Google has placed its first AI computing satellite into orbit. On October 1, 2026, a SpaceX Falcon 9 rocket carried the prototype — named MVP — into space aboard the Transporter-18 rideshare mission. Google said its team had confirmed contact with the spacecraft and that it was operating as expected.
The satellite carries four Tensor Processing Units (TPUs), the chips Google designed for machine learning and already runs in data centers on Earth. Built in partnership with Planet, which will operate it, the prototype is powered by solar panels providing one kilowatt of electricity.
Those chips will run Google's open-weight Gemma AI model, answering simple queries in 15-minute bursts before shutting down to cool. The limit is imposed by heat management — the single biggest engineering hurdle for computing in orbit.
On Earth, chips are cooled by air or water. In the vacuum of space that is impossible; waste heat can only be moved away through heat pipes and radiators. Google has called cooling "a crucial research challenge," and project lead Travis Beals said the radiators are among the heaviest components on the current mission — a drawback, since heavier objects cost more to launch.
The appeal of orbit is energy. Google calculates that solar panels in a sun-synchronous orbit, where the satellite is almost never in shade, can generate up to eight times more electricity than comparable installations on Earth. That orbit also removes the need for heavy backup batteries.
The mission is planned to last about a year. Google says it will gather data on how the TPUs handle the physical stress of spaceflight, radiation, and thermal extremes. The company has also published the scientific groundwork in a peer-reviewed paper in the journal Joule.
Google's longer-term vision is a constellation of satellites, each carrying dozens of TPUs, communicating with one another via high-bandwidth lasers. The company says it will test those laser links in 2027, when it plans to put two more satellites into orbit.
Google is not alone in pursuing the idea. SpaceX has said it expects to start deploying "orbital AI compute satellites" as early as 2028, while COO Gwynne Shotwell said in September that the company would deploy "supercompute in space" in 2027. The venture-backed startup Starcloud, which launched a spacecraft carrying an Nvidia H100 chip in November 2025 and demonstrated a version of Google's Gemini from space, has announced it secured $170 million in Series A funding at a valuation of roughly $1.1 billion.
Whether orbital data centers ever become economical is unresolved. Beals told NPR he does not see the approach being cheaper within the next five years, and noted it is difficult to know when it will be viable. Google's own analysis suggests launch prices near $200 per kilogram by the mid-2030s would be needed for orbit to compete with ground-based facilities — a steep reduction that depends on rockets that do not yet fly at the required scale.
What is confirmed: the prototype satellite is in orbit and operating; it carries four TPUs running Gemma in 15-minute intervals on one kilowatt of solar power; Google's peer-reviewed paper appears in Joule.
What remains unknown: how the chips perform across the planned year-long mission, whether satellite-to-satellite laser links can maintain the required bandwidth, and whether launch costs fall far enough to make orbital data centers competitive.
What is awaited: in-orbit test results over the coming weeks, the two-satellite laser demonstration planned for 2027, and further detail on SpaceX's own deployment timeline.