Martin Halliwell is a Partner at NewSpace Capital, one of the world's first private equity firms devoted exclusively to growth-stage companies working in the space technology sector. He formerly served as Chief Technology Officer of SES, where he led global technology and R&D from 2011 to 2019. Halliwell contributed this article to Space.com's Expert Voices section.
For most of the space age, satellites have been puppets on a string. Every meaningful decision — where to point a camera, when to fire a thruster, how to route a signal — was made by engineers on the ground and radioed up on a schedule.
That model is starting to break down, not because it stopped working, but because the number of satellites in orbit has exploded. A single company can now operate thousands of spacecraft at once, and no ground team can watch that many moving parts in real time. Artificial intelligence is stepping into the gap, and it's changing three things in particular: how individual spacecraft think for themselves, how whole constellations act as a team, and how the communication networks those satellites form get managed.
AI on the spacecraft itself
The simplest change is that satellites are starting to make more decisions on their own, onboard, without waiting for instructions from Earth. This matters because of a basic physics problem: a satellite in low orbit is only in radio contact with a given ground station for a few minutes at a time, and even geostationary satellites face communication delays and limited ground-station bandwidth. If a satellite has to wait for a human to notice a problem, approve a response, and send it back up, that can take hours. An onboard AI system can notice something interesting — a wildfire, a ship, a hardware fault — and act on it immediately.
NASA's Jet Propulsion Laboratory has been testing this idea directly, with onboard AI that analyzes data as it's collected and decides what's worth keeping, what's worth acting on, and what to ignore. The U.S. Air Force Research Laboratory has gone a step further, demonstrating a neural network that controlled a satellite's orientation in orbit without a human in the loop — AI managing not just the payload, but the spacecraft's core flight functions. Newer Earth-observation satellites are even carrying small, purpose-built processors (similar to the chips used in some laptops) so they can combine different types of sensor data on the spot and beam down a finished answer instead of a mountain of raw data for humans to sort through later.
AI for managing constellations
Once you have hundreds or thousands of satellites flying together — a "constellation" — a new problem appears: they need to behave like a coordinated fleet, not thousands of individuals. This is one of the areas where AI has the clearest, most practical role.
The most visible example is collision avoidance. Low Earth orbit is increasingly crowded with active satellites and debris, and a mid-size constellation can face dozens of close-approach warnings a day, each requiring a decision about whether to move the satellite out of the way. That volume is beyond what human teams can evaluate one by one, so operators are turning to AI systems that ingest tracking data, calculate collision risk, and in some cases plan or even execute an avoidance maneuver without waiting for a person to sign off. NASA's Starling project and related "distributed spacecraft autonomy" work go further still, teaching small satellites to share data with each other, divide up observation tasks, and adjust their collective plan on the fly — essentially letting the constellation manage itself as a team rather than as separate spacecraft that happen to share an operator.
AI for network orchestration
The third shift is in how satellites talk to each other and to the ground — and this is where AI may have the biggest impact on ordinary users, because it's what makes services like satellite broadband fast and reliable. A modern broadband constellation is a network in constant motion: every satellite is moving at roughly 17,000 mph (27,400 kph), links between satellites and ground stations are constantly breaking and re-forming, and demand for bandwidth shifts by time of day and location. Deciding which satellite should talk to which ground station, how to route a data packet through a mesh of satellites overhead, and how to shape each satellite's radio beam to serve the busiest areas — all of that has to be recalculated continuously.
This is a scale of real-time optimization that traditional, rule-based network management can't keep up with. Operators are turning to AI models — often the same family of technology used for demand forecasting or logistics — that can predict traffic patterns hours in advance and continuously re-optimize beam shapes, routing paths, and satellite handoffs across the whole constellation at once. SpaceX's Starlink network, for example, has described using this kind of AI-driven system to manage traffic across its laser inter-satellite links and adjust beam patterns in real time as demand shifts around the globe. Researchers are also applying newer AI techniques, including models that treat the whole constellation as a connected graph, to make these routing decisions faster and more efficient as networks scale into the tens of thousands of satellites.
Why it matters
None of this replaces human oversight — operators still set the rules, review anomalies, and can step in when something looks wrong. But it does change the job: instead of manually commanding every satellite, ground teams increasingly supervise AI systems that handle the routine, high-volume decisions on their own. As constellations grow from hundreds of satellites to the tens of thousands now being planned, that shift from "human commands every satellite" to "human supervises an AI-managed fleet" isn't just convenient — it's becoming the only way these systems can keep working at all.
Editor's note: This story was updated at 5:50 p.m. ET on Aug. 7 with a significantly revised version.
