🔗
Links & SMR
Loading...
🛰️ ADS-B NETWORK SAS · Independent Analysis · Verified sources · March 2026 📊 Baseline ratio 1 GW · 3:1 orbital/ground · McCalip / IEEE Spectrum 🛰️ ADS-B NETWORK SAS · Independent Analysis · Verified sources · March 2026 📊 Baseline ratio 1 GW · 3:1 orbital/ground · McCalip / IEEE Spectrum
ADS-B NETWORK SAS
Datacenters Orbitaux — Links & SMR
Revision 4 March 2026
 reads
Optical Links & SMR — Two Underestimated Problems
1200×800 · Generated illustration
Links & SMR · ADS-B NETWORK SAS · March 2026

Optical Links & SMR — Two Underestimated Problems

The RAND Corp error on optical pointing partially invalidates Johnston's $5/W. Nuclear SMRs remain out of reach for the decade.

20–50%
Optical unavailability
Cloud attenuation by latitude
100 Gbps
Max ISL throughput
Starlink Gen2 laser — but mechanical pointing
15–20 ans
Space SMR horizon
NASA Kilopower: 1–10 kW only today
🔴 Erreur
Economist/Johnston
Optical pointing ≠ phased array — RAND Corp
Major Technical Error · RAND Corporation

The Economist's False Claim on Pointing

🔴
Bonnie Triezenberg, Policy Analyst & Sr. Engineer, RAND Corporation

"The truth is OPPOSITE. Phased array beams are digitally formed and steered — all you need is knowledge of satellite attitude, not control. Optical crosslinks are MECHANICALLY pointed and their large inertias make attitude control DIFFICULT and ESSENTIAL. Given that whopper, why would any sane satellite engineer believe anything else in this article?" — Commentaire LinkedIn public, March 2026.

📡
Phased-Array Antennas (Starlink)

DIGITAL pointing. No moving parts. The beam is formed electronically by phase shifting. Only knowledge of satellite attitude is required, not precise attitude control. Conclusion: no costly pointing mechanism.

Digital pointing
💡
Optical Laser Links (Starcloud)

MECHANICAL pointing. Fine actuators. The laser beam must be physically directed toward the moving neighboring satellite. Large mechanism inertia = precise and essential attitude control. Conclusion: pointing cost comparable to or greater than phased-array.

🔴 Mechanical pointing

Direct consequence: the mass economy assumed by Starcloud (without phased-array antennas = cheaper satellites = $5/W) is partially negated by the mechanical pointing systems required for laser links. Johnston's $5/W is likely underestimated. Glenn Zorpette (IEEE Spectrum): the Economist article relies on a single non-neutral source and no independent validation.

Ground-Space Links · Practical Problem

Atmospheric Attenuation — The Ignored Factor

☁️
Clouds = Outage

Ground-space optical links are blocked by clouds. Unavailability rate: 20–50% depending on latitude and season. Paris: ~60% annual cloud cover.

20–50% unavailability
🔀
Solutions = Costs ×N

Redundant ground station network in clear-sky areas (deserts). Cost: ×3 to ×5 the number of stations. Or RF backup (×100 reduced throughput). KSAT solved this with its 40 sites — but at the cost of a global network.

×3–5 stations requises
🛰️
KSAT Hyper = Proof

KSAT is launching its own LEO constellation "Hyper" precisely to solve the periods without ground station coverage. This is proof that the problem is real and unsolved.

🟡 Problem acknowledged by KSAT
Nuclear SMRs · Complete Analysis

Space Nuclear Reactors — Good Idea, Wrong Timing

ObstacleDetailAssessment
1967 Outer Space Treaty + 1992 UN PrinciplesArt. IV prohibits WMDs in orbit. Civil reactors are in a legal grey zone — each launch requires international negotiation🔴 Major political risk
Radioactive debrisCosmos 954 (1978): nuclear contamination of Canada. A failed orbital reactor = permanent radioactive waste in LEO🔴 Unacceptable without guaranteed deorbiting
Self-irradiation shieldingThe reactor irradiates neighboring GPUs. Mandatory shielding = additional mass = counterproductive for mass economy🔴 Physical paradox
Current power (NASA Kilopower)1–10 kW per unit. A datacenter needs MEGAWATTS. 1 MW = 100–1,000 Kilopower units🔴 Insufficient scale
Orbital reactor maintenanceNuclear reactor not maintainable in orbit. Lifespan 5–15 years. Same problem as GPUs🔴 Maintenance impossible
Space SMR costeVinci (Westinghouse ground): $50–200M/unit. Space = ×10 minimum. For 1 MW: $5–20 Bn🔴 Prohibitive cost
💡
What is realistic long-term (2035–2040)

Space nuclear reactors of 1–10 kW to power crewed space stations or lunar missions — yes. For datacenters requiring MW in object-populated LEO — no, with current legal and technical constraints. The realistic horizon for a datacenter-scale space SMR is 2035–2045.