AI Data Center Sustainability: What's Real Progress And What's Still Just Talk

Every major tech company is racing to build more data centers to keep up with AI demand. For instance, US hyperscalers are planning as much as 725 billion dollars in capital expenditures this year alone just to keep pace. Worldwide data center electricity use is expected to nearly double by 2030, rising from about 415 terawatt-hours in 2024 to roughly 945 terawatt-hours. That growth is putting real pressure on local power grids and water supplies.

This is why AI data center sustainability, the effort to cut the energy, water, and land impact of these facilities, is now getting real attention. But opinions divide sharply on how much progress has actually been made. One side calls it all greenwashing, meaning companies exaggerate their environmental efforts to look better than they actually are. The other insists the industry already has it under control.

Want to know more? Read on as we cover:

• Why AI data centers use so much power and water

• What companies are actually doing to fix it

• Where the progress falls short

• What closing the gap actually requires

By the end, you will know whether the industry's sustainability efforts are keeping pace with AI's growth, or falling behind it.

Why AI data centers use so much power and water

So why exactly do data centers use so much power? That jump from 415 to roughly 945 terawatt-hours by 2030 comes down to the hardware. AI runs on graphics processing units, or GPUs, chips originally built for rendering images that turned out to be well suited for the huge number of calculations AI needs to run at the same time. A single GPU can draw 700 to 1,200 watts, compared to 150 to 200 watts for a standard server chip. Running thousands of these chips nonstop at full capacity drives that kind of surge.

That nonstop operation also generates a lot of heat, and chips can get damaged if they run too hot. To manage that heat, many facilities use water-based cooling, where water absorbs the heat and evaporates into the air instead of getting reused. That evaporation is why cooling draws so much water, especially in hot or dry regions where facilities need to cool harder just to keep pace.

What companies are actually doing to fix it

Several of the largest AI companies have made specific, public commitments to cut their energy and water use:

• Amazon reached 100 percent renewable energy across its global operations in 2023, buying power directly from new solar and wind projects instead of relying on the grid alone.

• Microsoft has contracted over 10 gigawatts of renewable energy capacity through similar long-term deals, and hit a self-set goal called water positive in its 2025 fiscal year, meaning it replenished more water than its facilities withdrew across its global operations.

• Google runs its data centers at a fleet-wide average power usage effectiveness (PUE) of 1.09, a facility efficiency score well below the industry average.

• Google, Microsoft, and AWS have all moved to direct-to-chip liquid cooling in their newer facilities, which cools chips directly with liquid instead of chilling large volumes of air. Closed-loop systems like these can cut freshwater use by up to 70 percent compared to older setups.

• Researchers backed by the European Commission found that data center waste heat, if captured and reused instead of vented away, could remove half a kilogram of CO2 and generate half a kilogram of water for every kilowatt-hour of computing energy used.

Where the progress falls short

The solutions covered above do cut how much energy and water each facility needs, but total demand for AI computing is growing faster than those cuts can offset, so overall energy and water use keeps climbing anyway.

Consider the following:

• US data centers' direct water use could climb from 17 to 19 billion gallons a year today to 60 to 110 billion gallons by 2030, even accounting for the efficiency improvements companies have already made. A single large facility can already use five million gallons a day, roughly what a town of 50,000 people uses.

• A peer-reviewed study in Nature Sustainability found the AI server industry is unlikely to meet its own net-zero targets by 2030, meaning the point where a company removes as much carbon as it emits, without heavy reliance on carbon offsets and water restoration projects, both of which carry real uncertainty about whether they deliver the results claimed.

• Most companies still do not publicly disclose their full water, carbon, and land footprint. In June 2026, the United Nations launched an initiative calling on every major AI company to measure and disclose that information, since voluntary reporting has been inconsistent across the industry.

• The unresolved water strain is showing up on the ground too. In the first quarter of 2026 alone, community pushback over local water use led to the cancellation of 20 proposed data center projects worth a combined 41 billion dollars, a sign that the fixes covered above haven't reassured the communities living closest to the impact.

What closing the gap actually requires

Fixing this will take more than individual company pledges. It requires changes at the policy and planning level that no single company can make on its own, such as:

• Turning voluntary disclosure into a real requirement. The UN's transparency initiative asks companies to report their footprint, but it carries no legal weight until individual governments turn it into binding regulation. The EU is already moving in that direction through its Water Resilience Strategy, which proposes mandatory water-saving measures for data centers instead of voluntary targets.

• Better site selection. Many data centers still get built wherever land is cheap, even in areas already short on water or grid capacity. The National Renewable Energy Laboratory has mapped where transmission lines and data center locations overlap, so future projects can be steered away from regions already under strain.

• Weighing tradeoffs together, not one resource at a time. Data centers that still pull power from the regular grid often rely partly on natural gas, and natural gas power uses 50 times more water than solar. Switching to wind to avoid that water use requires far more land instead, about 4 times more than solar and 42 times more than natural gas. There's no single fix that avoids every tradeoff, which is why energy, water, and land need to be planned together instead of separately.

Final thoughts

AI data center sustainability sits somewhere between the two camps from the intro. The renewable power deals and the shift to liquid cooling are measurable changes, not marketing spin. But they have not kept pace with how fast AI infrastructure keeps expanding.

The tools to close that gap already exist: mandatory disclosure, smarter site selection, and integrated planning across energy, water, and land. What is missing is the pace to match AI's growth. Whether that catches up will decide if AI data center sustainability becomes a real turning point or just a talking point.