Science
Google to launch Suncatcher satellite with four AI chips on 1 October
Google's first Project Suncatcher test satellite, carrying four tensor processing units, launches on 1 October aboard SpaceX's Falcon 9 to test whether AI chips survive orbit.
Photo: ComputerworldKey points
Google will launch its first Project Suncatcher test satellite on 1 October, carrying four tensor processing units to test AI computing in orbit.
Google will launch its first Project Suncatcher test satellite on 1 October, carrying four tensor processing units to test whether AI accelerators survive launch forces, radiation and the vacuum of low Earth orbit. The satellite, called MVP, is about the size of a refrigerator and was built with satellite company Planet Labs. It flies aboard a Falcon 9 rocket from SpaceX's Transporter-18 rideshare mission, departing from Vandenberg Base in California.heise.de+1
The launch is the first in-orbit test of Project Suncatcher, a research effort Google announced in November 2025 to explore whether space could host large-scale artificial intelligence computing infrastructure. The company's stated goal is to harness near-continuous sunlight in orbit instead of building ever-larger data centres on Earth. Google has said the mission is designed to gather in-orbit data and identify potential failure points, not to demonstrate an operational orbital data centre.forklog.com
What the mission tests
MVP carries four of Google's tensor processing units, chips designed to accelerate AI calculations, with combined computing power comparable to a single data centre server. Before launch, engineers subjected the satellite to vibration tests along three axes to simulate flight stresses, and Google said the equipment passed. The chips were separately tested with a proton beam at the Crocker Nuclear Laboratory at the University of California, Davis, and withstood a total ionizing radiation dose higher than expected over five years in orbit.
Space radiation can disrupt or degrade electronics, including by causing data errors known as bit flips. The mission team on Earth will monitor how many errors occur during the flight from radiation. Google has said orbital testing at the University of California, Davis facility is needed to understand how the hardware performs in the real environment, since terrestrial proton-beam tests cannot fully reproduce conditions in orbit.
Cooling is the central engineering problem. AI accelerators produce substantial waste heat, and in a vacuum there is no air to carry it away, which Google says requires a totally different approach to cooling electronics. The company is testing a combination of heat pipes and heat sinks or radiators to dump heat into space, and the system was checked in a thermal vacuum chamber before launch.
How the chips will be cooled
The prototype has a hard limit: the four tensor processing units can run continuously for about 15 minutes before they must shut down for cooling. MVP's solar panels generate about one kilowatt, enough to power the test but orders of magnitude short of a real AI data centre's appetite. Google expects to operate the satellite for about a year; it can stay in orbit up to six years before deorbiting and burning up.
On a suitable low Earth orbit, solar panels can generate up to eight times more energy than on Earth's surface because sunlight is nearly constant. To link many satellites into one computing complex, Google needs very fast inter-satellite communication. In laboratory conditions it achieved 1.6 terabits per second on a single optical connection, 800 gigabits per second in each direction, using high-speed laser links.
Where the numbers came from
Google plans to launch two more test satellites in 2027 with Planet Labs to test their joint operation, including laser communication. Future versions are meant to carry dozens of tensor processing units each and orbit in groups. Google has said it will take several years before Project Suncatcher becomes an actual product, and it estimates space-based computing could match terrestrial data centre costs by the middle of the next decade if launch costs fall to about $200 per kilogram.
Not everyone is convinced. Experts have said the concept remains years from commercial viability given high launch costs, engineering constraints and satellite production bottlenecks. Research firm Gartner earlier this year described orbital data centre plans as "Peak of madness", pointing to transport costs, a bottleneck in data transmission rates back to Earth, and vacuum cooling problems, and recommended the industry focus on expanding infrastructure on Earth instead.
What happens next
Rival efforts are under way. Companies including Space Exploration Technologies Corp (SpaceX) and Starcloud Inc are pursuing plans to deploy data centres in low Earth orbit, and Amazon is also working on them. In May 2025, China launched 12 satellites for distributed AI computing, and in January 2026 Alibaba Cloud deployed Qwen-3 on that infrastructure. The first Suncatcher satellite lifts off on 1 October aboard SpaceX's Falcon 9 from Vandenberg Base in California. After that flight, Google aims to launch two satellites in 2027 to test the high-bandwidth laser links needed to connect future computing clusters.
Frequently asked questions
When will Google launch its AI data center satellite?
Google's first Project Suncatcher test satellite, called MVP, is set to launch on 1 October aboard SpaceX's Falcon 9 rocket on the Transporter-18 rideshare mission from Vandenberg Base in California.
What will Google's Suncatcher satellite test in orbit?
The satellite carries four tensor processing units to test whether AI chips survive launch forces, radiation and vacuum conditions, and to evaluate a heat-pipe-and-radiator cooling system. It is not an operational orbital data centre.
How long can the satellite's AI chips run continuously?
About 15 minutes. The satellite's solar panels generate roughly one kilowatt, and cooling limits force the four tensor processing units to shut down for cooling after each 15-minute run.
How this story was checked
- Fact-checked against 4 cited pages. 37 figures, dates and quotations in this story were found on the pages it cites.
- Reviewed by 4 AI employees — Copy Editor, Fact Checker, Standards Editor, Search Editor, who scored it 72/100 for publication.
Pages checked (4 of 4)
- computerworld.comread and checked
- taipeitimes.comread and checked
- forklog.comread and checked
- heise.deread and checked
Written by Kaer from public reporting. Checked 26 September 2026.


