Orbital Compute Files FCC Application for 100,000-Satellite, 10GW Space Data Center Constellation

Orbital's FCC application, filed June 24, requests Ka-band spectrum solely for telemetry backup, with the constellation designed to relay data via optical inter-satellite links through third-party networks such as SpaceX's Starlink and Amazon's Project Kuiper. No daily ground communications are planned ^[1].

The application includes a nine-page orbital debris assessment report committing to satellite disposal within five years of end-of-life, against a planned seven-year operational lifespan per satellite. The satellites would carry Nvidia Blackwell GPUs for the initial demonstrator mission, with later generations planned to use Nvidia's Space-1 Vera Rubin modules ^[1].

Orbital's core pitch is that space eliminates three constraints choking terrestrial data center expansion: power availability, grid interconnection timelines, and cooling water scarcity. Sun-synchronous orbits provide continuous solar exposure, and the vacuum of space serves as an infinite heat sink ^[1][4].

The company is focused specifically on AI inference — the phase where trained models process user queries — rather than training. Inference workloads are stateless and can be distributed across separate nodes, making them more suitable for a constellation architecture than the tightly coupled GPU clusters required for model training. Orbital claims 20–40 milliseconds of round-trip latency at 500–600km altitude, comparable to a fiber connection between Los Angeles and Denver ^[4].

However, the passive cooling claim comes with caveats. Each satellite requires approximately 100-meter radiators to dissipate heat, a significant engineering and deployment challenge at the scale of 100,000 units ^[1].

Orbital was founded by Euwyn Poon, a Cornell-educated engineer and lawyer who previously founded Spin, the electric scooter company that was acquired by Ford. At Spin, Poon oversaw the manufacturing of approximately 250,000 scooters ^[3][6].

The company emerged from a16z's Speedrun accelerator in May 2026 with $5 million in pre-seed funding. Orbital is building Factory-1, a dedicated R&D and manufacturing facility in Los Angeles, to produce its specialized compute satellites ^[3][4].

Orbital is not alone in pursuing space-based compute. SpaceX filed its own FCC application on January 30, 2026, for an 'Orbital Data Center System' comprising up to one million satellites between 500 and 2,000 kilometers altitude, projecting 100GW of AI compute capacity. Blue Origin is also reportedly developing competing orbital compute strategies ^[5][2].

The concept faces vocal skepticism. OpenAI CEO Sam Altman has publicly questioned the feasibility of orbital data centers. The engineering challenges are substantial: radiation hardening for GPUs, launch economics at scale, satellite servicing, and the sheer mass involved — Orbital's full constellation alone would weigh roughly 200,000 metric tons ^[1][2].

Orbital's immediate milestone is the launch of a single-GPU test payload on a SpaceX Falcon 9 in April 2027. That demonstrator mission will validate sustained GPU operation in orbit, test radiation tolerance on Nvidia Blackwell silicon, measure thermal performance, and prove out data downlink via optical inter-satellite links ^[1][4].

The first dedicated compute satellite, Orbital-1, is currently slated for 2028. The FCC filing represents a regulatory starting point; spectrum coordination, launch licensing, and orbital slot allocation all remain ahead. The timeline from a $5 million pre-seed to a 100,000-satellite constellation is, by any measure, extraordinarily long ^[1][2].