Kepler Communications now operating 40 GPUs in orbit, the largest orbital compute cluster to date

The largest compute cluster currently in orbit went live in January, and it is already attracting customers who want to run workloads in space.\n\nCanadian satellite operator Kepler Communications launched the constellation with about 40 Nvidia Orin edge processors distributed across 10 operational satellites. All nodes are linked via laser communications. The company announced Monday that Sophia Space, a startup building passively-cooled orbital computers, has signed on as its newest customer.\n\nThis matters because the economics of orbital AI are brutal. Power is scarce. Cooling is harder than on Earth. Radiation flips bits. Until now, the idea of serious compute in orbit has been mostly theoretical. Kepler is treating it as a business.\n\nSophia Space will upload its proprietary operating system to one of Kepler\u2019s satellites and attempt to launch and configure it across six GPUs on two spacecraft. That kind of software deployment is routine in terrestrial data centers. It has never been done in orbit before. If it works, it validates a key assumption behind Sophia\u2019s own hardware roadmap. The startup plans its first satellite launch in late 2027.\n\nKepler CEO Mina Mitry has been clear that the company does not see itself as a data center operator. The pitch is infrastructure. Kepler wants to be the network layer that other satellites, drones, and aircraft plug into. The compute resources are a side effect of that architecture.\n\nExperts expect true large-scale orbital data centers - the kind SpaceX and Blue Origin have talked about - sometime in the 2030s. The near-term market is different. It is about processing sensor data where it is collected, reducing the amount of raw information that needs to downlink to Earth. That matters for Earth observation, for certain defense applications, and for any sensor network that generates more data than bandwidth allows.\n\nSophia\u2019s angle is cooling. Active cooling systems in space are heavy and expensive to launch. The company is betting that passively-cooled designs can handle powerful processors without the mechanical complexity. Testing that theory on Kepler\u2019s live cluster is cheaper than launching dedicated hardware.\n\nKepler now has 18 customers. The constellation is operational. The company is issuing calls for additional hosted payloads that need persistent, real-time data streaming in low Earth orbit.\n\nWhether orbital compute becomes a real market or remains a curiosity depends on whether customers like Sophia can make the economics work. The hardware is there. The next test is the software.