However, with 200 billion gallons of water (and counting) dedicated to AI workloads, how does that compare to other industries?
That's still a footnote compared to the or 26.4 trillion gallons used by U.S. agriculture in 2024 (as per USDA ), but it is closing in on international oil refining numbers. OPEC produced roughly 86 million barrels of oil a day in 2024, and with a rough conversion of 0.4 barrels of water per barrel of crude, that's in the region of 550 billion gallons a year.
There are still industries using much more water than AI, and much more even than all the data centers combined. However, water use is growing dramatically, and it's not really coming from the increased cooling demands. It's coming from power demands and hardware manufacturing.
As Americans' feelings towards data centers sour and, in turn, increase their opposition to their development, addressing the growing water needs of these new deployments may be key for developers who want to see their projects reach completion. For residents facing water shortages, droughts, and water contamination, the danger is more existential.
For data center developers, the main way they've looked to address growing water demand from their facilities is to improve cooling. Moving away from evaporative to closed-loop, direct-to-chip cooling can have a huge effect on the water used in the cooling process.
AI data center project secretly sucked 29 million gallons of water over 15 months before detected by residents complaining about low water pressure
AI data centers trigger massive 'irreversible' 76% electricity price spike in largest US region
After a $16 billion Stargate AI data center was built despite being voted down, Michigan towns rush to block new buildouts
That's a good thing. Indeed, Microsoft's Satya Nadella claimed that the company's newest AI data centers have cooling systems so efficient that they "can operate effectively with zero water consumption." He also compared these mega data centers to single restaurants in terms of annual water consumption.
Closed-loop cooling is absolutely an important innovation in data center development, and combined with more exotic ideas like fanless liquid coolers and immersion cooling (perhaps even undersea and in-orbit deployments ) could see data center cooling use very little water in the future.
But it's not the cooling that's the problem: it's the power being generated to run them. And the one downside to closed-loop cooling systems is it uses more power than evaporative cooling systems. As those power demands rise, so does the indirect water usage of these facilities.
The vast majority of indirect water use by data centers by 2050 will be down to power generation, the Xylem study suggests, and if recent generations of GPU development are anything to go by, those power demands are going to be enormous.
Nvidia's Ampere generation A100 enterprise GPU had a TDP of 300-400W. An H200 of the Hopper generation has a TDP up to 700W. A Blackwell GB200 GPU can pull as much as 1,200W. The next generation Vera Rubin? That's now up to 2,300W per chip .
Once you start scaling these GPUs up to their full racks, the power consumption is extreme. Where traditional data center server racks consumed between 10 and 15 KW, the latest GB300 NVL72 designs could consume upwards of 150KW a piece. Vera Rubin might be more energy efficient, but its racks could consume upwards of 230KW each.
Data centers just weren't traditionally designed with this kind of density of energy demands in mind. Powering them alone will be an enormous challenge and require enormous quantities of water to do it.
However, other industries also consume far more energy than all data centers, and thus, consume more indirect water usage from power generation. For instance, steel and iron manufacturing, chemicals and petrochemicals, cement and glass manufacturing, and many other industries each use multiple times more power than all data centers combined.
For hyperscalers, the near-term solution to data center power problems has been to use (occasionally law-breaking) mobile methane jet turbines. These aren't too heavy on water consumption, but have their own environmentally damaging effects with heavy carbon emissions.
A longer-term solution to this issue, and hopefully the extreme indirect water demands of these facilities, will be a combination of renewables, nuclear power, and water recovery.
The Switch Tahoe Reno exascale data center shows how it can be done. It's a 650MW facility built in 2017 that uses 100% renewable solar energy. In Portugal, the SINES DC Start Campus is a 1.2 Gigawatt facility that's partially online and uses 100% renewables, as well as using seawater cooling to offset its water usage. Although many of the newer data center projects are far larger, as the price of solar deployment continues to plummet, it's certainly possible that data centers powered by renewables can be effective and profitable, without consuming such vast quantities of water.
Another option is nuclear energy. New, smaller reactor designs are making it possible to get these facilities online faster and with a more modular design. There's even the possibility of repurposing old aircraft carriers and submarines with onboard nuclear reactors to power data center facilities.
Data center developers know this is at least one path for the future. That's why they're rushing to secure access to key materials like Uranium .
The future of data center power, especially AI data centers, and the water that they require, is almost certainly some mix of renewables and near-site nuclear power.
As the mobile methane turbines popping up at data centers have shown, the developers will often just use what they can get their hands on. Perhaps alongside moratoriums and pauses in construction, data center protestors could make sure that if data centers are built in their local area, the developers should also be required to invest in water infrastructure to offset their ever-growing demands, alongside more renewable energy solutions.
Jon Martindale is a contributing writer for Tom's Hardware. For the past 20 years, he's been writing about PC components, emerging technologies, and the latest software advances. His deep and broad journalistic experience gives him unique insights into the most exciting technology trends of today and tomorrow. ","collapsible":{"enabled":true,"maxHeight":250,"readMoreText":"Read more","readLessText":"Read less"}}), "https://slice.vanilla.futurecdn.net/13-4-24/js/authorBio.js"); } else { console.error('%c FTE ','background: #9306F9; color: #ffffff','no lazy slice hydration function available'); } Jon Martindale Freelance Writer Jon Martindale is a contributing writer for Tom's Hardware. For the past 20 years, he's been writing about PC components, emerging technologies, and the latest software advances. His deep and broad journalistic experience gives him unique insights into the most exciting technology trends of today and tomorrow.
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://www.tomshardware.com/tech-industry/SPONSORED_LINK_URL
- https://www.tomshardware.com/tech-industry/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#main
- https://www.tomshardware.com/my-account
- 3D-printed nozzles could revolutionize drug and self-healing material manufacturing — MIT-developed triaxial electrospray design makes cleanroom fabrication opt
- Seoul Purpose: How NVIDIA and South Korea Are Building the Future of AI
- Travlfi JourneyGo 5G mobile hotspot review – Affordably priced, but lacking in performance and features
- Linux developers are using AI vibe coding to keep vintage AMD GPUs alive — R600 driver cleaned up with GitHub Copilot gives HD 2000 to HD 6000 series a new leas
- NVIDIA Partners With Microsoft on Unified Stack for Agentic AI Deployment, From Windows Devices to Cloud to Local
Informational only. No financial advice. Do your own research.