⚛️
Physical Constraints
Computing...
🔥 Stefan-Boltzmann · P = σ × A × ε × T⁴ · 43 tonnes radiateurs/MW · non-negotiable ☢️ Rayonnement LEO 500km · 10–20 krad/an · GPU 4nm: 3900× more sensitive than RAD750 🚀 Gravitation Tsiolkovsky · 840× transport penalty · $0.10/kg ground vs $3,400/kg orbital 🔥 Starcloud-1 confirmed · Real overheating · Continuous H100 impossible without improved radiator 🔥 Stefan-Boltzmann · P = σ × A × ε × T⁴ · 43 tonnes radiateurs/MW · non-negotiable ☢️ Rayonnement LEO 500km · 10–20 krad/an · GPU 4nm: 3900× more sensitive than RAD750 🚀 Gravitation Tsiolkovsky · 840× transport penalty · $0.10/kg ground vs $3,400/kg orbital 🔥 Starcloud-1 confirmed · Real overheating · Continuous H100 impossible without improved radiator
ADS-B NETWORK SAS
Orbital Datacenters — Physics
Revision 4 March 2026
 reads
Loi Stefan-Boltzmann radiateurs thermiques espace satellite
1200×800 · Generated illustration
Physics · Absolute Laws · ADS-B NETWORK SAS

Non-Negotiable Physical Constraints

Three laws of physics govern the viability of orbital datacenters. They do not negotiate with marketing.

43 t
Radiators / MW
Stefan-Boltzmann — 24,000 m² surface area
5%/an
Actual GPU failure
Starcloud-1 (vs 9% estimated) — better, not solved
840×
Transport penalty
$0.10/kg ground vs $3,400/kg Falcon 9
5,6×
Satellite/DC mass ratio
169t orbital vs 30t ground for 1 MW
Law A · Thermodynamics

Stefan-Boltzmann — The Reigning Constraint

In the space vacuum, the only way to dissipate heat is infrared radiation. No convection. No water. No air. Only radiators.

P = σ × A × ε × T⁴
σ = 5.67 × 10⁻⁸ W/m²·K⁴ (Stefan-Boltzmann constant)
A = Radiating surface (m²) → 24,000 m² for 1 MW
ε = Emissivity (0.85–0.95 for black space coating)
T = Radiator temperature (K) → ~350 K for optimal GPU dissipation
🌡️
Minimum Surface Area

To dissipate 1 MW in LEO: 24,000 m² of radiators at 350 K. The ISS uses 51 kg/kW for its radiators. Starcloud-2 aims for 5 kg/kW (×10) — proprietary technology not yet demonstrated in flight.

24,000 m²/MW
⚖️
Radiator Mass

43 tonnes minimum per MW in orbit (ISS data). That is 25% of the total mass of a 1 MW orbital datacenter. Versus 0 tonnes for a ground datacenter using water or air.

43 t min/MW · ISS ref
🔥
Starcloud-1 Confirmed

Novembre 2025 : le H100 de Starcloud-1 n'a pas pu tourner en continu. Real overheating on demonstrator satellite. Starcloud-2 (oct. 2026) testera un radiator ×10 efficiency. Johnston : "tbd 😅" sur la masse exacte.

🔴 Overheating confirmed
🔴
Khayyam Wakil (Systems Architect) — Verified LinkedIn comment

"Radiators, people!! You'll have 100's of thousands of square feet of radiators for these data centers. Just getting those radiator panels up there is four trips lol." — Comment not refuted by Starcloud.

Law B · Particle Physics

Orbital Radiation — 4nm GPU vs LEO Environment

ParameterLEO 500 kmGround
TID (Total Ionizing Dose)10–20 krad/an~0 krad
GPU H100 (4nm) sensitivity×3 900 vs RAD750Reference
GPU lifespan with shielding2–3 ans5–10 ans
GPU failure/yr (2025 estimate)9%<0,5%
GPU failure/yr (Starcloud-1 actual)5%<0,5%
SEU (Single Event Upset)FrequentNegligible
Cumulative GPU degradation over 5 years in LEO (%)
5%/yr rate (Starcloud-1 data) · No replacement
⚠️
Partial good news

Starcloud-1 measures 5% GPU failure/yr, better than the 9% estimated in models. This improves economics (McCalip: $11.1 Bn vs $16.7 Bn at 5%). But Starcloud-1's nominal mission is only 11 months — behavior over 5 years remains unvalidated in flight.

Law C · Gravity

Tsiolkovsky — The Gravitational Penalty

Transport from the ground surface costs 840× more orbital than terrestrial. This penalty applies to every kilogram: satellites, GPUs, radiators, fuel, SSA sensors, optical terminals.

LauncherCustomer price/kg×ground
Ground road transport$0,10×1
Falcon 9 (actuel)$3 400×34 000
Falcon Heavy (actuel)$1 500×15 000
Starship (target)$100–200×1 000–2 000
Starship (SpaceX internal cost)~$20×200
Mass breakdown — 1 MW orbital vs ground datacenter (tonnes)
Source: ADS-B NETWORK SAS based on ISS and Starcloud data