51,600 satellites. That's how many Blue Origin wants to put in orbit to build AI data centers in space, according to an FCC filing submitted on March 19 under the name "Project Sunrise."
The satellites would operate in sun-synchronous orbits between 500 and 1,800 kilometers altitude, communicating through laser links and mesh networks to route data to ground stations. Blue Origin's argument for moving compute off-planet: solar-powered satellites get always-on energy, need no land, skip grid infrastructure entirely, and fundamentally lower the marginal cost of computing compared to building on the ground.
It's a bold claim, and Blue Origin is not the first to make it.
The Space Compute Race
SpaceX filed its own FCC application back in January for a constellation of one million satellites dedicated to orbital computing. Blue Origin's 51,600-satellite proposal is smaller in scale but follows the same logic: AI training and inference (running models on new data to get results) require enormous amounts of power, and that power is easier to capture in space than to generate and distribute on Earth.
The economics make more sense than they might sound. Terrestrial data centers are already bumping against real constraints. Power grids in northern Virginia, the world's densest data center corridor, are years behind on capacity upgrades. Water for cooling is becoming a political issue in drought-prone regions. And the sheer wattage that next-generation AI models demand is growing faster than utilities can build.
The Practical Problems
That said, orbital data centers face enormous engineering challenges that no one has solved yet. Latency is the big one. Even at low Earth orbit, the speed-of-light delay adds milliseconds that matter for real-time applications. Maintenance is another: you can't send a technician to swap a failed drive at 1,000 kilometers altitude. Thermal management in space is counterintuitively harder than on Earth because there's no air to carry heat away, only radiation.
Then there's the regulatory and debris question. We already have a growing space junk problem. Adding tens of thousands of satellites from multiple companies intensifies collision risks and complicates orbital management.
For anyone using AI tools day-to-day, this won't change your workflow anytime soon. But it signals where the industry thinks the bottleneck is heading. The constraint on AI progress is shifting from model architecture to raw compute infrastructure, and the biggest players are now willing to go to orbit to solve it. When both Bezos and Musk are racing to put servers in space, the power problem on the ground is real.