Nvidia Says Its Chips Run on Bathwater. Here's What 'Zero Water' Actually Means

The smarter AI gets, the hotter data centers run. And to cool that heat, the industry is pouring in staggering volumes of water. You’ve probably seen the line by now: every time you query a chatbot, a bottle of water evaporates somewhere. Now Nvidia is playing a new card — a cooling system that, the company says, uses almost no water at all. The catch is buried in that word “almost.”

Why AI Data Centers Drink So Much Water

Start with why data centers became water hogs in the first place. The chips inside a server throw off enormous heat while they work. Fail to cool them and they fry. The traditional fix is air. But AI accelerators run so hot that air alone can’t keep up.

Enter the cooling tower. It pulls heat out by evaporating water — and that’s exactly the problem. The water doesn’t get recycled. It evaporates into the sky, used once and gone. That’s why estimates put a single large AI data center’s daily consumption in the millions of liters.

Locals have every reason to be furious. Drop a data center into a drought-stricken region and suddenly the water people drink is competing with the water that cools GPUs. In dry US states like Arizona and Texas, that fight is already playing out in zoning meetings and lawsuits.

Nvidia’s Answer: Cool It With Hot Water

With its next-generation Rubin chips, Nvidia is pushing direct liquid cooling — running coolant straight across the chip to carry heat away. The twist is the temperature of that coolant.

A widely shared tech video recently pointed out that Nvidia’s coolant runs as warm as 45 degrees Celsius — “hotter than a hot tub,” as the creator put it. The clip pulled in close to 310,000 views, and on first listen it sounds absurd. You’re trying to cool something. Why use warm water?

Here’s the logic. To make cold water, you have to actively chill it — and that chilling step is exactly what sends water evaporating out of a cooling tower. But if the chip is happy being cooled by warm water in the first place, you skip that step entirely. Water at 45 degrees doesn’t need a tower to chill it down. You can dump its heat into the outside air and send it back around. That makes a closed loop possible — water that circulates instead of evaporating away.

What “Near-Zero Water” Really Means

Time to clear up the obvious misread. “Uses almost no water” does not mean a data center uses zero water. It means the warm-water approach nearly eliminates the water lost to evaporation.

A closed loop fills with water once and keeps recirculating it. You still top it off over time, but that’s a different universe from a cooling tower boiling off millions of liters a day. The headline number — a dramatic cut in water consumption — is real.

But there’s a hidden bill. Save water and you spend more electricity. Chilling that warm water with outside air means running chillers and fans. One industry breakdown projected that by around 2028, two-phase cooling — exploiting both liquid and vapor states — will be effectively mandatory. Which is the whole catch: you’re trading a water problem for a power problem.

Greenwashing or Genuine Progress?

So where does that leave us? My read: the direction is right, but it’s no silver bullet.

Cutting evaporation is unambiguous progress. It takes much of the pressure off the drinking-water-versus-GPU fight in drought regions. That’s not something to wave away as greenwashing.

But read the marketing carefully. “Zero water” sounds like the finish line of going green, while the bigger picture — electricity — sits there untouched. The real burden of an AI data center is power, and where that power comes from is what ultimately decides its environmental footprint. When you applaud the water saved, you should also ask how the added electricity gets generated.

One more caveat: this field is still short on verified, large-scale operating data. Once Rubin-class systems are running at full tilt, we’ll get actual water and power figures. That’s when the real scorekeeping begins.

Bottom line: Nvidia’s warm-water cooling is a meaningful answer to the water-scarcity panic. Just don’t forget the footnote under the “we saved water” headline — the one that reads “and used more electricity.” So where do you land? When we tally the environmental cost of AI, which should we worry about first: the water, or the watts?