The largest operational orbital compute cluster is currently in low Earth orbit, providing a live demonstration of space-based data processing rather than remaining a theoretical or presentation-only concept.

The system is built and operated by Kepler Communications, a company based in Toronto. It consists of approximately 40 Nvidia Orin edge processors distributed across 10 satellites. These satellites are interconnected via inter-satellite laser links, allowing the hardware to function as a unified compute fabric in orbit. The architecture is designed to process data in situ, reducing the need to transmit all raw data to ground stations.

Kepler reports 18 customers for its in-orbit services. One of the latest, Sophia Space, is using the constellation to evaluate an alternative approach to orbital data center design, specifically focusing on thermal management without conventional active cooling systems.

Sophia Space is developing passively cooled space computers that leverage the vacuum environment and temperature variations in orbit instead of pumps and mechanical chillers. Under its partnership with Kepler, Sophia will upload its proprietary operating system to one of Kepler’s satellites and attempt to deploy and manage it across six GPUs located on two spacecraft. While such distributed software operations are standard in terrestrial data centers, this type of deployment has not previously been executed on an orbital GPU cluster.

For Sophia, validating that its software stack can operate reliably in orbit is a key step in reducing technical risk before launching its own satellites. For Kepler, the activity serves as a demonstration of its broader objective: to provide networking and compute infrastructure that other organizations can use for their hardware in space and in near-space platforms such as high-altitude drones, with potential extension to aircraft.

Industry forecasts indicate that large-scale, cloud-style orbital data centers, as proposed by major aerospace companies, are unlikely to be deployed before the next decade. In the near term, commercial activity is concentrating on orbital edge processing, particularly for high-volume sensor data such as synthetic aperture radar and missile-tracking infrared systems. Government entities, including those in the United States, are already testing the offload of some processing tasks to space-based nodes to reduce latency and alleviate downlink bandwidth constraints.

Kepler’s leadership contends that the emerging use cases favor architectures built around numerous smaller GPUs running continuous inference workloads, rather than a small number of very large chips optimized for periodic training tasks. If this assessment is accurate, Kepler’s laser-linked constellation may serve as an early reference model for the design and scaling of orbital compute infrastructure.