A member of Nvidia’s Infrastructure Specialists team questioned Microsoft’s cooling approach during an early-fall 2025 Blackwell installation for OpenAI, writing in an internal email that the design seemed wasteful but offered flexibility and fault tolerance. The assessment concerned two server racks at one facility.
The employee wrote that Microsoft’s “cooling system and data center cooling approach for their GB200 deployment seems wasteful due to the size and lack of facility water use, but does provide a lot of flexibility and fault tolerance.”
The email described an installation of two GB200 NVL72 racks. Each rack contains 72 Nvidia GPUs and uses liquid cooling to remove heat from the processors. The employee’s assessment alone does not establish the efficiency of the facility’s cooling system.
Microsoft said it deploys closed-loop liquid cooling heat exchanger units in existing air-cooled data centers to increase cooling capacity for its own and third-party platforms. The company said this approach allows it to use existing facilities while improving heat dissipation and optimizing power delivery for AI systems.
Cooling servers and expelling heat
Shaolei Ren, an associate professor of electrical and computer engineering at the University of California who studies data center resource use, explained that cooling involves two stages: removing heat from servers and expelling it from the building.
Ren suggested the employee could have been referring to a building-level system that uses air rather than water to expel heat. Such systems tend to use more energy while avoiding water consumption at that stage, he said. His explanation was a possible interpretation of the email, rather than confirmation of the facility’s configuration or measured efficiency.
Microsoft acknowledged a related energy trade-off in its December 9, 2024, announcement about cooling designs for newer facilities. The company said replacing evaporative cooling with mechanical cooling would increase power use, although it expected warmer cooling temperatures and efficient chillers to limit that increase.
Installation required onsite support
The Nvidia employee also described additional work needed to complete the installation. The email said onsite support was necessary and that many hours went into preparing validation documentation and checking that the instructions worked for teams less familiar with cluster and system validation.
Handover processes between Nvidia and Microsoft “required a lot more solidification than what was performed before arrival,” the employee wrote.
The email gave a more positive assessment of the production hardware, describing its quality as better than samples supplied for early customer testing. It said both racks achieved a 100% pass rate on certain compute performance tests. Those results applied to the specified tests on the two racks; they did not establish cooling efficiency.
Nvidia said Blackwell systems deliver strong performance, reliability and energy efficiency across computing applications, and that customers including Microsoft had deployed GB200 and GB300 systems. Nvidia announced Blackwell in March 2024 as the successor to Hopper. The newer GB300 generation was available by December 2025.
Microsoft’s separate cooling plans
Microsoft has committed to becoming carbon negative, water positive and zero waste by 2030. Its December 2024 cooling announcement described a design that continually recirculates water between servers and chillers after an initial fill. Microsoft said the design avoids evaporation and the need for an ongoing fresh-water supply for cooling, while water would still be used in kitchens, restrooms and other administrative areas.
That announcement scheduled pilots in Phoenix, Arizona, and Mt. Pleasant, Wisconsin, for 2026, with new sites using the design expected to begin coming online in late 2027. Those plans were separate from the two-rack installation discussed in the Nvidia email.
