NVIDIA Unveils Full Liquid Cooling with 45°C Coolant Hotter Than Bath Water — Up to 100% Power and Water Savings

NVIDIA announced in June 2026 a full liquid cooling system for Rubin-generation AI servers that uses 45°C coolant. Compared with conventional air cooling or lower-temperature water cooling, the new approach significantly reduces both power consumption and water usage. It offers a practical way to lower operational costs and environmental impact in data centers.

📑Table of Contents

Background and Overview of NVIDIA Liquid Cooling System
Technical Benefits of 45°C Coolant
How Full Liquid Cooling Works and Differs from Conventional Methods
Specific Effects on Power Consumption and Water Usage Reduction
Application to Rubin-Generation AI Infrastructure and Future Outlook
Impact on Data Center Operations and Industry Reactions
Frequently Asked Questions (FAQ)
Summary

Background and Overview of NVIDIA Liquid Cooling System

Power consumption of AI servers continues to rise, with some GPUs exceeding 700–1000W TDP. Air cooling alone cannot handle the heat, and traditional water-cooling systems require substantial electricity and water for chillers and cooling towers. NVIDIA proposes a closed-loop full liquid cooling system using 45°C coolant to address these challenges. According to Gigazine, the system is fanless and efficiently transfers heat from inside the server to the outside.

Technical Benefits of 45°C Coolant

Setting the coolant inlet temperature at 45°C reduces the energy burden on the cooling side. Typical bath water is around 38–42°C, so this is indeed “hotter than bath water.” The higher-temperature liquid cooling allows dry coolers to suffice in more climate conditions, greatly shortening chiller runtime. NVIDIA documentation notes that raising the chiller setpoint by 1°C yields approximately 4% energy cost reduction.

The coolant is a 75% water + 25% propylene glycol mix. Combined with direct-to-chip cold plates, it achieves high thermal conductivity. The outlet temperature rises to around 55°C, but the closed loop re-cools it, minimizing water evaporation losses.

How Full Liquid Cooling Works and Differs from Conventional Methods

Traditional air-cooled servers use many fans to force air circulation, but fans themselves consume significant power and generate noise. Water-cooling systems circulate large volumes of low-temperature water, leading to evaporation losses in cooling towers and high chiller power use.

NVIDIA’s full liquid cooling eliminates all server fans. It uses direct-to-chip cold plates and closed piping. Coolant enters at 45°C, absorbs heat, and exits at approximately 55°C. This warmer fluid can be cooled by outdoor dry coolers, often eliminating the need for water-intensive cooling towers. Gigazine also highlights the ability to increase rack density (for example, fitting what was a 6U rack into a 2U footprint).

Item	Conventional Air Cooling	Conventional Water Cooling	NVIDIA 45°C Liquid Cooling
Fans	Many	Few	None
Coolant Temperature	–	Low (e.g., 20°C range)	45°C inlet
Water Usage	Low	High (evaporation)	Up to 100% reduction possible
Power Reduction	Baseline	Improved	Significant (up to 100% claimed)
Noise	High	Medium	Extremely low

Source: Gigazine (June 24, 2026) https://gigazine.net/news/20260624-nvidia-45-degree-liquid-cooling/ and NVIDIA official announcement

Specific Effects on Power Consumption and Water Usage Reduction

NVIDIA states that cooling can account for up to 40% of data center power consumption. The new system targets major reductions in that portion. For water usage, under conditions where outdoor dry coolers are sufficient, cooling towers can be eliminated, enabling up to 100% reduction. However, these figures are theoretical or calculated values dependent on climate and operational conditions.

In practice, operators will choose regions where dry coolers can run year-round or use chillers only as backup during peak summer. The closed-loop design also reduces the need for water replenishment, lowering environmental impact. NVIDIA suggests potential reuse of waste heat for nearby building heating.

Application to Rubin-Generation AI Infrastructure and Future Outlook

The system targets Rubin-generation AI accelerators that follow the current Blackwell generation. As GPU thermal density increases further in next-generation infrastructure, the limits of air cooling will become more apparent, accelerating the shift to liquid cooling. Deployment is expected from late 2026 through 2027, potentially changing data center design standards.

Impact on Data Center Operations and Industry Reactions

Operationally, the absence of fans dramatically reduces noise in server rooms and eases maintenance workload. Higher rack density allows more GPUs in the same floor space. On the other hand, durability of piping and connectors handling 45°C fluid and leak prevention become even more critical. Across the industry, liquid cooling adoption is spreading, including Microsoft’s two-phase immersion cooling and Google’s TPU deployments.

Frequently Asked Questions (FAQ)

Q: Is this coolant really “hotter than bath water”?

A: Yes, 45°C is set slightly higher than typical bath temperatures (38–42°C). It is a deliberate design choice to enable high-temperature liquid cooling suitable for servers.

Q: Is the claim of up to 100% power reduction realistic?

A: NVIDIA claims major efficiency gains over conventional systems. The 100% figure represents peak or condition-specific calculations. Actual savings vary with climate and operations.

Q: What are the benefits of fanless full liquid cooling?

A: Eliminating fans reduces power consumption and noise. The closed-loop design also minimizes water evaporation losses, creating a quieter and more energy-efficient data center environment.

Q: What is the Rubin generation?

A: It is NVIDIA’s next-generation GPU architecture following Blackwell, delivering higher-performance AI accelerators. Higher TDPs will require advanced cooling solutions.

Q: When will this technology be deployed?

A: Deployment for Rubin-generation infrastructure is anticipated from late 2026 through 2027. Actual adoption depends on data center operators’ plans.

Related articles:

Summary

NVIDIA’s 45°C liquid cooling system provides a practical approach to reducing power and water consumption in AI data centers. Based on Gigazine reporting and NVIDIA official information, the fanless closed-loop design offers significant savings under suitable climate conditions. It serves as a useful reference when planning or operating data centers.

For more details, refer to: – Gigazine: https://gigazine.net/news/20260624-nvidia-45-degree-liquid-cooling/ – NVIDIA official: https://blogs.nvidia.com/blog/liquid-cooling-ai-factories/

Author

krona23

Over 20 years in the IT industry, serving as Division Head and CTO at multiple companies running large-scale web services in Japan. Experienced across Windows, iOS, Android, and web development. Currently focused on AI-native transformation. At DevGENT, sharing practical guides on AI code editors, automation tools, and LLMs in three languages.

DevGENT about →

📚 Read Next