Servers have widely varying requirements for their environment

One of the things that I usually don't have much reason to think about is how servers can have varying requirements for the environment you put them in. Once upon a time this would have been quite relevant at work, because back in the days servers (then often called 'minicomputers') were physically quite large and had significant power requirements. You didn't power a (large) VAX or a serious Sun server by plugging it into a standard 120V wall outlet, for example. As time went by both servers and their environmental requirements shrunk and by no later than the mid 2000s the typical servers we bought ran off 120V power and didn't need anything more than being put in a machine room where their fans wouldn't bother people around them.

That's still the case for the typical servers we use, but it's increasingly not the case for all of the servers in the department. What changed the trend is simple; the advent of GPUs and especially multi-GPU servers vastly increased the amount of power that could sensibly be used in a relatively small amounts of physical space. This power often can't reasonably be delivered through plain 120V circuits, even with multiple 120V single phase PSUs (Power Supply Units) in the server. Instead you're looking at one or more high capacity, higher voltage and perhaps multi-phase circuits that can feed enough power to all of the server PSUs. The PSUs may still be 120V single phase (although probably 20 amps or more), but you'll use the big circuits so you don't have to have an entire breaker box for one server.

(Some of the servers we have can use 15,000 watts at full power, and a single 120V 15A circuit can apparently sustain about 1400 watts of power. A 120V 20A circuit gives you about 1900 watts of sustained power, meaning that you can theoretically run that server at full load with 8 such PSUs. But you probably don't want eight separate circuits and eight separate breakers and so on.)

A server with a high power usage has knock-on effects. That power will generate a lot of heat in a generally-confined space, so you have to dissipate that heat, and that will require a lot of fans running at high speed (hopefully). Servers with loud fans aren't new but modern high-power servers can take this to a whole new level, where the noise level is not merely obnoxious but actively dangerous. The department now has multiple servers that produce about 120 dB of noise, which is enough to damage people's hearing. As a result, one of the department's machine rooms is now a 'hearing protection absolutely required' area. And beyond the noise, you need the cooling capacity to handle all of that heat.

Not all servers are as demanding as this, of course. These days there's a whole spectrum of possibilities depending on how much the maker stuffed into any particular server. A server might need a single 120V 20A 'circuit' to itself, or have two 120V 15A PSUs that need to be on separate circuits because they draw near the rated 15A. For some of these, you can arrange a PDU that splits a higher power circuit to multiple lower-power ones, assuming of course that you have such a circuit available in the place where you want to put this server (or servers).

(Dedicated machine rooms (and datacenters) will typically have some number of high power circuits available, although maybe not enough. A room pressed into informal service as a place to put some servers may well not.)

All of this constrains where those increasingly demanding servers can go, both now (when they're new) and in the future (when they're used). For example, our main machine room simply can't accommodate such servers, because it has neither the power circuits nor the cooling that would be needed for even one of them. And in the long term, while we and other people will usually accept and run hand me down servers for as long as the hardware lasts, a big used server is not necessarily very useful if its banshee screaming requires dedicated, sound-isolated space that can supply all of that power (with special circuits) and all of that cooling.

(Room level and rack level constraints on how many servers we and other people in the department could cram in aren't new, but we haven't had high-power single servers for a while.)

PS: I'm only looking at servers that are intended to be used without specialized rack infrastructure. Some people are making entire integrated racks, sometimes with rack level DC power and liquid cooling. The systems going into those racks need the entire rack support infrastructure, making it very difficult to (re)use them outside that environment.