Space Data Centers Feasibility Overestimated
VraserX · x · 2026-07-12
The author argues that while space data centers aren't impossible, Elon's timeline is overly optimistic.
Citing a Google analysis, the text notes that for orbital computing to be competitive, launch prices must drop below $200/kg. Current costs remain at "thousands of dollars/kg," and Google predicts this threshold might only be approached by the mid-2030s.
The author lists major engineering challenges:
- Space is a vacuum, making direct air cooling impossible; GPU heat must be dissipated via massive radiators.
- Other hurdles include radiation, shielding, massive solar arrays, bandwidth requirements, and hardware failures.
- Broken GPUs in ground data centers can be replaced manually, but in-orbit hardware relies on resupply missions or autonomous repair systems that don't currently exist at scale.
- Meanwhile, AI hardware iterates rapidly, adding continuous commercial pressure.
Conclusion: Launching a functional space prototype next year is plausible, but it only proves it "can run," not that it's "economically viable." The author estimates that commercial scaling won't happen until the mid-to-late 2030s at the earliest, and the concept currently feels more like investor fantasy.
Related event: Space Data Centers Face Launch Cost Bottleneck(2 posts)→
More from Infra
- DeepSpeed shows AutoEP/AutoTP/AutoSP parallelism gains, hosts office hours — PyTorch · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- Gary Marcus: With Q3 data added, Anthropic's compute-efficiency growth shows deceleration — GaryMarcus · 2026-10-27
- SpaceX locks in $40B GPU financing deal with Apollo, the largest ever committed — TechCrunch AI · 2026-10-27
- Inference is just Kubernetes with a multi-trillion-dollar market, and the communities overlap — matt_slotnick · 2026-10-27
- vLLM eyes Rust migration as inference engines race to rewrite the stack — AccBalanced · 2026-10-27