Will we put data centers in space?

Putting large-scale data centers in Earth orbit is pitched as a way to tap 24/7 solar power, avoid land and permitting constraints, and potentially lower long‑term costs if launch prices plummet. Critics counter that the idea is economically and physically unrealistic for the foreseeable future, citing radiation, cooling by radiation only, maintenance and upgrade impossibility, orbital congestion, and vulnerability to attack or debris — all for workloads whose users and data remain on the ground. Many argue that cheaper, simpler options on Earth (from deserts and offshore platforms to underground sites and nuclear or renewable power) will remain more practical long before orbital facilities could handle a meaningful share of global compute.

Economic viability & launch costs

  • Many argue orbital data centers are uneconomic for the foreseeable future: launch costs, mass of radiators and shielding, and zero-maintenance constraints swamp benefits.
  • Supporters think the article’s conclusion is plausible: if launch costs fall to ~$50–250/kg and space-specific cooling and power are optimized, orbital compute might compete post‑2030.
  • Critics say the analysis “puts a thumb on the scale” by optimistic assumptions for space (huge launch cost drops, aggressive radiator performance, ignoring some mass) and pessimistic ones for Earth (permitting, energy constraints).
  • Several note a single rack’s launch cost is already comparable to its purchase price, but scaling to data-center scale looks prohibitive.

Engineering & physics challenges

  • Repeated concerns: radiation (error rates, hardware lifetime), heat rejection via radiators only, micrometeoroids, and limited power density from solar in orbit.
  • Some point to experiments (e.g., HBM under proton beams, small orbital TPUs) suggesting consumer hardware plus ECC and localized shielding might suffice, but others argue this doesn’t yet scale.
  • Cooling is seen as particularly underexplored at realistic AI-scale power (hundreds of kW to GW).

Alternatives on Earth & oceans

  • Many say problems space DCs ostensibly solve (land, power, NIMBY, cooling) are more cheaply addressed with:
    • Desert / rural DCs, underground facilities, skyscraper DCs.
    • Offshore platforms or underwater containers (though Microsoft’s trial is cited as likely not cost‑effective at scale).
    • Expanded solar/wind plus batteries, or nuclear.

Regulation, politics, and security

  • Some see space as a way around local permitting, NIMBYism, or data protection laws; others counter that jurisdiction follows customers and corporations, not server location.
  • Skeptics highlight vulnerability to anti-satellite weapons and cheap “luddite” attacks, making “security” a weak selling point versus hardened terrestrial bunkers.

Environmental, debris, and tax issues

  • Concerns about rocket emissions, orbital congestion, debris cascades, and eventual decommissioning of hardware.
  • Debate over whether orbital assets should be taxed as benefiting from terrestrial infrastructure.

Workloads and use cases

  • Doubts that high‑end, tightly coupled AI training workloads fit orbital constraints (latency, bandwidth, scale).
  • Niche possibilities mentioned: certain inference tasks, military uses, or politically insulated compute.

Meta‑discussion

  • Thread splits between “obviously stupid / investor scam” and “risky but not physically impossible; depends on future cost curves.”
  • Several emphasize extraordinary claims require strong evidence; others stress experimentation over armchair dismissal.