By Timothy Prickett Morgan
Publication Date: 2025-11-13 17:53:00
Space has always been at a premium in the data center, but there is pressure – literally – to increase GPU and XPU processing density not only because physical space is expensive, but because latency may be more expensive. The closer computing engines and their components can be brought together, the lower the latency between them and the greater the theoretical boost in utilization of those expensive resources.
Hence the mad engineering race to go from 30 kilowatts in a rack to 1 megawatt in a rack and the absolute inevitability of through-rack liquid cooling (it’s already happened) and silicon photonics for rack interconnects (scheduled for production in perhaps 2028 or 2029, depending on who you ask).
As the largest OEM supplier of traditional supercomputing equipment, largely due to its acquisitions of Compaq, Silicon Graphics and Cray and IBM’s abdication of the HPC market, Hewlett Packard Enterprise has its own ideas about how to build dense…
