There's Now a Computer Cluster in Orbit Running at 100% Utilization
Kepler Communications launched the largest compute cluster currently in orbit in January 2026. The Canadian company's network spans 10 operational satellites carrying approximately 40 Nvidia Orin edge processors, linked together by laser communications. It has 18 paying customers.
What It Actually Does
Kepler's CEO describes the workloads as inference-dominant rather than training. That distinction matters. Inference runs well on distributed, lower-power GPUs. Training requires dense interconnects and massive memory bandwidth that space hardware cannot realistically provide yet.
The utilization rate stands out: 100%. Ground-based data center hardware routinely sits idle between jobs. Orbital hardware apparently does not. Whether that reflects genuine demand or simply a small initial deployment is worth watching as the cluster scales.
The U.S. military is among the 18 customers. Kepler has demonstrated a space-to-air laser link for the U.S. government. Military applications focus on missile defense, using satellites to detect and track threats in real time.
The Hard Problems
Heat is the obvious one. Sophia Space is developing passively cooled space computers specifically to address overheating in orbital compute environments. In its first planned test, Sophia Space will upload its proprietary OS to one of Kepler's satellites and configure it across 6 GPUs on 2 spacecraft. That configuration has not been attempted in orbit before. Sophia Space plans its own first satellite launch for late 2027.
The larger vision, orbital data centers comparable to what SpaceX or Blue Origin have described, is not expected until the 2030s. Startups Starcloud and Aetherflux are currently raising capital toward that goal, but they are raising capital, not operating hardware.
Where This Sits
Kepler's cluster is real and operational. The utilization numbers are credible given the application profile. Inference at the edge, latency-sensitive military workloads, and customers willing to pay for satellite infrastructure are a coherent early market.
The gap between 40 Orin processors on 10 satellites and a functional orbital data center is significant. The 2030s timeline for large-scale deployment is probably not pessimistic.
Source: Techcrunch