Climate
Privacy
Bob Dylan
Drones
Data Centers
AI Governance
2026 Maine Senate Race
Last Update: 08/01/2026 at 2:01 PM EST
NVIDIA Raises Coolant Temperatures
Coverage from ScienceDirect, The Atlantic, and others
Articles
38
Active Days
50
The Topic
NVIDIA is promoting fully liquid-cooled AI systems that circulate coolant at roughly 45°C to 55°C, allowing dry coolers to reject heat with little or no ongoing onsite water consumption in many climates. The approach could reduce cooling-related water demand, energy use, noise, and rack space for new facilities, but adoption will take time and existing sites will retain older systems. The broader water impact remains difficult to assess because operators disclose limited data and electricity generation can shift water consumption upstream.
First Article: 06/08/26
Latest Article: 07/27/26
Summary
- NVIDIA’s Vera Rubin systems circulate coolant at about 45°C entering and 55°C leaving the racks, enabling heat rejection through outdoor dry coolers.
- Closed-loop liquid cooling can remove evaporative cooling equipment and sharply reduce ongoing onsite water use in suitable climates.
- Higher coolant temperatures may also reduce chiller energy, fan noise, and rack space, but reported savings are company claims and depend on facility conditions.
- The technology primarily applies to new high-density AI systems; existing data centers may need years to adopt comparable designs.
- Water concerns remain because electricity generation can consume substantial water, and onsite-only accounting may understate total demand.
- Research on immersion cooling above 700 W per chip finds that fluid choice, flow confinement, and heat-sink structure jointly determine temperatures, pressure drop, and pumping power.
- Comparable local water-use data remain scarce, complicating assessment of whether individual projects will strain reservoirs or municipal systems.
History
The story has narrowed from a broad industry shift in AI cooling to a specific NVIDIA-led push for higher-temperature, fully liquid-cooled systems that may cut onsite water use. The main new wrinkle is that the water question is now framed more explicitly as a total-footprint issue, because power generation can shift consumption upstream and facility-level data remain sparse.
The story has broadened from a general shift toward liquid cooling into a more specific policy-and-operations issue: water, wastewater, and permitting constraints are now central alongside heat and power. It also adds a sharper contrast between optimistic vendor claims of waterless designs and the continued dependence of existing facilities on conventional cooling.
Featured
07-14-2026
Researchers analyze single-phase immersion cooling for high-heat-flux AI chips, finding flow confinement and open-ratio control can reverse dielectric-fluid performance rankings.
07-16-2026
Kathy Hochul cited water needs in New York during a statewide moratorium as officials in Indiana and Georgia raised concerns about data-center cooling demand and water supply constraints.
06-22-2026
Nvidia announced warm-water closed-loop cooling aimed at reducing data center on-site water use, while analysis links remaining water impacts to USA fossil and hydropower electricity.
Timeline: 50 Days
Hover over any logo to see coverage summary, click for full article.
Additional Articles
⭐⭐⭐⭐⭐
PJM Interconnection and insurers report weather-driven curtailment and losses, while climate studies quantify high heat and flood risks for data center capacity.
Christina Karamperidou and University of Hawaiʻi at Mānoa researchers project increased temperature-humidity stress through mid-century that will limit direct air free cooling for data centers.
Valar Atomics and Nvidia announced Utah plans for a 30-megawatt AI data center using Ward250 nuclear power and water-efficient cooling.
UC Riverside and Caltech estimate U.S. community water systems may need $10 billion to $58 billion in new infrastructure by 2030 as peak-day demand from AI-linked data centers rises.
Val Frenkel and co-authors receive a 2026 American Water Works Association best article award for a data center water-energy framework and mitigation recommendations in the 2020s.
Water submetering guidance recommends cooling tower makeup and blowdown meters to improve WUE calculations and sewer-credit verification for evaporative-cooled data centers.
Researchers develop a robust electro-thermal planning framework for hyperscale AI datacenters to reduce thermal infeasibility during long-term power and cooling expansion.
Christina Karamperidou and University of Hawaiʻi at Mānoa researchers report in Scientific Reports that future humidity and temperature trends will reduce direct air free cooling availability for data centers.
