For years, “data centers in space” sounded like a category built more for keynote slides than for customers. That picture is starting to change, but not in the way the grandest hype promised. In TechCrunch's April 13 report, Kepler Communications said the largest compute cluster currently in orbit was launched in January and now links about 40 Nvidia Orin edge processors across 10 operational satellites using laser communications links.

The number that makes the story feel real is not just the hardware count. Kepler says it already has 18 customers, and its newest customer, Sophia Space, will test its operating system by configuring software across six GPUs on two spacecraft. TechCrunch described that as the first time this kind of data-center-style software setup will be attempted in orbit. That is a very different claim from the giant orbital AI factory narratives tied to later-decade plans from much larger players. This is narrower, more grounded, and arguably more useful in the short term.

Kepler's near-term thesis is that orbital compute will first make sense for inference and sensor processing close to where the data is collected, not for giant training jobs in space. The company explicitly contrasts that approach with visions of super-sized space data centers expected in the 2030s. CEO Mina Mitry told TechCrunch that the company prefers distributed GPUs that can run inference continuously rather than massive processors that consume large amounts of power while sitting underused. That is an edge-computing argument, just moved off the planet.

The commercial and strategic angle is broader than it looks. If satellites can process more of their own data in orbit, customers do not need to move every bit back to Earth before acting on it. That matters for remote sensing, radar, and government workloads where latency and bandwidth are expensive. Kepler even points to the U.S. military as an important customer for that kind of processing. Space data centers are still far from normal, but this cluster suggests the first viable market is not sci-fi cloud compute. It is targeted, always-on inference where the data is born.

#kepler

The first serious orbital GPU cluster is live with 40 Nvidia Orins in orbit