
Information Technology

Modern data centers face a significant thermal challenge as GPU server power demands skyrocket. NVIDIA's GB200 and GB300 NVL72 racks, for instance, now consume 130 to 140 kilowatts—loads that far exceed the capabilities of traditional air cooling, which typically hits a limit at roughly 35 kilowatts. Consequently, cooling management has transitioned from a routine facility concern to a critical boardroom priority. Because current chips like the H100 and H200 consume over 700 watts each, failure to implement advanced cooling strategies—such as direct-to-chip cold plates or immersion cooling—can lead to immediate cluster failure. As power demands for future generations of chips are projected to reach 240 kilowatts per rack, organizations must shift toward these advanced technologies to ensure compute reliability and operational efficiency.


Subscribe to our Newsletter