Starcloud: Orbital Data Centers — Bull Case, Bear Case, and Timeline Reality

Starcloud (formerly Lumen Orbit), a Redmond, Washington startup, is building GPU data centers in low Earth orbit. The company launched the first Nvidia H100 GPU to space in November 2025 and raised a $170M Series A at a $1.1B valuation on March 30, 2026, led by Benchmark and EQT Ventures — becoming the fastest Y Combinator company to reach unicorn status (17 months from batch entry). ## Bull Case **Energy:** Sun-synchronous orbits provide nearly continuous solar exposure. Solar panels in space receive approximately 10x more energy per square meter than ground-based installations (no atmosphere, no weather, no night in the right orbit). Data centers are the largest consumers of grid electricity; moving them off-grid entirely eliminates the energy cost bottleneck. **Cooling:** Space provides an infinite heat sink — no need for the water-intensive cooling towers that consume millions of gallons per day at terrestrial data centers. **Expansion:** No permitting, no land acquisition, no NIMBY opposition, no grid interconnection delays. Scale by launching more satellites. ## Bear Case **Cooling is actually harder in space, not easier.** Without atmosphere, heat can only be radiated (no convection, no conduction to air). Radiative cooling follows the Stefan-Boltzmann law and is extremely slow compared to convective cooling. The International Space Station uses massive radiator panels for this reason. "Infinite heat sink" is misleading — the thermal path to that sink is very narrow. **Radiation damage.** LEO exposes electronics to cosmic rays and solar particle events. Consumer GPUs are not radiation-hardened. Radiation causes bit flips, accelerated degradation, and eventual hardware failure. Radiation-hardened chips exist but are generations behind in performance and orders of magnitude more expensive. **Latency.** LEO altitude adds 20-40ms round-trip latency. Acceptable for batch processing and training workloads; unacceptable for latency-sensitive inference serving. **Maintenance.** Orbital hardware is unrepairable. A failed GPU or storage drive cannot be swapped. Terrestrial data centers achieve high availability through component replacement; orbital data centers must rely on redundancy alone. ## Independent Assessment Google's internal Project Suncatcher analysis concluded that orbital data centers will not be economically competitive with terrestrial facilities until approximately 2035, contingent on significant reductions in launch costs and improvements in space-grade electronics.