Just read about plans to build data centers in space.

My initial take was: But why?

Will we run out of space? Probably not.

To minimize impact on the environment? That is the core rationale of this architecture and the paper. I agree with that. But practicality is equally important. Ideas need to be executable. https://lnkd.in/gf9-CSrU

But first, you need to design a different kind of data center, optimized for this architecture.

Why?

One key aspect that has not been touched upon in this article is breakdowns and troubleshooting.

On Earth, a minor issue like a failed fan, a memory module, or a firmware patch gone wrong is trivial. A technician can swap parts or reboot in minutes.

In orbit, physical intervention is impossible unless you dispatch a crewed or robotic servicing mission, which is orders of magnitude more expensive and slower.

Even diagnostics can be hard if the fault affects communications or power systems, because you lose telemetry before you can analyze the cause.

Not being an expert in data center design, I asked AI about possible redundancy design ideas. Here are the suggestions.

START——–

” Triple-modular redundancy (TMR) in compute nodes — i.e., three copies of every critical circuit voting on outputs to tolerate bit flips from cosmic radiation.

Radiation-hardened chips (e.g., AMD or ARM designs rated for LEO radiation).


Autonomous fault detection & reconfiguration, using onboard AI that can reroute workloads to healthy nodes and even down-clock faulty sections to isolate damage.

Redundant power and comm links (multiple solar arrays, RF links) to prevent single-point failure.

⚙️ What “troubleshooting” would look like in practice

Remote telemetry: Continuous health monitoring of each subsystem (thermal sensors, voltage lines, memory parity checks). Engineers can send diagnostic scripts from Earth, much like you’d SSH into a remote server — but through a high-latency satellite link.

Automated self-healing software: Machine-learning-based fault-management systems that can isolate faulty modules, reallocate workloads, or even perform virtual “reboots” at the node level.

Robotic servicing: Companies like Astroscale and Northrop Grumman are already developing spacecraft that can dock, refuel, or swap modules — a capability that could eventually extend to orbital data centres. But today it’s experimental and high-cost. “

———–END OF RESPONSE

Will the juice be worth the squeeze?

There are a plethora of aspects that will still need human trips to address….unless….there are stations built in space to host such crews.

Maybe a few decades from now. At this point, the best bet is to focus on two broader areas. One is to optimize AI algorithms and methodologies to become more efficient. Second is to design data centers that are not only efficient in terms of energy utilization but also space utilization (like vertical scaling architecture).