While the concept of data centers in space has garnered significant attention, the actual deployment of Graphics Processing Units (GPUs) in orbit has been limited. This landscape is beginning to shift, with the near-term business of orbital compute taking distinct shape. Leading this evolution is Canada's Kepler Communications, which in January launched what is currently the largest compute cluster in orbit. This cluster features approximately 40 Nvidia Orin edge processors distributed across 10 operational satellites, all interconnected by advanced laser communications links. Kepler currently serves 18 customers and recently announced a new partnership with Sophia Space.

Kepler Communications does not define itself as a data center company but rather as an essential infrastructure provider for various space applications. Its vision is to serve as a networking layer, offering services for other satellites in space, as well as for drones and aircraft operating within the Earth's atmosphere. The company's current operations involve carrying and processing data uploaded from the ground or collected by hosted payloads on its own spacecraft. As the orbital compute sector matures, Kepler anticipates expanding its services by linking up with third-party satellites to provide extensive networking and processing capabilities. CEO Mina Mitry notes that satellite companies are now designing future assets around this model, recognizing the benefits of offloading processing for power-intensive sensors, such as synthetic aperture radar. The U.S. military, a key customer, is particularly interested in this capability for developing new missile defense systems that rely on satellites for threat detection and tracking; Kepler has already demonstrated a space-to-air laser link for the U.S. government.

Sophia Space, the new partner, is focused on addressing one of the primary hurdles for large-scale orbital data centers: thermal management. The company is developing passively-cooled space computers designed to prevent powerful processors from overheating without the need for heavy, costly, and complex active-cooling systems. In their collaborative effort, Sophia will upload its proprietary operating system to one of Kepler's satellites, aiming to launch and configure it across six GPUs on two separate spacecraft. This undertaking, considered standard practice in terrestrial data centers, marks the first time such an operation will be attempted in orbit. Successfully ensuring the software's functionality in space is a critical de-risking exercise for Sophia, ahead of its inaugural satellite launch scheduled for late 2027. For Kepler, this partnership serves to validate the utility and robustness of its in-orbit network.

The initial value proposition of orbital data centers will be found in edge processing—the ability to process data directly where it is collected to achieve faster responsiveness. This focus on distributed inference rather than centralized training distinguishes Kepler and Sophia from other prominent space companies like SpaceX and Blue Origin, and startups such as Starcloud and Aetherflux, which are raising substantial capital for visions of large-scale data centers utilizing traditional data center-style processors. Mitry elaborates on this distinction, stating, “Because we have the belief it’s more inference than training, we want more distributed GPUs that do inference, rather than one superpower GPU that has the training workload capacity.” He adds that running kilowatt-consuming GPUs only 10% of the time is inefficient, whereas Kepler's GPUs are designed to operate 100% of the time.

Experts suggest that large-scale, terrestrial-like data centers in orbit, as envisioned by giants like SpaceX and Blue Origin, are unlikely to materialize until the 2030s. The immediate future involves processing data collected in orbit to enhance space-based sensors for private and governmental applications. Once these foundational technologies are proven in orbit, the possibilities expand significantly. Sophia CEO Rob DeMillo also highlights external factors, pointing to recent municipal bans on data center construction, such as one enacted in a Wisconsin city, and similar legislative pushes in Congress. Such restrictions on terrestrial data centers, in his view, amplify the appeal of space-based alternatives, suggesting a future where orbital solutions become increasingly relevant: “There’s no more data centers in this [city],” Demillo mused. “It’s gonna get weird from here.”