So, you know how a good thermos keeps its contents hot or cold, depending on what you put in it?

You know the trick they use to do that? They use a vacuum.

Heat can be passively transferred using convection, conduction and radiation. Convection requires the movement of fluid, there's no fluid in a vacuum so that can't happen. Conduction requires that one vibrating molecule causes another adjacent molecule to vibrate. Again, not possible in a vacuum because there are no molecules to transfer that vibration. That leaves radiation.

(In a thermos it's not quite that simple. A thermos isn't floating in space, the outside needs to connect to the inside somehow, so even though there's a big vacuum bottle, the heat gets transferred through the parts that aren't in a vacuum.)

Radiation is possible in a vacuum, it's just really inefficient. People used to see radiation transfer in a vacuum all the time: incandescent light bulbs. They're much more rare these days. But, in that case a filament was heated up so much that it glowed white hot. To reduce the heat transfer, that white hot filament was put in a vacuum -- the bulb of the light bulb. If you ever touched one, you know they got hot. But, the filament was 2500 celsius, and the bulb itself was maybe slightly over 100C. That shows how inefficient radiation is. Even with something that's thousands of degrees, it only heated the glass of the bulb from say 20C to 120C.

So, it's hard to dump heat in space. The only thing that works is radiation, and radiation is much less effective when things are not glowing white hot. In addition, anything in space is getting heated up by the sun. So, not only do you need to dump any heat the device produces itself, it also needs to dump any heat it's receiving from the sun.

What are two of the main things that datacenters need? Power and cooling. The need for power is famous because it's driving up electricity costs. The need for cooling is well known because of all the water being used.

To meet the power needs, a datacenter would need massive solar panels to collect as much sunlight as possible. The solar panels on the ISS provide about 30kW of power. A small datacenter (and I mean small), uses 1 MW. So, for one small DC you'd need 30x the number of solar panels as on the ISS. If you're talking a conventional pre-AI datacenter, it's more like 100MW, so you'd need 3000x the number of solar panels as on the ISS.

Solar panels create electricity from sunlight, but by design they have to be facing the sun, which means they get heated up too. So, before you even turn on a computer in the DC, you have to find a way to dump all that heat that the solar panels are creating. Then, let's say you want to actually turn on the computers in the DC. On earth can be cooled by dumping water on something hot and letting water evaporate. That's really easy to do, and really cheap. You just move the heat from the computers into the water and it goes into the sky. But, in space, there's no sky to evaporate into, and no source of water. So, you need radiative cooling. Now, just imagine how much radiative cooling you'd need to match the performance of a simple "dump water on it" cooling system.

tl;dr: The idea of datacenters in space is ludicrous. Even someone as stupid as Musk knows that. But, investors are dumber than Musk -- or at least investors think that there are other investors that dumb, so they won't get stuck with the shares if they need to sell them. Musk needs the belief that datacenters in space to be possible or his shares all crash in value.