Google to launch experimental AI satellite MVP next week for space data center project

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

Rigorous Testing in San Francisco

Last month, a team of technicians wearing protective suits and hairnets conducted extensive testing on the refrigerator-sized satellite, built to Google's specifications, at a San Francisco laboratory.

The testing process included:

• Solar panel inspection: Engineers checked the panels that will deploy and align with the sun in space
• Vibration testing: The satellite was mounted on a test stand and subjected to intense shaking to verify that internal chips would not be damaged and the structure would remain intact during launch
• Screw monitoring: Technicians drew fine alignment marks above each screw to detect any loosening during vibration

The satellite passed vibration testing successfully, with all screws remaining tight and chips intact. James Manyika, Google's Senior Vice President and Head of Research, described the results as "very ideal," though he expressed lingering uncertainty about the satellite's actual performance once in orbit.

Project Suncatcher: Google's Orbital Ambition

This effort is part of Project Suncatcher, an ambitious Google initiative to deploy AI data centers in space, powered by solar energy. On October 1, the satellite — named MVP by Google and built in the San Francisco lab — will become the tech giant's first step toward that vision. It will launch aboard a SpaceX Falcon 9 rocket from Vandenberg Space Force Base near Santa Barbara, California.

Just one year ago, this concept seemed like science fiction.

Elon Musk, Jeff Bezos, Sam Altman, and others have expressed support for the orbital data center concept. However, the idea faces significant challenges, including protection against space radiation and the high cost of launching equipment beyond Earth's atmosphere. As ground-based data centers encounter public opposition and physical limitations, the appeal of deploying computing infrastructure above Earth continues to grow.

A Prototype, Not a Full Data Center

It is important to note that Google is not launching a complete data center — only an experimental prototype. The MVP carries four specialized chips called Tensor Processing Units (TPUs), providing computing power equivalent to a single server in a ground-based data center. The satellite's solar panels can deliver only about 1 kilowatt of power to the chips, roughly equivalent to the power consumption of a hair dryer.

However, this power is sufficient for Google to test hardware performance in the harsh space environment. The satellite can handle simple AI queries and has a design life of one year, though it could remain in orbit for up to six years. Eventually, Earth's gravity will pull it back into the atmosphere, where it will burn up upon reentry.

Cautious Expectations

Manyika said Google maintains cautious expectations for the satellite.

"Frankly, we don't expect to have a stable, practical system within the next few years," he said, comparing the mission to Google's early autonomous vehicle development. "Remember, Google worked on that for about 15 years before seeing results. I think this project will follow a similar timeline."

Brandon Lucia, Professor of Electrical and Computer Engineering at Carnegie Mellon University, noted that expanding from a single satellite to a massive satellite cluster — building a giant distributed data center — would require extensive time and substantial funding.

"Once you enter large-scale deployment, even more engineering challenges emerge," he said. "This is uncharted territory."

Origins of the Idea

Project Suncatcher was conceived by Bryce Aguerra y Arcas, a 51-year-old Google Vice President and AI researcher who leads a team focused on intelligent technology research. Approximately three years ago, he attended a gathering of entrepreneurs and AI researchers where the central topic was the rising energy consumption of AI systems. After the meeting, Aguerra y Arcas concluded that data centers must move to space to fully harness solar energy.

"I've been thinking about this concept since I was a child," he said. "Science fiction has long featured the idea of using stellar energy for computing."

Aguerra y Arcas quickly presented the idea to Manyika, who was initially skeptical but agreed to conduct experiments to determine whether AI chips could be cooled and protected from space radiation.

Radiation Testing at a Nuclear Lab

In February 2025, Google sent AI chips to the Crocker Nuclear Laboratory at the University of California, Davis. A large particle accelerator — a cyclotron whose magnets were used in equipment similar to that of the Manhattan Project — irradiated the chips with a radiation dose equivalent to five years of exposure in space.

Space radiation can cause bit flips, faults that alter the physical state of circuits, turning binary code 0s into 1s or 1s into 0s — potentially triggering serious system failures. However, the cyclotron test results were encouraging: in most cases, restarting the chip could repair bit flip issues.

Green Light from Leadership

In May 2025, Manyika organized a meeting where Aguerra y Arcas presented the project to Google CEO Sundar Pichai, requesting approval to proceed. To his surprise, Google co-founder Sergey Brin also attended.

Brin and Pichai quickly approved the project. Aguerra y Arcas recalled Brin's words: "Okay, this idea is feasible. Let's discuss how to implement it."

With senior leadership approval, Project Suncatcher accelerated. Google has not disclosed the project's investment amount, but estimates that as satellite launch costs decline, orbital data center costs could roughly match ground-based data centers by the mid-2030s.

Building the Satellite

Google quickly contracted with Planet Labs, a satellite imaging company in which it had previously invested, to build the satellite carrying the AI chips. James Mason, Planet Labs' Chief Space Officer, said he had discussed space computing concepts with Brin over the years, but the ideas had always remained vague.

"Back then, it was still very far off," Mason said. "Large models hadn't emerged yet, and AI computing demand hadn't entered its exponential growth phase."

Planet Labs had originally agreed to launch two satellites for Google in 2027. But the internet giant wanted to reach orbit this year and was willing to accept the risks of an accelerated timeline. To speed things up, Google directly installed its own chips into Planet Labs' existing satellite platform for testing, according to Eric Stevens, Planet Labs' Director of Systems Engineering.

Solving the Cooling Challenge

The biggest technical challenge was cooling the AI chips, which generate significant heat during computation. Ground-based systems typically use fans for cooling, but fans cannot operate in the vacuum of space. The Silicon Valley company developed a custom cooling solution using multiple layers of thermally conductive materials to dissipate heat into space.

The thermal management stack includes:

• Bottom layer: Google's AI chip, mounted on a green circuit board
• Thermal interface material: A light green paste-like sheet, similar in form to a fruit roll-up, connecting the chip to aluminum and copper散热 layers
• Radiator panel: The outermost layer, which radiates heat into space

Travis Bills, Senior Director of Product Management for Project Suncatcher, explained that the chip can operate continuously in space for about 15 minutes before it must shut down to cool. During that window, it can process brief queries and support responses from Google's Gemini large language model.

Future Plans

Google has already begun planning subsequent phases. The company aims to launch a pair of satellites next year and has designed a cluster of more than 80 satellites that would fly in close formation, communicating with one another to collaboratively process AI queries. Google is also in discussions with designers about building a custom satellite the length of a football field.

"If, five years from now, everything we've done has gone perfectly, that would only mean we didn't take enough risks and didn't learn enough," Bills said. "In the long run, if the project is truly successful, this will eventually become ordinary — people won't think twice about the possibility that their query to Gemini might have been computed in space."