Why doesn't anyone consider cooling servers and CPUs with a cryogenic cooler that uses the heat generated by the servers/CPUs and produces cooling on the spot -- i.e., within the racks?
Cryocoolers, or cryogenic coolers, are heat engines based on the principles of Stirling engines; they've been used for decades in many industries. A Stirling-engine-based cryogenic cooler is a theoretical option for the data center, but they need to be large enough to be effective. And with increasing densities of equipment such large CPU-mounted systems are not acceptable.
Cryocoolers, however, can be useful for in-cabinet or in-row cooling. But the problem is being able to route enough hot air to the cryocooler to run it effectively, and then use the cold junction generated to create a closed-loop cooling system. This would require more powered fans and ducting to ensure air-flows are working adequately.
A power-free type of Stirling engine system needs a temperature differential to work effectively; ambient temperatures of 77 degrees Fahrenheit (25 degrees Celsius), for instance, will require a warmer air temperature to drive the cycle that will stress the components of a data center environment.
Applying power to drive the compressor provides a standard refrigerant approach, as with the Cold Head StirlingCool-10, a two-part assembly that needs 400 W to operate at full capability and is rated for cooling a 10-W thermal load to 80 degrees Kelvin. Server chips have higher thermal output than this (around 80 to 150 W), but don't need as much cooling.
Stirling Cryogenics -- the spin-off from Philips N.V. that began manufacturing cryogenic coolers in 1960 -- focuses on liquefaction and other low-temperature systems; it doesn't seem to have any capability for data center cooling. At the moment, I would say that the costs of implementing a Stirling-engine-based cryocooler system would be far too high and with little overall benefit for a data center.
This was first published in July 2013