52 min
Max downtime/yr SLA 99.99%
Enterprise standard — impossible in orbit
6–18 mois
Orbital RTO
Satellite replacement delay vs 4h ground
🔴 Bloquant
CLOUD Act EU
Starcloud = US company → US jurisdiction
5–10×
Reduced TAM
Real addressable market vs announced general cloud
Enterprise SLA vs Orbital Constraints
Fundamental Incompatibility
Enterprise contracts demand guarantees that orbital physics makes impossible to fulfill with current technology.
| SLA Component | Ground Standard | Orbital Reality | Verdict |
|---|---|---|---|
| Compute uptime | 99,99% (52 min down/an) | Eclipses + 5%/yr irreplaceable GPU failures + shadow zone passes | 🔴 Incompatible |
| Guaranteed latency | <1 ms intra-DC / <50 ms cloud | LEO round-trip: 5–40 ms + limited laser bandwidth | 🟠 Partial inference only |
| Corrective maintenance | Technicien sur site en 4h | Impossible — zero hardware replacement until Starship operational | 🔴 Incompatible |
| Geographic redundancy | Multi-AZ / multi-region | 88,000 sat constellation not yet deployed — vision not yet reality | 🟠 At scale only |
| RTO (Recovery Time Objective) | <15 min enterprise standard | Satellite replacement: 6–18 months (fab + launch) | 🔴 Incompatible |
| RPO (Recovery Point Objective) | Near-zero (sync replication) | Limited uplink/downlink bandwidth | 🔴 Problematic |
| GDPR / HDS (EU) compliance | ISO 27001, SOC 2, GDPR, HDS | US CLOUD Act — Starcloud = US company → sovereignty not guaranteed | 🔴 Blocking EU legal |
| Contractual penalties | SLA credits 10–30% invoice | Impossible to honor without possible physical intervention | 🔴 Incompatible |
Direct impact on the business case
Enterprise cloud (Fortune 500, Finance, Healthcare, EU Defense) represents ~70% of AWS/Azure/GCP revenues. Without 99.99% SLA, an orbital datacenter can only access the spot/batch market, 5 to 10× less lucrative than Starcloud's projections.
Use Cases · Real TAM
What Is Feasible vs What Is Not
| Use Case | Required SLA | Orbital feasible? | Horizon |
|---|---|---|---|
| SAR inference (Capella Space) | Best-effort | ✅ Demonstrated Starcloud-1 | 2026 |
| US sovereign satellite edge compute | Best-effort | ✅ In-Q-Tel probable client | 2026–2027 |
| Non-real-time batch training | Flexible SLA | ⚠️ Partial (NanoGPT demonstrated) | 2027–2029 |
| General enterprise cloud | 99.99% + GDPR | 🔴 Incompatible SLA + EU legal | 2035+ si jamais |
| High-frequency trading | <1ms, 99.999% | 🔴 Physics (LEO light) prohibits it | Jamais |
| Healthcare / HDS / EU data | GDPR + HDS + sovereignty | 🔴 CLOUD Act blocking | Never without US legal reform |
| Cold backup / LEO archiving | Long RPO acceptable | ⚠️ Niche possible (Lonestar) | 2027–2030 |