Reusable Thermal-Structural Spines for Orbital Computing: A 16-Architecture Engineering and Lifecycle Cost Study
We compare sixteen orbital and terrestrial computing architectures at 1, 4, and 16 MW over 5-, 7-, and 20-year horizons, for general AI for terrestrial users (Regime C3) and for space-native work (Regimes C1/C2), using a Stefan–Boltzmann radiative model with a coupled junction-to-radiator thermal chain, a reliability and availability model, Starship packaging bounds, and a lifecycle cost model including launch, servicing, insurance, and silicon refresh. The study opens by falsifying its own most attractive screen result: a thin-film "truly light" bus claimed at 11.93 kg/kW rises to 19.44 kg/kW once omitted booms, harness, a realistic 20 K junction-to-radiator drop, view-factor shadowing, and launch strongback are restored (+62.95% hidden mass), and to 20.66 kg/kW as a full system with end-of-life degradation. Three invention rounds then recover the mass honestly: direct 800 V HVDC with a backside structural heat-pipe spine (Arm N, 14.58 kg/kW), a 20-year reusable spine with dry-clamped hot-swappable compute blades (Arm O, 15.60 kg/kW), and a dual-use variant (Arm P). Arm O cuts 20-year cost by 36.7% against the conventional modular orbital incumbent (Arm D) and by 33.4% against the modeled SpaceX AI1 reference (Arm B), and cuts re-entry alumina by 94.7% against Arm D. It still loses to terrestrial clean-grid compute for general AI: $272.65M/MW ($14,960.66/kW-yr) against $220.0M/MW ($12,077.26/kW-yr), a ratio of 1.239x at $500/kg launch — and 1.169x even at $0/kg launch, because the gap is mostly non-launch hardware, servicing, and insurance, not launch. Parity needs launch at or below $200/kg and at least a 45% space-hardware and insurance cost reduction, or launch at or below $65/kg with a 70% reduction. The dual-use Arm P reaches $176.4M/MW net only after an assumed $4,850/kW-yr co-tenancy credit, which is an assumed input, not a market price, and is reported strictly separately. The split verdict is the finding: a reusable spine makes orbital computing substantially cheaper than any conventional orbital design, and still not cheaper than Earth for general AI. The decisive physical test — a thermal-vacuum coupon measurement of the dry-clamp interface — is outstanding and could reject the central claim. Computational, AI-assisted research. All results are modeled outputs of the accompanying simulation code. No orbital prototype has been built, no vacuum thermal test has passed, and no independent environmental advantage has been established. Nothing in this paper is demonstrated hardware performance. Evaluation was performed by an internally separated harness; it has not been externally independently accepted. Accompanying code and data: https://github.com/terryncew/orbital-datacenter-002 (Apache License 2.0); release: https://github.com/terryncew/orbital-datacenter-002/releases/tag/v1.0.0.
Authors
- Terrynce A. White
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24294710
- Primary Topic
- Spacecraft Design and Technology
- Type
- preprint