> It's a lot less dumb if you can drastically reduce the launch costs AND drastically increase the launch mass and size. If the Starship thing works out, he will have achieved that.
No, it's still dumb.
No matter how cheap they manage to make SpaceX launches realistically, there's really no situation that a space datacenter makes any sense compared to putting datacenters in, for example, Antarctica. If they built in Antarctica, it would still be cheaper than launching into orbit. You'd have lots of free cold air to potentially cool the computers, and you wouldn't need trained astronauts to fix things when things break. I dont' even think that building a datacenter in Antarctica is a good idea, I'm just saying it's less dumb than launching into space.
Even if you make CPUs that are able to work at a hotter temperature, you still have to contend with the fact that space is effectively one giant insulator and these CPUs cannot work at infinitely high temperatures no matter what.
Even for something like Akamai space data centers are a dumb idea. Keep in mind, this would be space, where people can't easily get to, so you'd need considerably more physical servers to be installed in order to have fault tolerance. Even if the servers weighed nothing, which they wouldn't, you'd need to power them, and in order to power that many servers you'd need solar arrays considerably larger than the ISS.
And what exactly would this buy you? Slightly lower latency for Starlink? With a potentially spotty connection, I'm not even convinced on that; I suspect any latency savings you'd have would be eaten by retries when packages drop.
Outside of a neatness factor, I just don't see exactly would be won by doing this compared to just setting up gigantic solar array in the middle of large deserts and building here on earth. You know, the planet we live on, where technicians can go and repair things in datacenters, because servers break all the fucking time, and these technicians don't have to get into a rocket to do that.