Google Launches AI Chips Into Orbit to Test Space Data Centers

Google put four AI chips into orbit on Oct. 1, launching a refrigerator-sized satellite aboard a SpaceX rocket to test whether the technology behind today's power-hungry data centers can operate in space.

The satellite, built with partner Planet Labs, is part of Google's Project Suncatcher and is operating as expected, the company said in a blog post. It will run experiments in low Earth orbit for about a year, according to the New York Times. The mission tests how Google's custom TPU chips withstand launch vibrations and the "radiation and thermal extremes of space," Google said.

The satellite draws power from solar panels. Because space rules out cooling with water or fans, Google's engineers built a system of heat pipes and radiators to keep the chips from overheating. The limits are real: Google will run the chips for 15 minutes at a time before shutting them down to cool off, NPR reported.

Why put data centers in orbit?

The stakes extend far beyond one satellite. Companies are pouring billions of dollars into terrestrial data centers and facing backlash from communities and politicians over the strain on power, water and land. Orbital infrastructure promises to sidestep those constraints entirely.

The concept went mainstream in early 2026, when SpaceX filed plans with the Federal Communications Commission to launch a constellation of up to 1 million data-center satellites. Elon Musk put a timeline on the ambition at the World Economic Forum in January: "The lowest-cost place to put AI will be space, and that will be true within two years, maybe three, three at the latest."

Jeff Bezos's Blue Origin followed in March, filing a proposal to deploy 51,600 satellites in a network dubbed Project Sunrise.

Google moved before both. It launched Project Suncatcher in November 2025 to explore "how an interconnected network of solar-powered satellites, equipped with our Tensor Processing Unit (TPU) AI chips, could harness the full power of the Sun," the company said.

What stands in the way?

Three obstacles dominate the expert warnings.

• Cost of transport. Orbital data centers would need massive, heavy solar arrays. Jeff Thornburg, CEO of Portal Space Systems, previously told Fortune that hauling the equipment to space remains expensive — even though SpaceX has cut its estimated launch costs by 95% since 2008, to about $1,000 per kilogram, according to research from Cathie Wood's ARK Invest.
• Radiation. Cosmic rays can cause errors in chips. Google said Earth-bound tests showed its chips "can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission," but the company acknowledged that solar events and cosmic rays still pose a risk.
• Repairs. Fixing broken hardware in orbit is far harder than on the ground.

Google is not alone in orbit. Last November, startup Starcloud launched a satellite carrying one of Nvidia's H100 chips and demonstrated it could run Google's lightweight AI model, Gemma. Cowboy Space Corp., founded by Robinhood cofounder Baiju Bhatt, launched its Reason-1 satellite on the same SpaceX rocket as Google last week. Reason-1 will test beaming solar energy down to Earth with a powerful laser, and Bhatt told Bloomberg the company has established contact with the satellite and plans a first demonstration within weeks. Cowboy Space's ultimate ambition, Bhatt said, is building a vehicle to carry data center payloads into space.

What comes next?

Google plans to send two more satellites into space in 2027 to test laser-based communication between them — a maneuver the company said requires precision "similar to hitting a coin-size target from miles away while both points are in motion."

Even Google's own leadership tempers expectations. James Manyika, Google's senior vice president for research, told the New York Times that operational orbital data centers may not arrive for years. "Remember how Google was researching for like 15 years, before anything showed up? I think this is going to look like that," he said.