Orbital Data Centers Unlock GW-Scale AI Training
Shift AI training to space for 22x cheaper energy ($0.002/kWh via 95% capacity factor solar), radiative cooling, indefinite GW scalability, and rapid deployment without Earth permitting delays.
Harness Space Solar and Cooling for Massive Cost Savings
Terrestrial data centers face tripling electricity demand, silicon shortages, and transformer bottlenecks within 1-2 years, limiting clusters to 100MW-1GW while multi-GW setups are needed for AGI-scale LLMs like Llama 5 or GPT-6 by 2027. Orbital data centers solve this by tapping uninterrupted solar power with >95% capacity factor (vs. 24% median US terrestrial solar), 40% higher irradiance without atmospheric losses, yielding 5x more energy per array. At $0.03/W solar cells and $5M launches for 40MW modules (amortized over 10 years), energy costs drop to $0.002/kWh—22x below US wholesale $0.045/kWh.
Cooling leverages deep space's -270°C as a heatsink via deployable radiators (half the solar array size), achieving 838W/m² radiation at 20°C (net 633W/m² after sun/Earth absorption). This eliminates water use (1.7M tons/40MW cluster on Earth), chillers ($7M/10yrs), and backup power ($20M), yielding $8.2M total 10-year cost for 40MW vs. $167M terrestrial. PUE matches hyperscalers without overprovisioning for 45°C peaks, using two-phase loops, direct-to-chip, or immersion cooling in pressurized modules.
A 5GW cluster (4km x 4km silicon array at 22% efficiency) costs less than equivalent Earth solar farms, with <0.15%/year degradation via radiation-hardened thin-film cells (>1000W/kg, foldable for launch).
Scale Indefinitely Without Earth Constraints
Earth's permitting delays (10+ years for GW projects) and physical limits block multi-GW clusters exceeding largest US power plants. Space enables linear modular scaling: dockable 40MW containers in 3D architectures for low-latency AI training (containers within 200m, daisy-chain networking with spine switches for bisection bandwidth). Speed-of-light 35% faster in vacuum vs. fiber aids tight coupling.
Deployment skips grid/transmission hurdles (e.g., xAI's gas generators in Memphis); launch reusables hit $5M/100t SSO ($30/kg, potentially $10/kg). Single universal ports minimize failure points; resiliency ensures graceful degradation. Data shuttles (e.g., Snowcone-scale, petabyte/exabyte hauls) or laser links to Starlink/Kuiper/Kepler handle I/O, bypassing RF spectrum limits.
Orbit choice (SSO) minimizes debris via maneuverability, tracking, underused paths; solar arrays self-heal debris hits per ISS data. No astronomy impact at dawn/dusk visibility; EU ASCEND study confirms lower GHG emissions, zero water use.
Proven Design Principles Ensure Feasibility
Starcloud's concepts follow modularity (independent dock/undock), maintainability (10+ year life, easy swaps), minimal moving parts (one power/network/cooling port), resiliency (no single failures), and incremental scaling (profitable from container 1). HVDC transfers power; ADCS controls huge deployable arrays (Z-fold/roll-out). Radiation shielding ($1.2M/40MW at 1kg/kW, $30/kg) and UV/thermal mitigation enable GW viability. Heat pumps boost radiator output via T^4 law if needed, but passive designs suffice.