NVIDIA liquid cooling is how Rubin stays running. The system runs at 45 degrees Celsius — 113 Fahrenheit, hotter than most hot tubs. That is not a bug.
NVIDIA announced June 21, 2026 that its Rubin-generation AI infrastructure — part of the DSX AI factory reference design — is the world’s first to achieve 100% liquid cooling. Every chip. Every networking component. No fans anywhere in the system.
The coolant is 75% water and 25% propylene glycol, cycling through a closed loop that gets filled once and runs for the life of the facility. At 45C, it is warm enough to reject heat efficiently in the right climate without ever turning on a mechanical chiller.
Sound too good to be true? The footnotes are worth reading.
What NVIDIA Liquid Cooling Actually Does
The DSX is a reference design — a guide for how to build an AI factory around Rubin architecture. It is not a mandatory spec, but NVIDIA is the only game in town for hyperscale AI compute right now, so the ecosystem follows.
The cooling picture is genuinely different from anything that came before it.
Historically, cooling has accounted for up to 40% of a data center’s electricity consumption. That is not the AI compute workload. That is just keeping the chips from overheating.
Rubin changes that calculus. NVIDIA liquid cooling captures heat directly at the chip via cold plates and transports it through liquid loops running at 45C.
Outdoor dry coolers can reject that heat without mechanical chillers for most of the year in temperate climates. The result: water use drops from roughly 2.6 million gallons per megawatt per year (standard cooling-tower setup) to near zero.
Rack density also increases. Six rack units compress into two. And fan noise disappears — traditional data centers run at 85-plus decibels. Liquid-cooled AI factories run quiet.
Ali Heydari, NVIDIA’s director of data center cooling and infrastructure, put it this way in the official blog post:
“The NVIDIA DSX reference design for AI factories has zero water consumption — we have eliminated massive amounts of power usage and pretty much all water usage.”
Richard Whitmore, president and CEO of Motivair (the advanced cooling division of Schneider Electric), has been working alongside NVIDIA’s roadmap for nearly a decade. His verdict: “Once the watts per chip crossed a certain level, liquid cooling became mandatory.”
That part is real.
The Climate Fine Print
Here is where the “zero water” headline gets qualification.
The DSX architecture achieves chiller-less, zero-water operation in favorable climates. That means temperate regions with dry enough air and low enough ambient temperatures that outdoor dry coolers can do the job.
In Arizona and Nevada, it is a different story. When outdoor temperatures approach 115F — the coolant’s upper limit — chillers still kick in. The system still uses less water than a conventional setup in those climates, but “100% reduction” does not apply.
NVIDIA acknowledged this directly in the blog post:
“A data center in the Scottish Highlands and one in Phoenix, Arizona face very different realities.”
Even in hot climates, the shift to 115F coolant is a genuine improvement. Chillies may run a few days a year instead of constantly. But that is a long way from the headline NVIDIA led with.
The Cost Nobody Disclosed
A 50-megawatt hyperscale facility can save over $4 million annually in cooling-related energy and water costs by moving to liquid-cooled infrastructure, according to NVIDIA.
That is NVIDIA’s number. It has not been independently verified.
NVIDIA did not disclose what the infrastructure costs to build or retrofit. The capital expenditure premium for liquid cooling over air cooling remains unknown. Analyst estimates on that gap do not appear in any public report I could find.
Heat recovery is a real potential upside — waste heat can warm nearby buildings — but that requires geographic pairing that does not exist at every data center site.
The honest read: the operational savings are plausible at scale, but the transition cost is not public, and nobody outside NVIDIA has signed off on the math.
The Bigger Ecological Picture
Cutting water use is real. It does not solve the carbon problem.
On the same day NVIDIA’s announcement dropped, the Wall Street Journal reported that Microsoft had inked a deal with Chevron to build a new gas-powered data center in West Texas.
The two announcements landed on the same calendar date. They tell very different stories about where AI infrastructure is actually heading.
Many AI data centers are still powered by fossil fuels. A facility that uses zero water but runs on natural gas has reduced its environmental footprint on one dimension while doing nothing on another.
The public is paying attention. A recent Pew Research Center survey found that the majority of Americans familiar with data centers view them as mostly bad for the environment.
Google and Microsoft have both announced water-reduction initiatives in response to that pressure. NVIDIA is now making the same move, in public, with a technically impressive announcement timed to land in that exact news cycle.
Space-based data centers came up in coverage as a fringe alternative — heat dumps into the vacuum of space, no water required. That is still theoretical.
What This Means for AI Costs
This is where the story gets practical.
NVIDIA liquid cooling eliminates a massive electricity line item. Cut the fans, reduce chiller dependency, and the operational cost structure of running AI changes at scale. If those savings pass through to token pricing, AI gets cheaper to run.
Rubin’s higher rack density also means more compute per square foot. For hyperscalers burning through hundreds of megawatts, that is real estate and construction savings too.
NVIDIA did not break down the cost savings by geography or climate zone, so it is hard to know which facilities qualify for the full benefit. But the direction is clear: liquid cooling at 45C is a better engineering solution than air cooling at 40C, and it will get cheaper as the industry scales it.
The Bottom Line
NVIDIA’s water-free cooling is a genuine technical achievement. The engineering checks out. The efficiency gains are real. In the right climate, NVIDIA liquid cooling delivers zero water consumption along with lower power bills.
But the announcement is also a carefully timed PR move. It lands the same week as the Microsoft-Chevron gas data center story, in a moment when most Americans think data centers are bad for the environment, and when the AI industry is desperate to show it can be a better neighbor.
The technology is not the hard part. The hard part is deploying it at scale, at what cost, and whether cutting water use gets used to deflect from the harder question of what actually powers these facilities.
I have been tracking NVIDIA’s infrastructure announcements for a while now — including their broader platform play with the Groq inference licensing deal.
DSX is another piece of that strategy. Rubin is not just a chip. It is a whole way of building the factory around the chip.
The fine print on this one is worth knowing.
