Despite the excitement surrounding space-based data centers, the number of GPUs currently in orbit is limited. However, as this begins to shift, the emerging business model for orbital computing is becoming clearer.
The most extensive compute cluster in orbit was deployed by Kepler Communications from Canada in January. It features approximately 40 Nvidia Orin edge processors housed within 10 active satellites, all interconnected through laser communication links.
Kepler now serves 18 customers and recently added Sophia Space as its newest client. Sophia Space, a startup, will test its unique orbital computer software within Kepler’s satellite network.
Industry experts predict that large-scale data centers, like those proposed by SpaceX and Blue Origin, won’t materialize until the 2030s. Initially, the focus will be on processing data collected in space to enhance the capabilities of sensors used by both private companies and government agencies.
Kepler does not consider itself a data center company. Instead, as CEO Mina Mitry explained to JS, it views itself as providing infrastructure for space applications. Kepler aims to offer network services for other satellites, as well as drones and aircraft below.
Sophia Space is working on passively-cooled space computers, addressing a significant challenge for large-scale orbital data centers: preventing processors from overheating without relying on heavy, costly active-cooling systems.
Through their new partnership, Sophia plans to upload its proprietary operating system to a Kepler satellite and test its deployment across six GPUs on two spacecraft. While commonplace in terrestrial data centers, this will be a first in orbit. Successfully running the software in space will be a critical validation step for Sophia ahead of its planned satellite launch in late 2027.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
For Kepler, this partnership demonstrates the effectiveness of its network. Currently, it processes data sent from the ground or collected by payloads on its satellites. As the industry evolves, Kepler anticipates connecting with third-party satellites to offer networking and processing services.
Mitry notes that satellite companies are now designing future assets around this model, highlighting the advantages of offloading processing for sensors that require more power, such as synthetic aperture radar. The U.S. military is a significant client for this type of work, developing a missile defense system that relies on satellites to detect and track threats. Kepler has already showcased a space-to-air laser link in a demonstration for the U.S. government.
Edge processing, which involves handling data where it is collected for quicker responsiveness, is where orbital data centers will initially prove their worth. This vision distinguishes Sophia and Kepler from established space companies like SpaceX and Blue Origin, as well as startups like Starcloud and Aetherflux, which are investing heavily in large-scale data centers with traditional processors.
“We believe that inference is more important than training, which is why we prefer distributed GPUs for inference rather than a single powerful GPU for training,” Mitry told JS. “If it uses kilowatts of power and runs only 10% of the time, it’s inefficient. In our setup, our GPUs operate 100% of the time.”
Once these technologies prove successful in space, the possibilities are limitless. Sophia CEO Rob DeMillo notes that Wisconsin recently banned data center construction, a move some Congress members also support. Such restrictions make the prospect of space-based alternatives more appealing to them.
“There are no more data centers being built in this country,” DeMillo remarked. “Things are going to get interesting from here.”
