We had a similar thread in November when Google announced a moonshot project to explore datacenters in orbit:
Here is the announcement:
Which also has a link to their whitepaper that outlines the challenges (radiation, heat, cost, networking, etc.). It includes cost projections as well.
Google’s strategy here is to use a single project or goal to further development in multiple areas. Even if the goal ends up failing, the hope is their is collateral innovation.
In any case, datacenters in inaccessible places like space or the ocean will be running commodity HW for common use-cases. It could be tasked with batch processing corporate data overnight, for example. The cutting edge stuff would stay local-ish.