Economics of Orbital vs. Terrestrial Data Centers

Original Article ↗ Hacker News Discussion ↗

Use Cases and “Why”

  • Many commenters struggle to see any compelling workload that needs orbital compute.
  • Plausible technical use cases are limited to:
    • Processing data generated in space (imaging, sensing) to reduce downlink volume or latency.
    • Possibly caching static content or model weights in a large LEO constellation.
  • Most other workloads (ML training, generic cloud compute) can be done cheaper and more flexibly on Earth.
  • A recurring suggestion is regulatory arbitrage: escaping local data, environmental, and power regulations, or hosting “unsavory” services. Others note this is illusory because operators, launch states, and spectrum use are still Earth‑jurisdictional.

Economics and Launch Costs

  • Multiple back‑of‑the‑envelope analyses (including the linked one) find orbital power costs several times terrestrial, even under optimistic assumptions; estimates range from ~3–5× to 50–100×.
  • The only way the numbers come close is by assuming:
    • Starship‑level launch costs falling to $100/kg or less, and
    • Very cheap, mass‑manufactured satellites.
  • Commenters highlight: aerospace has a long history of over‑promising cost/kg reductions; even if launch costs drop, the same capital could further cheapen ground nuclear/renewables.
  • Some argue regulatory delay and permitting could make higher orbital $/W tolerable, analogous to people paying AWS premiums, but others note you can already jurisdiction‑shop on Earth much more cheaply.

Engineering Challenges: Cooling, Power, Radiation, Data

  • Strong pushback against the meme that “space cooling is easy”:
    • No convection; only radiative cooling, requiring vast radiator area and complex fluid loops.
    • Radiators, pumps, pipework, and batteries add large mass, making economics worse.
  • Power: 1 GW‑scale orbital systems imply square‑kilometer solar and radiator farms; distributing heat and maintaining them in orbit is non‑trivial.
  • Radiation:
    • Non‑rad‑hard silicon in LEO can work with redundancy and error correction, but at performance and mass cost; higher orbits are much harsher.
    • Shielding adds more mass; long lifetimes for expensive GPUs worsen the problem.
  • Maintenance: high real‑world GPU failure rates + no easy physical access is seen as a show‑stopper.
  • Communications: bandwidth, spectrum licensing, ground stations, and latency are significant, mostly unmodeled costs.

Risk, Law, and Security

  • Claims that an orbital data center is “impossible to raid” are countered by:
    • International space law tying operations to sponsoring states;
    • The relative ease of destroying large, trackable objects in orbit with missiles or killer satellites;
    • Debris and “rods of god”‑like consequences of catastrophic failure.
  • Some see potential national‑security or “tyranny‑proof” motives (harder for mobs or local governments to shut down), but others argue this is politically and militarily unrealistic.

Alternatives and Broader Framing

  • Undersea or polar/remote terrestrial data centers, or solar+storage built adjacent to ground facilities, are repeatedly judged more sensible.
  • Orbital compute for niche, sensor‑adjacent processing already exists (e.g., Jetsons on satellites) and seems sufficient.
  • A minority think long‑term trends (cheaper launch, solar, manufacturing) could eventually make orbital compute “close enough” to be interesting, but most see current hype as PR, investor bait, or sci‑fi fantasy rather than a rational near‑term infrastructure strategy.