The heat you have available in a data center is pretty low-quality (cold) heat. If you’re not familiar with the field, a (very) basic introduction is looking at the Carnot efficiency: In principle, you could increase the pressure in the water with a pump, then let it evaporate, before extracting work in a turbine. Then, you condense the steam (by heat-exchanging with the ambient) before sending it back into the pump.

Now, if this process is ideal (frictionless pumps and turbines, perfect heat exchangers, etc.) we can figure out how much of the heat energy that can be converted to useful work (turbine output - pump input). Assuming the ambient (our cold side) is about 25 C, and the racks we’re cooling (our hot side) operate at around 100 C, we get a Carnot efficiency of about 0.2. That means only 20 % of the heat can actually be converted work. Again, this is the ideal case. It is not thermodynamically possible to get better than this. Realistically, you could maybe get 10 % or something.

So, bottom line: The racks aren’t really hot enough to extract meaningful work. A better proposal would probably be to build things like this in places that are cold and require heating, so that you could use the waste heat as a district heating source. In that case, you could more or less completely eliminate the need for other heating sources in homes (which are far too often electrical). Then, we would be using the electrical power (which is high-quality) for something “useful” (disregarding whether or not a data center is useful in the first place), and use the low-quality heat for what it does best (heating things to moderate temperatures).