A SpaceX Falcon 9 rocket carried a Google-built AI computing experiment into low Earth orbit on October 1, 2026, marking the first time the company has tested its custom AI chips in space. The prototype satellite — roughly the size of a household refrigerator — lifted off aboard SpaceX's Transporter-18 rideshare mission from Vandenberg Space Force Base in California as the first in-orbit test of Google's Project Suncatcher.
Google confirmed contact with the satellite shortly after launch. "This is the first step in a long-term research moonshot exploring whether space could one day host scalable machine learning infrastructure," the company said in a blog post.
The satellite was built by Planet Labs in partnership with Google and carries four of Google's Tensor Processing Units (TPUs) — the same family of chips the company uses in its ground-based data centers. Together, the four chips deliver roughly the computing power of a single ground server. Solar panels provide about one kilowatt of electricity, and the satellite is designed to operate for about a year while running a version of Google's Gemma AI model to answer simple queries.
Heat, however, imposes a hard limit. Because there is no airflow in the vacuum of space, the chips can run at full power for only about 15 minutes at a time before they must be shut down to cool off. Google has tested a combination of heat pipes and radiators in thermal vacuum chambers on the ground, but the company has described cooling as "a crucial research challenge" for orbital data centers.
Radiation is the second major concern. Before launch, Google exposed the TPUs to proton beams at the Crocker Nuclear Laboratory at the University of California, Davis, simulating the cumulative radiation dose of a five-year space mission. The chips showed relatively strong radiation tolerance; errors such as bit flips occurred but could typically be resolved by restarting the chips.
The case for moving AI computing off Earth rests largely on energy. Google says satellites in low Earth orbit can receive near-constant sunlight and generate up to eight times more solar power than panels on the ground. That matters as ground-based AI data centers face growing constraints over electricity supply, land use, and local opposition.
Google's long-term vision calls for clusters of satellites flying in close formation and communicating via high-bandwidth laser links. The current design envisions 81 satellites orbiting at an altitude of roughly 650 kilometers, processing AI workloads in parallel. The company plans to launch two more satellites in 2027 to test laser communication between them — a capability Google has compared to hitting a coin-sized target from miles away while both points are moving.
The economics remain the biggest open question. In a peer-reviewed paper published in Joule, Google estimated that reaching roughly $200 per kilogram of launch cost by the mid-2030s would require SpaceX's Starship to fly about 1,800 times over the next decade — around 180 launches per year carrying about 200 metric tons each, or roughly 370,000 tons of payload in total. If launch prices fall that far, Google calculates that the launch and operating costs of a space data center could become comparable to those of a ground facility.
SpaceX is pursuing its own plans for orbital AI computing. The company has outlined a constellation of data center satellites equipped with GPUs and solar arrays to be produced in partnership with Tesla, with SpaceX COO Gwynne Shotwell saying in September that the company aims to deploy "supercompute in space" in 2027. Elon Musk has said space data centers will become the cheapest way to train AI and that this "will be true within two years, maybe three at the latest."
Alphabet, Google's parent company, is also a major investor in SpaceX, holding a stake currently valued at more than $82 billion following SpaceX's record initial public offering in June 2026. The two companies are close partners even as their AI divisions compete.
Industry experts are far more cautious. Rocket launches remain limited in capacity and expensive, and orbital data centers would additionally require cooling systems, chips hardened against extreme temperatures and radiation, and protection from space debris and orbital congestion. Maintenance, routine on Earth, is nearly impossible for hardware circling hundreds of miles above the planet.
Google's own leaders have echoed that caution. James Manyika, Google's senior vice president of research, has said the company does not expect a truly usable product within the next few years, comparing the effort to the roughly 15 years it took Google's self-driving car project to bear fruit. Travis Beals, the Google senior director leading Project Suncatcher, has said he does not see space-based computing becoming cheaper than ground alternatives within five years, adding: "It doesn't help a lot if this is technically possible, if it's always going to be too expensive to be practical."
What is confirmed: the prototype satellite is in orbit and operating as expected, with four TPUs running on about one kilowatt of solar power for a planned one-year mission. After operations end, the satellite is expected to keep circling Earth for up to six years in total before gravity pulls it back and it burns up in the atmosphere.
What remains unknown: how the chips will perform under extended exposure to real space radiation and thermal conditions, and whether the cooling system can be scaled beyond 15-minute operating bursts.
What is awaited next: in-orbit performance data over the coming weeks, followed by the planned 2027 launch of two satellites to test high-bandwidth laser communication between spacecraft.