AI data centers are being criticized for guzzling water, yet NVIDIA says its next AI factory design can cool servers with nearly none of it. That is the tension behind NVIDIA liquid cooling: a real engineering shift, but not a full environmental pardon for AI.
XOOMAR Intelligence
Analyst Take
The claim centers on NVIDIA’s new liquid-cooling approach for next-generation AI infrastructure, which can reduce cooling water consumption to “nearly zero” in certain environments, according to Tom's Guide.
The caveat matters. NVIDIA is talking mainly about water used inside the facility for cooling. AI’s broader water footprint also includes electricity generation, chip manufacturing, construction, and the scale of the buildout itself.
Why AI users should care about NVIDIA liquid cooling now
Every chatbot query, image prompt, coding assistant session, and enterprise AI workflow ultimately lands on servers. Those servers run on dense clusters of GPUs, many designed by NVIDIA, and those chips throw off serious heat.
The old assumption was simple: bigger AI models mean bigger data centers, and bigger data centers mean more local water stress. That assumption has become politically expensive as communities question whether new AI facilities are putting pressure on local resources, especially in regions already facing drought or water shortages.
NVIDIA’s counterclaim is narrow but important. If cooling water is one of the most visible costs of AI infrastructure, then removing most of that water use inside the data center changes the public argument around siting and permitting.
It doesn’t end the argument. It moves it.
For readers tracking NVIDIA’s reach beyond hyperscale training clusters, XOOMAR has also covered how the company’s AI stack is being aimed at finance in Nvidia AI Fraud Detection Hunts $403B Card Crime Rings, and how GPU demand shows up in consumer hardware coverage like 5 Prime Day GPU Deals Slash $175 Off Nvidia, AMD Cards.
How AI data centers became a water problem
AI chips run hot. Dense GPU racks concentrate that heat in a small physical footprint, and operators have to remove it continuously to avoid throttling or hardware failure.
Many data centers have historically used cooling towers that rely on evaporation. Water absorbs heat, evaporates, and carries that heat away. It works well, but it can consume large volumes of water.
Tom’s Guide cites the concern plainly: conventional cooling-tower-based systems can use roughly 2.6 million gallons per megawatt per year for facility cooling.
There are two water stories here:
- Direct water use: Water consumed on site, mostly for cooling.
- Indirect water use: Water consumed elsewhere, including electricity generation and semiconductor manufacturing.
That split is where many corporate claims get slippery. A facility can slash on-site cooling water and still carry a large water footprint through the power it buys and the hardware it installs.
How NVIDIA liquid cooling cuts water inside the facility
NVIDIA’s design changes where heat gets captured. Instead of pushing chilled air across server components, the system moves liquid directly near the hardware.
According to NVIDIA’s own blog, its Rubin-generation infrastructure is designed for 100% liquid cooling, with every chip and networking component cooled by liquid in a closed loop. NVIDIA says there are no fans anywhere in the system.
The coolant is a mix of 75% water and 25% propylene glycol. It flows through cold plates that sit on processors, enters racks at up to 45°C (113°F), and exits at roughly 55°C after absorbing heat from the chips.
That warm operating temperature is the trick. If the coolant is already hot enough, outdoor dry coolers can reject heat like large radiator coils, without relying on evaporative cooling. In favorable climates, NVIDIA says that can cut facility cooling water consumption from roughly 2.6 million gallons per megawatt per year to near zero.
“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,” said Ali Heydari, director of data center cooling and infrastructure at NVIDIA.
NVIDIA also argues that higher liquid temperatures improve energy efficiency. Its blog says cooling alone has historically accounted for up to 40% of a data center’s electricity consumption, and that raising chiller plant temperatures by one degree can cut cooling energy costs by about 4%.
The near-zero claim covers cooling, not AI’s whole water bill
This is the key distinction: near-zero water does not mean zero environmental footprint.
NVIDIA’s claim is strongest inside the data center boundary. The coolant loop is filled once and recirculated. Dry coolers can reject heat without evaporating fresh water. That is a major reduction if the alternative is a cooling tower.
