Source Review: Andrew Cote — Do Orbital Data Centers Make Sense? (2026)

Verdict: Mixed Confidence: medium-high

Cote's bandwidth-binding-constraint argument is one of the strongest skeptical positions in the 2026 literature. The 500-800 Tbps total orbital comms vs 5-20 Tbps per ground DC is a real constraint, but Cote's "niche-only" conclusion assumes ground-station-routing rather than OISL-interconnected compute clusters; this assumption is contestable. The 10-20× space-qualified-chip performance lag is the strongest single point — if true, orbital DC viability depends on radiation-tolerant compute architecture or accepting the performance penalty. Cote's piece is load-bearing for q6.c14 and correctly tempers TAI-C optimism by factor-of-5 sensitivity.

Abstract

Andrew Cote's technical analysis of orbital data center economics identifies bandwidth as the binding constraint rather than mass or cost. For a 1 GW orbital DC, radiative cooling at 300 K requires ~2.6 km² radiator area, manageable within Starcloud's proposed 16 km² envelope. Critical vulnerability: space-qualified chips lag commercial H100 performance by 10-20×, neutralizing the cheap-orbital-energy advantage. Global orbital comms capacity is "500-800 Tbps total" while single ground DCs achieve 5-20 Tbps each — implying orbital deployment is viable only for niche workloads served via in-space laser cross-links, not for hyperscaler training workloads. Launch costs at $1,000/kg currently, halving every few years per Cote.

Key claims

• radiator-area-1gw: 1 GW system needs ~2.6 km² radiator at 300 K.
• starcloud-envelope: Starcloud's proposed deployment envelope is 16 km².
• space-qualified-perf-gap: Space-qualified chips "lag H100 perf by 10-20×," erasing orbital energy economics.
• global-orbital-bandwidth-cap: 500-800 Tbps total orbital comms capacity vs 5-20 Tbps per single ground DC.
• niche-only-conclusion: Orbital DC viable only for specific niches (in-space processing via laser crosslinks), not hyperscaler replacement.
• launch-cost-current: $1,000/kg currently with reusable rockets; halving every few years.

Reviewer notes

Tier C: credentialed-expert blog. Cote's bandwidth-constraint argument is one of the most technically grounded skeptical positions in the 2026 literature. The 10-20× space-qualified-chip performance lag is load-bearing — if true, the orbital DC market is fundamentally niche-only at any launch cost. q6's source-review should explicitly weigh Cote's bandwidth argument against the Handmer / Luminix optimistic forecasts.