NVIDIA has proposed a liquid-cooling design for AI data centers that it says can eliminate virtually all on-site water consumption for chip cooling, replacing traditional evaporative cooling with a sealed warm-liquid loop that rejects heat directly to ambient air. The design, detailed on its official blog, recirculates the same coolant in a closed loop, cutting on-site water use from millions of gallons per facility to near zero.
Key facts
- What: A new NVIDIA cooling design claims to use almost no water inside the data center, though critics say that's only part of AI's water bill.
- When: 2026-06-25
- Primary source: read the source
Traditional data centers cool hot chips the way a swamp cooler works, by evaporating enormous amounts of water—often millions of gallons a year for a single large facility. As AI compute scales, so does that water demand, creating both environmental and public-relations problems. NVIDIA's alternative cools chips with liquid rather than air, and does it with warm liquid specifically. Coolant runs directly against every chip in a sealed loop and carries the heat away. Because the system is engineered to function even when that coolant is fairly warm—warmer than a hot tub—the heat it carries is hot enough to be dumped into the outside air through simple radiators, the same principle as a car radiator, for most of the year. This matters because the water-guzzling step in traditional cooling is the evaporation used to chill things down; if heat can be rejected to open air instead, both the evaporation and the water it consumes can be skipped.
The payoff NVIDIA claims is dramatic: closed-loop recirculation that consumes essentially no new water for chip cooling, down from the millions of gallons a comparable conventional facility would evaporate. There is an energy bonus as well. Cooling can account for a large share—by some measures close to half—of a data center's total electricity, and running the system warm means the power-hungry chillers can be switched off for much of the year in favorable climates. Less chilling means less water and less power simultaneously.
The environmental footprint of AI has become a competitive battleground, not just an activist talking point, and NVIDIA is positioning itself as the vendor with a sustainable answer—one that builds not just the chips but the blueprint for the building they sit in. As AI data centers multiply, a design that genuinely cuts on-site water use at scale is a real selling point to operators and to the communities and regulators deciding whether to allow these facilities nearby.
The caveat, raised by critics immediately, is a good one. Both TechCrunch and Fortune pointed out that eliminating the water used inside the data center does not eliminate the water used to generate the electricity that powers it. Much of that power still comes from plants that consume large amounts of water for their own cooling—water that doesn't appear on the data center's books but is part of AI's true footprint. "Zero cooling water" is a real and useful efficiency win, narrowly scoped. It is not the same as "zero water," and the larger, system-wide question of AI's energy and water appetite remains very much open.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)