But AI’s total footprint extends past the facility wall. TechCrunch reported that water use outside the data center, mainly from electricity generation and chip manufacturing, can double or triple the total water footprint of a facility. It also estimated that NVIDIA’s solution addresses about a quarter to a third of AI data centers’ total water consumption.
Power source matters. TechCrunch cited figures showing natural gas power plants use 1.17 liters of water per kilowatt-hour generated, while coal plants use 2.2 liters. Wind and solar are far lower in the cited data, at about 0.01 liters and 0.03 liters per kilowatt-hour, respectively.
So the honest read is this: NVIDIA liquid cooling can make AI facilities far less water-hungry on site. It can’t erase the water tied to power generation, chip fabrication, or the decision to build ever-larger AI infrastructure.
A 100-megawatt AI campus shows the scale of the claim
Take NVIDIA’s own conventional-cooling figure and scale it.
A hypothetical 100-megawatt AI campus using cooling-tower-based systems at 2.6 million gallons per megawatt per year would imply roughly 260 million gallons of facility cooling water annually.
With NVIDIA’s closed-loop liquid cooling and dry coolers, that cooling water figure could fall to near zero in favorable climates.
That is not a rounding error. In a water-stressed community, avoiding evaporative cooling could remove one of the most visible local objections to a data center project.
The before-and-after looks like this:
| Cooling approach | Facility cooling water use | Main condition |
|---|---|---|
| Conventional cooling-tower system | Roughly 2.6 million gallons per megawatt per year | Uses evaporative cooling |
| NVIDIA 45°C liquid cooling with dry coolers | Near zero in favorable climates | Needs climate and design conditions that support dry heat rejection |
The example has limits. Actual savings depend on climate, utilization, rack density, cooling design, and electricity source. NVIDIA itself frames the strongest claim around favorable climates.
Liquid-cooled NVIDIA AI data centers still face the power test
NVIDIA’s cooling advance is credible as an engineering answer to one specific problem: water-intensive facility cooling for dense AI hardware. It is especially relevant because next-generation GPU clusters are pushing beyond what traditional air cooling can handle cleanly.
The harder question is whether the industry uses those efficiency gains to reduce total impact, or simply to build more capacity faster.
Microsoft has made similar zero-water cooling claims for its newest data centers, according to the supplied Tom’s Guide material. That suggests the direction of travel is clear: hyperscale AI builders want to make cooling less dependent on local water supplies.
Readers should press every “near-zero water” claim with four questions:
- Boundary: Does the number cover only facility cooling, or the full supply chain?
- Climate: Is the site actually suited to dry cooling most of the year?
- Power: What electricity source feeds the facility?
- Verification: Are the reported savings independently checked?
NVIDIA may have changed the cooling math. The next test is whether AI builders disclose enough data for communities and investors to see the full water and energy math, not just the cleanest slice of it.
Impact Analysis
- NVIDIA's cooling claim could reduce one of the most visible environmental objections to AI data centers.
- The broader water footprint of AI remains unresolved because power generation, chip production, and construction still consume resources.
- Communities weighing new AI facilities may see the debate shift from cooling water use to total infrastructure impact.
AI Data Center Water Use: Traditional Concern vs NVIDIA's New Claim
| Issue | Traditional AI Data Centers | NVIDIA's New Liquid-Cooling Design |
|---|---|---|
| Cooling water use | Criticized for consuming significant local water | Claims cooling water consumption can fall to nearly zero in certain environments |
| Environmental scope | Often judged by visible local water demand | Addresses facility cooling, but not the full AI water footprint |
| Remaining concerns | Local drought and resource stress | Electricity generation, chip manufacturing, construction, and rapid buildout |
Primary Sources & Disclosures
Written by
XOOMAR Insights Team
Research and Editorial Desk
The XOOMAR Insights Team pairs automated research with human editorial judgment. We track hundreds of sources across technology, fintech, trading, SaaS, and cybersecurity, cross-check the facts, and explain what happened, why it matters, and what to watch next. We do not just rewrite headlines. Every article is fact-checked and scored for reliability before it goes live, and we link back to the original sources so you can verify anything yourself.
