Next Thursday, Google will send an experimental satellite into orbit. The satellite has enough computing power to answer simple artificial intelligence queries from space.

Last month, inside a San Francisco laboratory, a team of technicians wearing protective suits and hairnets ran a series of tests on a refrigerator-sized satellite commissioned by Google, repeatedly adjusting and inspecting it. They first checked the satellite's solar panels, which will unfurl in space and point toward the sun. The satellite was then placed on a test rig and subjected to violent vibration testing to verify whether the internal chips would be damaged and whether the whole unit would come apart during launch. Technicians carefully drew thin lines above each screw so they could tell whether any screw had loosened during the vibration test. The satellite passed the vibration test smoothly, with all screws remaining tight and the chips intact.

James Manyika, Google's senior vice president and head of research, called the test results "very ideal," but said he still has questions about how the satellite will actually perform once in orbit.

The work is part of an ambitious Google project called Project Suncatcher. The project aims to deploy artificial intelligence data centers in space, using solar power to run them. On October 1, the satellite built in the San Francisco laboratory, which Google has named MVP, will become the tech giant's first step. The satellite will launch into orbit from Vandenberg Space Force Base near Santa Barbara, California, aboard a SpaceX Falcon 9 rocket.

Just a year ago, the concept still seemed like a science fiction dream. Elon Musk, Jeff Bezos, Sam Altman and others have all voiced support for the idea of orbital data centers. But the vision faces numerous challenges, including shielding against space radiation and the high cost of getting equipment out of the atmosphere. As ground-based data centers run into public opposition and physical constraints, the appeal of placing computing facilities above the Earth is steadily rising.

It should be noted that Google is not launching a complete data center this time, only a preliminary experimental prototype. The MVP carries four specialized chips called tensor processing units (TPUs), with computing power equivalent to a single server in a ground-based data center. The satellite's solar panels can supply only about 1 kilowatt of power to the chips, roughly the same as a hair dryer. But that power is enough to help Google test the hardware under the harsh conditions of space. The satellite can handle simple AI queries and is designed for a one-year operational life, though it could orbit the Earth for up to six years. Eventually, it will re-enter the atmosphere under Earth's gravity and burn up.

Manyika said Google's expectations for the satellite are very cautious. "Frankly, we don't expect to have a system that can be reliably put into practical use within the next few years," he said. He compared the mission to Google's early work on self-driving cars. "Don't forget, Google researched that for about 15 years before seeing results on the ground. I think this project will follow a similar pace."

Brandon Lucia, a professor of electrical and computer engineering at Carnegie Mellon University, said scaling from a single satellite to a massive constellation and building a giant distributed data center will take a long time and enormous sums of money. "Once you get into large-scale deployment, more engineering problems will emerge," he said. "This is an area no one has set foot in."

Project Suncatcher was proposed by Blaise Agüera y Arcas, a 51-year-old Google vice president and AI researcher who leads a team researching intelligent technologies. About three years ago, he attended a gathering of entrepreneurs and AI researchers whose central topic was the steadily rising energy consumption of AI systems. After the meeting, Agüera y Arcas concluded that data centers must move into space to make full use of solar power. "I've been thinking about this idea since I was a child," he said. "Science fiction has long contained the notion of using stellar energy for computing."

Agüera y Arcas quickly reported the idea to Manyika. Manyika was initially skeptical but agreed to run experiments to verify whether AI chips could be cooled in space and withstand space radiation. So in February 2025, Google sent AI chips to the Crocker Nuclear Laboratory in Davis, California. A large particle accelerator there, a cyclotron whose magnets were once used in a similar device for the Manhattan Project, irradiated the chips with a radiation dose equal to what they would absorb over five years of exposure in space. Space radiation can cause bit flips, failures that change the physical state of a circuit and turn a 0 in binary code into a 1, or a 1 into a 0, easily triggering serious system failures. But the results of Google's cyclotron test were encouraging: in most cases, restarting the chip fixed the bit-flip problem.

In May 2025, Manyika organized a meeting, with Agüera y Arcas attending, to report to Google CEO Sundar Pichai and seek approval to continue the project. To his surprise, Google co-founder Sergey Brin also attended. Brin and Pichai quickly approved the project. Agüera y Arcas relayed Brin's words: "Okay, this idea works. Let's discuss how to implement it."

After receiving top-level approval, Project Suncatcher accelerated. Google has not disclosed how much it is investing in the project, but it expects that as satellite launch costs fall, the cost of orbital data centers will roughly match that of ground-based data centers by the mid-2030s. Google soon signed a contract with Planet Labs, a satellite imaging company it had previously invested in, commissioning it to build satellites carrying AI chips. James Mason, Planet Labs' chief space officer, said he had discussed the idea of space computing with Brin for years, but the idea had remained vague at the time. "Back then, this was still very far off," Mason said. "At that time, large models had not yet risen, and demand for AI computing power had not yet entered a phase of exponential growth."

Planet Labs originally agreed to launch two satellites for Google in 2027. But the internet giant wanted to reach orbit this year and was willing to take on risk to compress the schedule. Eric Stevens, Planet Labs' director of systems engineering, said that to speed things up, Google directly installed its own chips into Planet Labs' existing satellite platform for testing. The biggest technical difficulty lies in dissipating heat from the AI chips: the chips generate a great deal of heat when processing information. Fans are commonly used for cooling on the ground, but they cannot work in the vacuum of space. The Silicon Valley company therefore developed its own cooling solution, using multiple layers of thermally conductive materials to radiate heat into space. At the bottom is Google's AI chip, mounted on a green circuit board; above it is a thermal interface material, a pale green paste-like sheet resembling a fruit roll-up, connecting the chip to aluminum and copper heat-spreading layers and drawing heat away from the board; the outermost layer is a radiative cooling panel that emits heat into space.

Travis Beals, senior director of product management for Project Suncatcher, said the chip can run continuously in space for about 15 minutes at most before it must shut down to cool. During that time, the chip can handle brief queries and support Google's Gemini large model in producing responses.

Google has already begun planning later phases. It plans to launch a pair of satellites next year and has also designed a constellation of more than 80 satellites flying in close formation, communicating with one another and jointly handling AI queries. Google is also in talks with designers about possibly building a custom satellite the length of a football field.

Beals said: "If, five years from now, everything we've done lands perfectly, that would only show we didn't take on enough risk and didn't learn much. In the long run, if the project truly succeeds, this will eventually become utterly ordinary, and people won't think that a query they send to Gemini might possibly be computed in space."