Technology

The limiting factor for space data centers is absolutely not the cost to build them. The AI industry has already proven that they will burn the GDP of a small country just to have an LLM that shits out slightly better text that has coin flip level odds of being true.

The limiting factor here is one of thermodynamics. A travel mug for coffee works because it has a near vacuum between the inner lining and the outer shell. Vacuums are fantastic insulation because there’s no atoms there to transfer heat away. Very useful if you want hot coffee for a few hours. Space is a big vacuum, and data centers are giant heat generators. You’re basically putting a computer in a perfect insulation medium. It’s really stupid.

Completely different level of difficulty.

And be pedantic self-driving cars are not a solved problem. If you try driving a Tesla in autopilot and don't intervene when it makes mistakes you'll be in a wall within about 2 minutes.

Building data centres in space isn't just a technical problem, although it's a major technical problem, it's also an economic problem. Obviously yes it's physically possible to do it but there's no way of building a data centre in space for anything close to the price of just doing it on earth and with no other obvious advantage to it being in space there's no reason to do it.

It's the same reason we don't build a transatlantic railroad tunnel. Obviously it's technically possible it's just a very long tunnel, but it would be hella expensive.

Self landing rockets were practically impossible not that long ago

Technically, they still are.

Self driving cars were practically impossible not that long ago.

There have been implementations of a sell-driving vehicle since the 1980s, and we're still far away from "true" full self diving.

Both of these examples demonstrate the adage of "the first 90% of the work takes 10% of the effort, and the last 10% work takes 90% of the effort".

The main factor against it is the $/kg of payload into space.

Oh my lord no. Although, technically yes but not for the reason you think.

The number one issue is heat dissipation. To radiate the heat from one DC satellite (at the power levels needed to run AI workloads) would need a football sized dissipation array. Even if Space X can invent some magical new physics and cut that down to a quarter of that size (hint: they can't), we're still talking about an order of magnitude increase in payload per satellite.

Next on the list is volume. We're currently at around 14k man-made objects in low earth orbit. As it is, satellites (including the ISS) have to perform collision avoidance maneuvers every so often. The calculated limit of satellites we can put up to low earth orbit before orbital collision maneuvers start to become unmanageable is 100k. Basically after that amount we enter into a state where several corrections for each satellite are made regularly, and a single collision at the 100k limit would result in a cascading series of collisions that will render low earth orbit impossible to use. Basically after that anything you put up will get shredded by the insane amounts of debris.

Space X wants to put up a MILLION massive satellites that will require extremely large structures to dissipate the heat from the very power hungry AI chips.

They fully know the impossibility, and when challenged about the over crowding issue during an interview, an engineer brushed it off as "it's not a problem". People who speak that way about science and engineering issues are not serious people.

That’s one of the many issues SpaceX is working to solve, and there’s nothing to suggest they won’t get there.

There are countless engineering and physics reasons why they won't. Stop sniffing Elon's farts. They're not good for your brain.

Edit: all of this is to say: space datacentres are the dumbest idea yet to come out of that idiots face hole. And he's said a lot of really really dumb things.

Other people have already explained the topic in sufficient detail, so I'll just leave a quote from a former NASA engineer and a link to their article.

Taking the NVIDIA H200 as a reference, the per-GPU-device power requirements are on the order of 0.7kW per chip. These won't work on their own, and power conversion isn't 100% efficient, so in practice 1kW per GPU might be a better baseline. A huge, ISS-sized, array could therefore power roughly 200 GPUs. This sounds like a lot, but lets keep some perspective: OpenAI's upcoming Norway datacenter is intending to house 100,000 GPUs, probably each more power hungry than the H200. To equal this capacity, you'd need to launch 500 ISS-sized satellites. In contrast, a single server rack (as sold by NVIDIA preconfigured) will house 72 GPUs, so each monster satellite is only equivalent to roughly three racks.

Source: Datacenters in space are a terrible, horrible, no good idea.