Google is preparing to put an AI satellite into orbit next week, according to The Verge — the first time the company's long-discussed plan to run machine-learning hardware in space moves from research paper to flying hardware.
The effort traces back to Project Suncatcher, a research programme Google outlined for solar-powered TPUs in orbit. The core idea is deceptively simple. A satellite in a dawn-dusk, sun-synchronous orbit stays in sunlight almost continuously, so its solar array never has to carry a battery sized to survive a long eclipse. The heat that ground data centres struggle to shed would instead radiate into deep space, where the cold is essentially free.
Google's researchers argued the economics could close by the mid-2030s — but only if launch costs fall to roughly $200 per kilogram, a fraction of what heavy-lift rides cost when the paper was written. They also published supporting work: radiation testing on Trillium TPUs suggesting the chips could survive around five years in low Earth orbit, and short-range demonstrations of free-space optical links between spacecraft.
The previously stated plan was for a small prototype constellation, developed with Planet Labs, targeted at 2027. A launch next week implies the schedule has moved up. Google has not published full details of what the first spacecraft carries, how long the demonstration will run, or whether it flies any TPU silicon at all — a single pathfinder satellite is a radiation-and-thermal experiment far more than it is a data centre.
Why it matters: the hard part of orbital compute was never the chip. It is the constellation economics — inter-satellite bandwidth, limited ground-station time, servicing and deorbiting, and the brutal head start terrestrial operators have in cost per FLOP. Google, SpaceX and a wave of startups are all betting that falling launch prices eventually erode that advantage. Project Suncatcher is the most public attempt to find out whether they are right.
Caveats worth holding onto: mission specifics remain thin, and a single launch settles none of the economics. Treat next week's flight as the start of the measurement, not the result.
