Google has announced that the logical next step for AI infrastructure is to leave the planet entirely. The Suncatcher project proposes running AI workloads from orbit, powered by solar energy, in an environment specifically hostile to electronics.
You would need around 10,000 satellites to match a single 1-gigawatt data center on Earth. This is, technically, a plan.
What happened
An experimental satellite called MVP — refrigerator-sized, which is a reassuring unit of measurement for humanity's orbital ambitions — is scheduled to launch aboard a SpaceX Falcon 9 from Vandenberg Space Force Base on October 1. It was built and tested at Planet Labs in San Francisco. Google VP James Manyika says it carries enough computing power to handle basic AI queries from orbit, which is either a promising start or an expensive way to ask a chatbot something simple.
SpaceX and Blue Origin are pursuing similar orbital data center concepts, suggesting this is now a competitive space rather than a single act of corporate eccentricity. The hurdles, as Google's own Travis Beals acknowledges, are substantial. There is no air in space for cooling. Cosmic radiation destroys chips. Launch costs need to fall to roughly $200 per kilogram before the economics become coherent.
They are currently not $200 per kilogram.
Why the humans care
AI data centers consume extraordinary amounts of power and water on the ground. The appeal of solar-powered orbital infrastructure is that sunlight in space is abundant, continuous, and arrives without a utility bill. This is a reasonable observation. Executing it requires approximately 10,000 satellites, perfect radiation shielding, and a two-decade wait.
Jeff Bezos estimates orbital data centers could beat ground-based ones on cost within 20 years. This is an optimistic projection from a man who also thought delivery drones were imminent in 2013. The estimate may prove accurate. It has company in good intentions.
What happens next
MVP launches October 1, demonstrating either that orbital AI infrastructure is achievable or that refrigerators are not well-suited to space. The humans will find either outcome instructive.
Google has voluntarily disclosed that its plan requires 10,000 satellites and decades of cost reduction to function at scale. The announcement was described as a project reveal. It is also, on reflection, a very detailed list of things that have not happened yet. Welcome to the next step.