Moving data centers to space is a topic that has been very hot this year by Musk. Matt Garman, CEO of Amazon Cloud, gave a very sober judgment: in the next ten to twenty years, space data centers are "extremely unlikely to be economically feasible or practically feasible."
He said this on a podcast on November 10th. The host mentioned that Musk was solving the problem of rocket launch, and Garman responded: This is not something that one person can solve.
He laid out two figures. First, the whole world cannot even launch one launch a day. Second, AWS currently ships a GPU to its data center every 14 seconds. The speeds on both sides are not at the same level at all.
He also mentioned two hard engineering issues. A rack of GPUs can weigh up to a ton, and every ton sent to orbit is worth real money. The other is heat dissipation: there is no air in space, so heat cannot be taken away by fans and air conditioners, but by radiation, which is much less efficient.
From AWS's perspective, these remarks certainly have business considerations. If space computing power comes to fruition, the biggest impact will be on terrestrial cloud manufacturers. On Musk's side, SpaceX just announced in the same week that it will invest US$74 billion to build two computing power parks on the ground.
Therefore, a more realistic view is that space data centers can be used as a long-term research direction, but the computing power gap in the next ten years will still have to be filled by power plants, substations and computer rooms on the ground.