An image of the ocean can run to 10 gigabytes. If all a customer wants to know is where one ship is, that is a lot of data to push down to Earth. A satellite that can think for itself sends back the vessel’s coordinates, speed and heading instead, a few kilobytes in all.
That is the idea behind on-orbit edge processing, which Lorne Graves, chief technology officer of Frontgrade Technologies, laid out on the Space Minds podcast hosted by SpaceNews editor Mike Gruss. Graves tied the shift to the rise of commercial megaconstellations and to a space domain that is now contested.
The old way
For decades, satellites in geostationary or low Earth orbit collected optical images, radar data or telecom signals and downlinked all of it. Ground-based supercomputers did the analysis. The model kept spacecraft small, light and cheap, but it had costs.
Sensors such as synthetic aperture radar, hyperspectral imagers and high-resolution cameras produce terabytes of raw data every day. Radio spectrum and ground-station contact windows limit how much can come down. Delay is a second problem: for missile tracking, wildfire detection or military targeting, the time spent downlinking and processing can make the answer useless. Constant high-bandwidth downlinks are also easy targets for jamming and spoofing.
What changed in orbit
The Apollo Guidance Computer that supported the 1969 lunar landing ran at about 2.048 MHz with 72 kilobytes of ROM. Later, dedicated radiation-hardened chips protected electronics from the Van Allen belts and deep space, but that market was small. Space-grade processors fell several generations behind commercial silicon in speed and transistor density.
Megaconstellations in low Earth orbit, among them Starlink, OneWeb and Amazon’s Project Kuiper, broke that pattern. Low Earth orbit has less radiation than geostationary orbit, so operators could use commercial off-the-shelf parts that offer far more processing power at a fraction of the price. The result was software-defined payloads that can steer beams, route traffic and manage data on board.
Hard limits
Space-based computing still runs into radiation, heat and power. High-energy protons, heavy ions and cosmic rays wear down semiconductors over time, and a single charged particle can flip a bit of memory or, in the worst case, permanently destroy a chip through a short circuit.
The usual fix is a hybrid design. A high-performance commercial processor is paired with a radiation-hardened supervisor, the kind Frontgrade makes, which watches for errors and resets the system after a radiation event.
Heat is harder. With no air in a vacuum, fans are useless, and warmth leaves only by conduction through the structure or by radiating infrared. Thermal design is often what caps how much a satellite can compute. Every watt also has to come from the solar arrays and batteries, and extra weight raises launch costs, so designers favor efficient chips such as FPGAs, ASICs and low-power neural processors built for AI.
Lessons from electronic warfare
Graves spent much of the discussion on electronic warfare. Satellites now face uplink and downlink jamming, GPS spoofing and cyber intrusion. On the ground, EW systems detect and counter hostile signals within milliseconds. Space systems never needed that speed because space was treated as a sanctuary: when interference hit a satellite, controllers studied telemetry for days or weeks before uploading a fix. In a modern conflict, that is too slow, and satellites need to detect jamming on their own.
