Source Review: Luminix — Data Centers in Space 2029-2031 (2026)

Verdict: Mixed Confidence: medium

Luminix's 42 t/MW per-MW orbital DC mass figure is the third independent engineering analysis converging on the 40 t/MW central estimate; this is load-bearing for q6.c1. However, the report's viability-probability estimates (<15% 2029-31, 30-40% 2032-35) are not first-principles-derived and McCalip's methodological critique applies. The $200/kg launch threshold cited matches NASA OTPS and the Suncatcher-Google reference. The 50-100 GW "addressable AI compute demand" figure is also industry-narrative-driven rather than derived. Net: use mass-density figure with high confidence; treat viability probabilities as scenario inputs not derivations.

Abstract

Luminix industry analysis provides the most detailed per-MW orbital data center economic breakdown in the public 2026 literature. For a 1 MW orbital cluster: ~42,000 kg total equipment mass (20 t compute

• 8 t solar + 8 t radiators + 6 t structure), $255-505M capex (hardware-dominated; launch only $12.6M at $200/kg target). A 100 MW constellation requires ~50+ dedicated Starship launches. Market-capture thesis: $10-50B annual market by 2030 from "overflow training capacity for power-starved hyperscalers" — i.e., the addressable subset of 50-100 GW global AI compute demand that cannot connect to terrestrial grids before 2030. Probability assessment: <15% economic viability by 2029-2031, rising to 30-40% by 2032-2035 conditional on Starship $200/kg + COTS hardware radiation survival + persistent terrestrial power bottleneck.

Key claims

• mass-per-mw-luminix: ~42 t/MW for 1 MW orbital cluster (20 t compute + 8 t solar + 8 t radiators + 6 t structure).
• capex-per-mw: $255-505M capex per 1 MW cluster ($12.6M of which is launch at $200/kg).
• launches-per-100mw: ~50+ Starship launches for 100 MW constellation.
• market-size-2030: $10-50B annual addressable market by 2030.
• addressable-demand-50-100gw: 50-100 GW of AI compute demand "cannot physically connect to terrestrial grids before 2030."
• starship-200kg-threshold: $200/kg LEO is the "viability threshold" Luminix references.
• 2029-2031-viability-15pct: <15% probability of economic viability by 2029-2031; 30-40% by 2032-2035 conditional on three factors.
• speed-to-compute-arbitrage: A 1 MW cluster generates ~$70M/yr at cloud rates; 5-year grid queue = $350M foregone, justifies 2-3× capital premium for speed.

Reviewer notes

Tier C: industry report with quantitative breakdown. The 42 t/MW figure is corroborated by peraspera's 30-50 t/MW range and is the most-citable single number for orbital DC mass demand. The market size and viability probabilities are scenario assumptions; q6's calc should treat them as one estimate among several and propagate the uncertainty. Tier C scalar verdict applies; the speed-to-compute arbitrage thesis is a notable conceptual contribution.