The advantage of applying immersion cooling is mostly spatial [1]: it saves space by allowing a very high energy (hashing) density, and it loosens the constraints on where the miner rigs can be deployed. However, space doesn't seem to be a bottleneck as of now, but it might be in the (near) future. With immersion cooling the cost of the cooling fluids will become a nonnegligible (if not major) part of the whole operation, a consistent and calculated design of the PCB boards and the metal cooling tanks can minimize the consumption of cooling fluids and thus reduce costs. So, there's a lot of work to do other than manufacturing efficient hashing chips.
[1] Google data centers can achieve a PUE (Power Usage Effectiveness) of less than 1.06 [2], so I don't think a PUE of of less than 1.02 [3] achieved with immersion cooling is spectacularly impressive.
[2] https://www.google.com/about/datacenters/efficiency/internal/
[3] http://www.allied-control.com/publications/Full_Version_Bitcoin_2-Phase_Immersion_Cooling_and_the_Implications_for_HPC.pdf
[1] Google data centers can achieve a PUE (Power Usage Effectiveness) of less than 1.06 [2], so I don't think a PUE of of less than 1.02 [3] achieved with immersion cooling is spectacularly impressive.
[2] https://www.google.com/about/datacenters/efficiency/internal/
[3] http://www.allied-control.com/publications/Full_Version_Bitcoin_2-Phase_Immersion_Cooling_and_the_Implications_for_HPC.pdf
Saving 5% on electricity is definitely important for any large datacenter and the methods google uses to cool are very extreme and expensive.
The average datacenter would save something like 20% on electricity costs in addition to the high density as you mentioned.
this time 4chips
