For years, satellite operations have followed a familiar pattern: collect data in orbit, downlink it when possible and analyze it later. That model has supported a wide range of Earth observation missions, but it has also constrained how quickly insights can be generated and acted on.
Artificial intelligence is beginning to change that paradigm. As operators explore ways to move processing closer to the sensor, satellites are starting to shift from passive data collectors toward systems that can interpret, prioritize and act on information in orbit. At the same time, that transition is exposing new technical constraints, operational trade-offs and questions about how these systems should be used.
Read our top four takeaways from our conversation with Paul Lasserre, general manager, AI for Space at Loft Orbital, or listen to the full episode.
Takeaway 1: The shift from “pixels to events” is redefining how satellites are used.
Satellite workflows today are still largely built around delayed analysis. Operators collect large volumes of imagery and other sensor data, then wait for a ground pass before processing it – a model that inherently limits responsiveness.
“We operate today basically the same way we have for the past decade… you need to wait for your satellite to fly over a ground station, dump all of these heavy data, and later analyze it,” Lassere said.
Because of that latency, most satellite-derived insights have historically been used for strategic, after-the-fact decision-making rather than real-time operations, he said.
“But what we’ve witnessed on Earth across industries over the past decade… where AI today helps make decisions in real time has not traveled to space yet,” he said.
That gap is now beginning to close as AI models are deployed closer to the sensor, enabling systems to identify specific conditions and transmit only what matters.
“That’s really what we observe today, this fundamental shift from pixels to events,” Lasserre said.
Instead of downlinking full datasets, satellites can send small, targeted alerts tied to predefined conditions – such as a wildfire or a vessel leaving a harbor – in near real time, according to Lasserre.
“[You can] send an insight – just a few kilobytes of data… to let someone know in seconds that the wildfire is starting,” he said.
Takeaway 2: Moving AI into orbit is less about scale – and more about constraints.
While discussions around AI in space often focus on replicating cloud-like infrastructure in orbit, current approaches are far more constrained by physics, Lasserre noted. Power, radiation and connectivity limitations all shape what can realistically be done onboard, he said.
“AI in space does not mean… to replicate the infrastructure on Earth in space,” Lasserre said.
For Earth observation, the goal is to leverage AI to bring as little intelligence to the sensor as possible, instead bringing that intelligence to the edge, he said.
But constraints in space are different from those on Earth, said Lasserre. Satellites must operate with limited power budgets, hardened compute and intermittent connectivity, making it impractical to run the same models used in terrestrial data centers, he said.
“There is no power plant in space… you just need to deal with whatever solar panel you have,” Lasserre said.
As a result, system design becomes a question of tradeoffs – deciding which tasks require immediate, low-latency processing and which can wait for more robust analysis on Earth, Lasserre said.
“You need to think hard about the tradeoff on what you want to process at the edge... and what to really process with more generous power budgets,” he said.
Takeaway 3: AI changes workflows by prioritizing signals – not replacing human judgement.
As satellites begin filtering and prioritizing data in orbit, the definition of what constitutes a meaningful event remains grounded in human input, Lasserre said. Operators or customers set the conditions, and models identify when those conditions may be met.
“The conditions will… always be set by a human… and the model is then going to… surface a risk of this condition being met,” Lasserre said.
Even with humans and models working together, uncertainty is inherent, especially when conditions around a satellite can be unpredictable, said Lasserre.
“You will have false positives… [and] false negatives by nature of… the laws of physics,” he said.
Rather than delivering perfect answers, these systems function as an early warning layer – scanning large areas and surfacing signals that warrant closer inspection, Lasserre said.
“You basically look at vast amounts of territories or at sea, and you act as a broad net that you’re casting, and the model is helping you and giving you hints on where to look more carefully,” Lasserre said. “And if anything, you can treat the system as a reordering queue workflow.”
That shift has implications for how analysts work. Instead of reviewing all incoming data equally, they can prioritize based on AI-generated alerts, he said.
“You can treat the system as a reordering queue workflow,” he said.
Takeaway 4: The next phase of AI in space is coordination across constellations.
Beyond individual satellites, AI is beginning to enable coordination across multiple assets – a capability that has historically been limited, Lasserre said.
“Until now AI in Earth observation was mostly processed at the edge on the single satellites. And so this notion of coordination across assets was not really a thing,” Lasserre said.
That is starting to change as systems evolve to translate high-level intent into coordinated actions across a constellation, he said.
Satellites in network can start with a mission objective, then work backward, Lasserre said. Sensors identify things like cloud cover or wide‑area activity, then cue other satellites to capture precise, high‑resolution imagery., coordinating the entire system around that goal, he said.
This kind of orchestration allows satellites to detect, validate and refine observations collaboratively, improving both coverage and responsiveness, Lasserre said.
Over time, that coordination could extend beyond a single operator’s fleet, pointing toward more interconnected systems, he said.
“You can imagine this coordination not only happening at the constellation level, but… at a pooling and sharing level of resources,” Lasserre said.
For more, listen to the full episode.
Explore More:
Rethinking Satellite Operations: More Autonomy Key to Fewer Collisions
Agentic AI for the Satellite Operator
AI, Spacecraft Autonomy and Space Data for Humanity
