Orbital aims for space AI data centers

Orbital, a startup building AI data centers in space, has set a 2027 target for its first test mission. The company aims to bypass terrestrial power and cooling limitations by operating in low Earth orbit, a move backed by funding from a16z Speedrun.

The demand for AI compute is outstripping available power, a bottleneck Orbital believes can only be solved by generating energy in orbit. Its strategy focuses on using solar power and radiative cooling to overcome the energy constraints that limit current AI infrastructure.

Factory-1 Opens in Los Angeles

Orbital is opening Factory-1, its research and development facility in Los Angeles, signaling a commitment to in-house innovation.

Orbital-1 Mission Details

The company's first satellite, Orbital-1, is slated for a SpaceX Falcon 9 launch in April 2027. This mission will validate sustained GPU operation in space, test radiation hardening, and demonstrate commercial AI inference workloads. Orbital is also seeking FCC approval for a constellation of satellites.

Addressing AI's Energy Ceiling

Orbital CEO Euwyn Poon highlights that AI progress is constrained by the power grid. He states that in orbit, continuous solar power and distinct cooling methods offer a scalable solution.

The company's architecture is tailored for AI inference, which is more distributed and suitable for a constellation of satellites than the tightly coupled GPUs required for large model training.

Orbital's approach could redefine the scalability of AI infrastructure, moving beyond Earth's energy limitations. This venture into Orbital AI data centers is part of a growing trend exploring space for compute resources, as detailed in AI's Next Frontier: Datacenters in Space and the Race for Supremacy. The broader implications for energy and the new space race are explored in Powering the Cosmos: AI, Nuclear, and the New Space Race.