It's also a lot less dumb if you can make your chips work well at higher temperatures.
It's also a lot less dumb if you can space harden your chips better than anybody else can at the moment. This is what the Terafab thing is about (for now). Not about pumping out insane amounts of chips but about doing practical R&D for chips that work better in space: hardening and higher temperatures.
Such chips would also be useful for Starlink/Starshield and for Starship itself.
Putting something similar to Akamai/Cloudflare up there would work very well with Starlink. If the costs could be made low enough, of course.
Will it make any sense whatsoever for AI training? Not unless he manages to scale a whole lot of things drastically, and probably not even then. It might make sense for AI inference in a few years, though. Faster inference responses (via Starlink) might be worth some money.
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.
Solar energy is going to expensive in Antarctica.
We can imagine a situation where the server hardware becomes so cheap that the energy cost dominates. In that case, sticking the things in space could make sense, particularly if extremely low mass space PV (just a few microns thick) can be made to work and also work cheaply.
I am harping on this point because You can’t just say “but in future computes won’t need maintenance” because at that point you’re just engaging in speculative fiction that you have no way of knowing. I could say “in the future we’ll have cold fusion” and maybe that’s true but I have no way of knowing.
So given that you would need 2x the power that you’d need on earth. Compared to just putting a shitload of solar panels in a desert where non-astronauts can easily access it I don’t see the point.
And of course there’s the nuclear-power-plant sized elephant in the room; if power is the constraint it would still almost certainly be more economical to have a nuclear reactor than trying to get a data center in space.
Or, and hear me out, Elon is just a drug addict who makes shit up based on his 12-year-old-boy fantasies because it sounds neat and investors just eat it up.
Also...hardened electronics have been a thing for decades. It's not big because shielding is cheaper and far more effective. The only practical use is military, and there are already DoD suppliers who are generations ahead of SpaceX on the hardened chip front.
We probably disagree on the meaning of "smartness"
I guess he's good at making shit up and making his investors forget his failed promises? I guess that requires some level of intelligence.
Things like space datacenters are probably just part of the promotion side. It pumps up share prices for a gullible public. Where's he's been clever is in creating companies that do enough real things to convince many people that he might be able to pull off the unrealistic things as well.