A University of Manitoba simulation found that orbital data centers placed near imaging satellites could return processed results to spacecraft in about two seconds. Ground customers waited much longer: the modeled median was 14 minutes, compared with 22 minutes for ground processing.
Peng Hu, Mustafa Alhassan, and Amjad Iqbal’s October 23, 2026 preprint combines small computing nodes near customer spacecraft with larger nodes in a separate orbit. Nearby nodes handle immediate processing; larger nodes provide heavier computing capacity.
Processing reduced each modeled image from 2 GB to a 20 MB product. This changes the transmission requirement: a user receives an extracted result instead of the complete image.
A geostationary relay eliminated the mixed architecture’s ground-latency advantage in the simulation. Faster computing alone did not remove the wait for communications access. These are simulation results: processor speed was extrapolated from hardware specifications, and the model excluded weather.
Why Computing Capacity Depends on Connections
Google’s Project Suncatcher research explains why orbit and communications must be designed together. Its November 2025 architecture proposed a dawn-dusk orbit to collect sunlight almost continuously, reducing battery requirements. It also proposed tightly grouped satellites connected by optical links.
Large distributed artificial intelligence workloads exchange data repeatedly between processors. Google estimated that its proposed system would need inter-satellite connections carrying tens of terabits per second. Its laboratory demonstration achieved 800 gigabits per second in each direction with one transceiver pair. This was a bench test, not an orbital network.
The Destination Changes the Service
The distinction between processing and delivery separates two services. A spacecraft can use a nearby processor’s result without sending it to Earth. A ground customer needs a complete path through the space network and a downlink.
New Space Economy’s analysis of orbital AI workloads identifies proximity to space-generated data as a specific reason some applications benefit from computing in orbit. Processing near a sensor can discard unwanted observations, prioritize detections, or return a compact product instead of transmitting the entire collection.
