
I don't think that it is harder than putting 10K nodes of Nvidia B300 in space. But who knows, they might manage to make it work. A lot of people also thought that reusable rockets would not be profitable back in the day.
But what I am saying is that they don't need to compete with regular datacenters for the project to make sense. Even if it is more expensive and the performance is worse, there are benefits to having an AI model running in space, and there will be people willing to pay for that.











