Source Review: Marcy 2026 — The Impact of Computing Data Centres Orbiting Earth

Overall verdict: Consistent

Two-sentence summary: Marcy's 100 GW US AI compute demand by 2035 projection is the most-cited authoritative figure underlying q6.c2's TAI-C SDC scenario. The paper's astronomy-impact analysis is tangential to q6's mass-demand question but reinforces orbital-debris and collision-risk considerations relevant for q6.c14 bandwidth + congestion ceiling.

Key claims

• US data center power demand >100 GW by 2035 → Consistent. Directly underlies q6.c2 TAI-C 1,000 GW scenario by extrapolation (orbital DC captures fraction of full demand). Also corroborates scientificamerican-2026-sdc.

• 5 GW orbital facility = 4×4 km array → Consistent. Implies ~12.5 kg/m² total system mass density at the SDC 40 t/MW range (5 GW × 40 t/MW = 200,000 t in 16 km² → 12.5 kg/m²). Marcy's geometric description is plausible for thin-film panels with integrated radiator + structure mass.

• Orbital-brightness magnitude g = -5 to -7 → Consistent (tangential to q6). The astronomy concern reinforces that aggressive scale-up faces non-launch-cost constraints.

• Visibility window 1.5 h post-sunset / pre-sunrise → Not directly relevant to q6 mass-demand; orientation context.

• Collision-and-debris risk for multiple structures → Consistent with q6.c13's industrial-explosion debris-mitigation implication.

Anti-hallucination check

Quotes appear in extract.md ✓. Single-author preprint; not peer-reviewed yet (submitted to MNRAS Letters). Tier A by venue ambiguity; full submission status as of fetch is "submitted." Future iteration should re-check publication status.

Notes

Marcy's 2026 preprint is one of two tier A sources; the load-bearing contribution is the 100 GW US demand baseline. The astronomy-impact section is tangential to q6 but valuable for q5 / q8 synthesis of non-mass-demand constraints on orbital build-out.

Abstract

Marcy projects that orbital computing infrastructure will be deployed in sun-synchronous low-Earth orbit to address terrestrial AI power constraints. The paper situates U.S. data center demand on a trajectory beyond 100 GW by 2035 and global demand toward 1 TW. A reference 5 GW orbital facility would require a solar array measuring 4×4 km at approximately 500 km altitude. Marcy emphasizes the visibility-and-collision consequences: the resulting structures would appear at magnitude g = -5 to -7 (roughly 100× brighter than the brightest stars), span approximately 0.4° on the sky, and be visible for roughly 1.5 hours after sunset and 1.5 hours before sunrise. The paper raises astronomical, technical, and cultural concerns about dozens of such structures forming visible chains.

Key claims

• us-data-center-power-2035: U.S. data center power demand projected "beyond 100 gigawatt by 2035" with global trajectory toward 1 TW.
• reference-5gw-facility: A 5 GW orbital facility requires solar arrays measuring 4 km × 4 km at approximately 500 km altitude.
• orbital-brightness: Magnitude g = -5 to -7 at the orbital tier; comparable to the Moon's apparent size (~0.4°).
• visibility-window: Visible 1.5 hours after sunset and 1.5 hours before sunrise from any given Earth location.
• collision-and-debris: Multiple such structures pose serious astronomical, technical, and cultural concerns including increased collision risk.

Reviewer notes

This is the most directly load-bearing single document for the q6 demand calc: it explicitly couples U.S. AI compute power demand (100 GW by 2035) to orbital deployment as a candidate sink. The 5 GW per facility scale and 4×4 km array size let us back out ~kg/W requirements when combined with peraspera/luminix mass estimates. Tier A rather than S because it is a preprint with single-author astronomy focus rather than peer-reviewed economic or engineering analysis. The astronomy concerns are tangential to q6's structural-mass-and-propellant question but useful for constraint-mapping in synthesis.