption required just under a trillion gallons of water in 2025, with AI workloads accounting for roughly 200 billion gallons — or about 20% of that total.

Exploding GPU Power Demands

The driving force behind this accelerating water consumption is the relentless increase in GPU power requirements. Each new generation of AI accelerators draws significantly more power than the last:

• Nvidia A100 (Ampere): 300–400W TDP
• Nvidia H200 (Hopper): Up to 700W TDP
• Nvidia GB200 (Blackwell): Up to 1,200W TDP
• Nvidia Vera Rubin (Next-Gen): Up to 2,300W per chip

At the rack level, the escalation is even more dramatic. Traditional data center server racks consumed between 10 and 15 kilowatts. The latest GB300 NVL72 designs can pull upwards of 150KW per rack, and future Vera Rubin racks could exceed 230KW each — representing a 15-fold increase in rack-level power density in just a few generations.

Comparing AI's Water Use to Other Industries

While 600 billion gallons sounds immense, context is important. U.S. agriculture used roughly 26.4 trillion gallons of water in 2024 — more than 40 times the projected AI water consumption. Global oil refining consumes approximately 550 billion gallons annually, putting AI's water footprint in a similar league. However, the key difference is growth trajectory: while established industries have relatively stable water demands, AI's consumption is projected to double from 20% to 40% of total data center water use by 2030.

Cooling Innovations: Progress and Limitations

Data center operators are not standing still. Major advances in cooling technology are helping reduce direct water consumption. Microsoft CEO Satya Nadella recently claimed that the company's newest AI data centers can "operate effectively with zero water consumption" using closed-loop, direct-to-chip cooling systems — analogous to AIO liquid coolers in enthusiast PCs. Similarly, immersion cooling and fanless liquid cooling technologies are gaining traction.

However, there's a catch: closed-loop cooling systems typically consume more electricity than evaporative systems. As power demands rise, so does the indirect water footprint. This makes power generation — not cooling — the central challenge.

The Path Forward: Renewables and Nuclear

Addressing AI's water footprint ultimately requires tackling the energy source. Hyperscalers are exploring several pathways:

• Renewable energy: The Switch Tahoe Reno exascale data center runs 100% on solar power at 650MW. Portugal's SINES DC Start Campus (1.2GW) combines renewables with seawater cooling.
• Nuclear power: Smaller modular reactor designs could provide dedicated, carbon-free power for data centers, with some developers even exploring repurposed naval reactors.
• Water recovery: Capturing and reusing water from cooling processes can dramatically reduce net consumption.

The challenge is immense but not insurmountable. With GPU power demands nearly doubling each generation and data center construction accelerating worldwide, the pressure to find sustainable solutions has never been greater. As local communities increasingly push back against new data center developments — particularly in drought-prone regions — the industry's ability to innovate on both energy and water efficiency will determine whether AI's environmental costs become a crisis or a solvable engineering problem.

Image credits: BalticServers data center (CC BY-SA 3.0), NERSC data center (CC0), UNC Data Center (CC BY-SA 4.0) — Wikimedia Commons.