Milk Road AI@MilkRoadAI
The world is running out of places to put data centers (Save this).
And two of the most powerful companies on the planet think the solution is to leave the planet entirely.
Every AI model you use runs on infrastructure that drinks electricity like a small city and consumes water like a farm.
Grid constraints are already threatening to block roughly 40% of planned data center builds because there simply isn't enough power or land to support it at the scale AI demands.
So SpaceX filed an application with the FCC to launch up to a million satellites, each one functioning as an orbital data center running on free solar energy, tracking the dawn-dusk line in sun-synchronous orbit for near-constant sun exposure.
Blue Origin followed months later with its own proposal, Project Sunrise, a constellation of over 50,000 dedicated compute satellites using laser inter-satellite links to pass data between nodes.
The pitch is elegant but the engineering is brutal.
Cooling, trivially solved on Earth with air and water, has no equivalent in space.
Satellites would need closed-loop radiative panels, circulating coolant over the chips, radiating heat as infrared into deep space, and cycling it back and those panels are heavy, which drives up launch costs.
Radiation is a second problem with no clean solution.
Cosmic particles randomly flip bits inside chips corrupting computation., you either detect and correct the errors, run every calculation three times in parallel and vote on the right answer, or physically shield every server.
All three approaches add mass, cost, or latency and bandwidth may be the most underappreciated constraint of all.
Getting data from Earth to orbit and back over radio frequency is nowhere near fiber optic speed.
The most viable near term use case isn't general cloud computing, it's AI inference specifically launch the satellite preloaded with model weights, send up a short query, receive a short answer.
That's feasible but streaming training data back and forth is not.
There's also a competitive subplot worth noting, Amazon whose founder Jeff Bezos is building orbital computing through Blue Origin formally petitioned the FCC to reject SpaceX's application, calling it speculative and facially incomplete.
This isn't a debate about whether orbital data centers make sense. but rather about who captures the orbital spectrum and the regulatory runway before anyone else can build.
One million satellites would be more than 60 times every satellite currently in orbit and the debris and spectrum consequences alone are enormous.
Engineers are clear: large-scale orbital data centers are a 2030s reality at the earliest dependent on Starship dramatically cutting launch costs, radiation-hardened AI chips that don't yet exist commercially, and cooling architectures still being designed.
Whoever wins this race owns something unprecedented, compute infrastructure running on free energy, above every country, beyond every grid, serving the entire planet at once.
