Analysis
Google launched the first satellite in its Project Suncatcher program on October 1 aboard a SpaceX Falcon 9 from Vandenberg Space Force Base, carrying four of the same custom TPU chips that power its ground data centers into orbit for roughly a year of testing. The refrigerator-sized satellite, built by Planet Labs and fitted with Google's accelerators, is meant to answer a single question: can AI compute eventually run on solar power in space, where there is no water to draw for cooling, no land to clear, and no utility queue to wait behind. Fortune first reported the launch details on October 10.
Why Google Is Looking Up
Project Suncatcher has been running as a research effort since November 2025, born out of the same resource crunch reshaping terrestrial AI infrastructure: Big Tech is pouring hundreds of billions into data centers even as power grids, water rights and local permitting fights slow construction. Google's James Manyika framed the timeline bluntly: "Remember how Google was researching for like 15 years, before anything showed up? I think this is going to look like that" -- a sign this is a multiyear bet, not a near-term fix for today's power shortages. The solar panels on the test satellite supply only about one kilowatt, enough to run a microwave, and the TPUs can only operate in roughly 15-minute bursts before the radiators need to catch up, according to TechTimes. Heat, not radiation, turned out to be the harder problem: in a vacuum there's no air to carry heat away, so Google routes it through heat pipes to external radiators tested in a thermal vacuum chamber before launch.
“Google's James Manyika framed the timeline bluntly: "Remember how Google was researching for like 15 years, before anything showed up?”
A Race Google Didn't Start
Google isn't first. Starcloud beat it to orbit in November 2025, flying an Nvidia H100 chip that ran Google's own Gemma model in space before Suncatcher's satellite ever left the ground. Blue Origin filed plans in March for Project Sunrise, a constellation of 51,600 satellites built around the same orbital-compute thesis. Elon Musk has said SpaceX could eventually loft up to a million data-center satellites, predicting "the lowest-cost place to put AI will be space, and that will be true within two years, maybe three." Cowboy Space Corp added a third entrant this month, launching its Reason-1 satellite to test laser-based power transmission between orbiting units -- the same capability Google plans to test with two more satellites in 2027.
The economics turn on one number: SpaceX's launch costs have fallen roughly 95% since 2008, to around $1,000 per kilogram, according to Fortune's reporting. That curve is the entire bet. At today's prices, lofting enough compute to matter would cost far more than building an equivalent terrestrial data center even where power is scarce -- Google's own $43 billion nuclear power deal with Constellation, signed weeks ago to secure electricity for ground-based AI infrastructure, is a far more immediate and far larger commitment than anything in the Suncatcher program today.
That gap is the catch the headline misses: four TPU chips drawing one kilowatt of power for 15-minute bursts is a research instrument, not a data center. Google has not announced a timeline for commercial orbital deployment, and the laser-communication test that would prove satellites can work together as a cluster doesn't happen until 2027 at the earliest. Whether cooling, radiation-hardening and launch costs ever converge on numbers that beat building more nuclear or gas-fired capacity on the ground remains an open question the program is designed to answer, not one it has answered yet.
For now, Suncatcher is a hedge against a scenario terrestrial buildout can't fully solve: a world where AI compute demand keeps compounding faster than grids, water tables and local zoning boards can accommodate. If that scenario plays out, Google wants flight data already in hand. If it doesn't, the company has spent a modest sum -- alongside the billions it's committing to nuclear, gas and conventional data centers -- on an insurance policy. Blue Origin and SpaceX have each filed plans for satellite constellations far larger than Suncatcher's single test unit; neither has launched one yet, and Google's own mission won't produce a cost-per-compute figure until its year in low Earth orbit ends in 2027.