A Guardian analysis links planned U.S. data-center siting in drought areas to larger AI water impacts from chipmaking and power generation.
Valar Atomics and NVIDIA demonstrated a helium-cooled, waterless AI data center concept at a Utah site on Wednesday using a small modular nuclear reactor.
⭐⭐⭐
NVIDIA outlines Rubin AI infrastructure liquid cooling, using closed-loop coolant and outdoor dry coolers to reduce energy, water use, and noise in data centers.
Nvidia announced at London Climate Week that warm recirculated liquid cooling could reduce data-center water consumption for next-generation AI infrastructure.
Nvidia announced Vera Rubin AI server cooling that uses hotter closed-loop coolant and dry coolers to reduce onsite water use in data centers.
Zurich and technology vendors highlight extreme weather and heat-driven cooling and grid stress as data center operators plan updated environmental specifications for Uptime.
NVIDIA reports Rubin AI infrastructure with closed-loop liquid cooling and dry-cooler-based operation to reduce hyperscale cooling energy and water use.
Accelsius research on two-phase coolant systems for Nvidia B200 accelerator racks claims higher facility water temperatures can reduce datacenter energy use, citing hardware and supply barriers.
AirJoule Technologies announced a 15.0 million registered direct offering to commercialize AirJoule Core and AirJoule Prime systems for waste-heat-driven cooling and distilled water production.
Nvidia announced closed-loop liquid cooling for new AI servers, aiming to reduce data center water use as water and power scrutiny increases.
Valar Atomics and NVIDIA plan to scale a helium-cooled waterless AI data center in Emery County, Utah, after reactor criticality in a San Rafael Energy Lab sandbox.
Vertiv expanded thermal management capacity and launched PurgeRite NearZero, aiming to reduce water waste for hyperscale liquid cooling in AI data center buildouts.
University of Illinois researchers propose aquifer thermal energy storage for data center cooling in Illinois to cut electricity use and enable underground heat reuse.
AI data center operators are moving from CFD-only thermal design to liquid-cooling system simulation and digital twins to handle GPU-localized heat loads.
Stantec discusses liquid cooling designs for AI data centers, linking closed-loop heat removal, higher rack power density, and LEED Optimize Energy Performance modeling.
Daikin Industries and NTT DATA Japan begin an AI cooling proof of concept in July 2026 with testing at an NTT DATA data center in Japan through March 2027.
Data-center operators can cut potable water use for cooling tower makeup by capturing rainwater and stormwater, recovering condensate and blowdown, and reusing gray water while meeting plumbing codes.
Motivair reference designs for GPU-based AI factories outline liquid cooling architectures and heat-rejection choices in Dallas and San Francisco.
An AI data-center reporting initiative details water and electricity impacts, cooling-system tradeoffs, and community permitting considerations in US corridors including Ashburn and Loudoun County.
Microsoft, Amazon, and Google water approaches highlight how data-center cooling withdrawals and NPDES-relevant blowdown chemistry affect local water planning in multiple U.S. regions.
Engineers discussed hybrid air/liquid cooling for AI data centers, linking direct liquid cooling thresholds and free-cooling operation to local water-use code constraints.
Water demand forecasts and cooling designs are reshaping AI data center expansion, with waterless two-phase systems reducing water consumption and potential permitting friction.
Parker engineers discuss pumped two-phase cooling requirements for high-density data center racks, emphasizing vapor quality control, refrigerant selection, and low-permeation tubing.
Walsh argues that closed-loop cooling and privately funded power generation and grid upgrades reduce water and grid stress tied to AI data-center expansion.
Nvidia hosted a mini-datacenter tour in Sunnyvale, California, to showcase Vera Rubin NVL72 AI Factory hardware, emphasizing power-constrained tokens per watt metrics and resource concerns.
Vertiv opened a Johor, Malaysia manufacturing facility in 2026 to produce thermal and power modules for AI and high-density computing, targeting full operations in 2027.
Digital Realty operations program manager discusses hot standby cooling and load-aware redundancy to reduce cooling PUE and enable power flexibility as AI raises data center load density.