Why hot-water cooling matters for data centers 

• Traditional cooling requires massive chiller plants to refrigerate air and coolants, consuming significant power.
• Liquid cooling with advanced CDUs and cold-plates enables much more efficient heat transfer from GPUs, which allows for coolant temperature to be warmer, up to hot-water at 45°C (113°F), such that it does not need chilling anymore.
• When cooling and power architectures are co-designed, the control logic in intelligent CDUs can allow for communication and coordination with the power systems for even greater levels of efficiency and reliability.
• Hot-water coolant for NVIDIA systems exits at even higher temperatures and is easily cooled by simple dry cooling systems most of the time (no evaporative water cooling needed). 

Next generation CDUs and cold plates for high-power chips

• Intelligent CDUs that coordinate cooling systems co-architected with the power delivery architecture.
• Precision-engineered liquid cooling cold plates with finer features deliver efficient, reliable heat removal for GPUs and central processor units (CPUs) – allowing for hot water use.
• Advanced manufacturing processes will produce high quality cold plates and intelligent CDUs  at large volumes.

Impact – increased AI factory computing power without additional energy demand

• Intelligent CDUs can coordinate coolant and power delivery architecture, delivering greater efficiency and reliability.
• Advanced cold plate designs and cooling loops can capture up to 80–90% of heat with hot water and allowing the data center to use dry coolers with minimal energy.
• This shift moves the chiller energy budget to expanded electrical architecture and AI GPUs, creating customer value.

Data center operators can save 20–40% in energy costs, reinvesting in more electrical infrastructure and GPUs for higher AI throughput—potentially 33% more compute, token per watt without expanding grid capacity. 

Why it matters - these technologies allow data centers to do more with less power 

Intelligent CDUs and hot water cooling will bring dramatic efficiency improvements that’ll enable data centers to move power and cost out of facility cooling, and into the electrical architecture, power distribution and additional GPUs to support denser AI factories. While it may not lead to less power consumption, it’ll mean that AI factories can do more with the power they have.   

More for you