Martin Halliwell knows his stuff. He’s a Partner at NewSpace Capital and former CTO of SES. He’s watched this space get crowded. Now, it’s exploding.
We have about 16,000 satellites up there right now. That’s a lot of metal and silicon buzzing around Earth. But the numbers are going to get wild. Market intelligence firm Novaspace says we’re looking at 43,000 launches over the next decade. Goldman Sachs? They think 70,000 low-Earth orbit (LEO)satellites alone could hit the skies in just five years.
The exact count doesn’t matter. What matters is the scale.
These constellations are growing. And they’re getting too big for humans to manage by hand.
Why satellites need onboard AI processing
You can watch a dozen satellites from a ground station. You can even handle a hundred if you’re really good. But thousands? Forget it.
A network this size has to do things instantly. Share data. Avoid crashing into itself. React to sudden shifts in customer demand. Fix technical glitches while moving at 17,000 miles per hour. Humans are too slow. The latency alone would break the system.
So we need automation. We need artificial intelligence. Not to replace humans, but to handle the stuff that happens in seconds. Satellites will need to process their own data up in orbit, acting within strict boundaries set by us.
This is onboard processing for satellites —the shift from sending raw data to Earth to making decisions in space.
Take Earth observation. Right now, these satellites collect massive amounts of imagery. Weather patterns. Crop health. Disaster zones. Emissions. They beam it all down. Ground teams then clean, sort, and analyze it before customers can actually use it. That’s a bottleneck.
If satellites could clean that data in orbit, they’d send back only what matters. You cut costs. You save time. In defense? Speed wins wars. Faster data means faster action. Plus, it reduces reliance on ground stations, which are vulnerable to attacks or simple disruptions. The satellite itself becomes part of the network’s resilience.
Dynamic capacity management with AI
Satellites aren’t just cameras or sensors. They’re routers in the sky.
And the demand for their bandwidth is chaotic. It spikes over cities during rush hour. It vanishes over the ocean during lulls. It’s needed urgently in disaster zones or for military ops.
A fixed beam pointing at the same spot 24/7 is inefficient. It wastes power. It wastes spectrum.
AI can watch these patterns. It can spot the shift. It can reorient beams in near real-time. It can decide how much power to dump into a beam over a hurricane zone versus a sleepy suburban area. It’s dynamic. It’s responsive.
This means better use of limited resources. The satellite network becomes a smart utility, not a dumb pipe.
Who is responsible when the AI messes up?
Here’s the rub.
The tech is ready. The physics works. The software exists. But the legal and organizational frameworks? They’re stuck in the 1990s.
Most laws assume a human is at the helm. A named person signs off. A named person makes the call. If something goes wrong, you know who to sue or fire.
AI blurs that line.
What if an autonomous system points a high-gain beam at the wrong receiver? What if it disrupts a nearby satellite? What if it causes physical damage? Who is at fault?
The operator? The manufacturer? The software dev?
It’s murky.
Operators are cautious. They’re happy to let AI suggest moves. They’re not ready to let AI move capacity without human approval. Not yet.
Adoption will be slow. Painfully slow. We need clear rules. Boundaries. What can the machine do alone? What needs a human thumb on the scale? Until we answer that, the industry will hesitate.
Designing spacecraft with AI assistance
AI isn’t just for flight ops. It’s changing how we build these things.
Engineers can use AI to write code. To scour technical documents for past failures. To test hypotheses faster than a human brain ever could. It can suggest structural designs. Antenna layouts. Power configurations.
It generates early drafts. Not the final product, but a starting point.
This shortens development cycles. Cuts costs. Allows smaller teams to take on projects that once required armies of engineers. It’s not about replacing skill. It’s about offloading the routine. So the humans can focus on safety. On the hard problems. On innovation.
Factories will use it too. Spotting defects. Predicting maintenance needs. Keeping the line moving.
The trust deficit
To make this work, we need trust.
And trust requires security.
Organizations won’t feed their sensitive data into public cloud models if they can’t control where it goes. They need secure, private environments. Strong cybersecurity. Clear access rules. Data segregation.
Some models might need to be trained inside government networks. On-premises. Where the data never leaves the building.
Technical safeguards are part of it. But the real barrier is confidence.
The industry needs better standards. Clearer legal liability. Reliable testing protocols.
The technology is advancing by leaps and bounds. The regulations? They’re crawling.
AI might be ready to take the wheel of satellite networks next year.
The question is whether we’re brave enough—or careful enough—to let it.
