A refrigerator-sized satellite carrying Google's Tensor Processing Units is now circling the Earth, after a SpaceX rocket delivered the Project Suncatcher prototype to orbit — the search giant's first tangible step toward a goal that once sounded like science fiction: data centers in space, powered entirely by sunlight.
The prototype launched as part of SpaceX's Transporter-18 rideshare mission, which carried Google's satellite alongside more than a hundred other payloads. It was developed in partnership with Planet, the satellite imaging company, whose team will commission the spacecraft before Google fires up its chips. On board are four TPUs — the custom machine-learning accelerators that already do heavy lifting in Google's terrestrial data centers — which will run a version of Google's open-weight Gemma model in short bursts, answering simple queries while engineers monitor how the silicon copes with orbit. NPR reported the details of the mission and its goals. It is a milestone for one of the more audacious infrastructure bets in AI, one we have tracked since Google first put its space data center research on the record — more on that story and the rest of the AI industry at aibuzzwire.news.
A Minimal Test With a Big Backdrop
Google's own team is quick to lower expectations. Travis Beals, the senior director who leads Project Suncatcher, called the satellite "a very minimal test" whose job is simply to confirm the chips can run in space. In a blog post, Google said the mission will gauge how the TPUs handle "the physical stress of spaceflight and the radiation and thermal extremes of space."
The satellite is headed for a sun-synchronous orbit, where its solar panels will almost never pass through shadow. That is the entire appeal of the orbit: near-continuous sunlight eliminates the need for heavy batteries and backup power, which are among the costliest mass on any spacecraft. The goal is for the mission to remain operational for a year.
Why Google Thinks Space Is the Answer
The logic starts with energy. "The sun puts out almost all of the power in our solar system. All of the other power sources that humanity has tapped into are just a tiny fraction of a percent," Beals said. Project Suncatcher is, in his framing, an attempt to tap the best possible solar power for AI compute.
The timing is no accident. Demand for AI services has outrun the supply of power and chips on Earth. Google CEO Sundar Pichai said at the company's developer summit in May that he expects capital expenditures of $180 billion to $190 billion this year — more than six times the 2022 figure — much of it going into computing infrastructure. At the same time, opposition to power-hungry data centers on the ground is hardening in community after community. Orbit offers virtually unlimited solar energy and no zoning protests.
The Heat Problem — and Every Other Problem
The first test flight also illustrates why this remains hard. Heat is the immediate constraint: the TPUs on board will run the Gemma model for only about 15 minutes at a time because there is no wind or water in a vacuum to carry heat away. Google calls cooling "a crucial research challenge" and is working on pipe-and-radiator systems to wick heat off the chips — and Beals said the radiators are among the heaviest components on the current mission, which is a problem, because mass is what makes launches expensive.
Maintenance is the next wall. Carnegie Mellon professor Brandon Lucia, who studies computing in space, noted that a technician can walk into a terrestrial data center when something breaks — an option that disappears a few hundred miles up. "Those all have their cost and complexity amplified by a factor of 10, maybe a factor of 100," he said. "And so there has to be a big payoff."
Even communication is harder in space. Google eventually envisions clusters of satellites, each carrying dozens of TPUs and linked to one another and to the ground by laser. Next year the company plans to launch two more satellites specifically to test those optical connections, which must punch through the atmosphere and over hundreds of miles without the reliability of fiber.
The Race to Put Compute in Orbit
Google is not alone. Last November, the startup Starcloud launched a spacecraft carrying an Nvidia H100 chip and demonstrated a version of Google's Gemini model running from orbit. SpaceX, which is conveniently also the launch provider everyone else depends on, has said it expects to begin deploying "orbital AI compute satellites" as early as 2028, and Elon Musk has argued space is the "only way to scale" data centers for future AI demand.
Environmental questions linger as well. The satellites may run on sunlight, but the rockets that carry them burn carbon-based fuel, and defunct spacecraft eventually re-enter the atmosphere and burn up, with pollution effects researchers are still working to quantify.
Economics: The Open Question
Beals is candid that technical feasibility is not the same as a business. "It doesn't help a lot if this is technically possible, if it's always going to be too expensive to be practical," he said. His own estimate: "I don't see this being something where it's cheaper to do this in the next five years. I think it will take longer than that."
For now, the milestone is narrower but real: Google's machine-learning chips are running its own model in orbit, and the company has a year of data collection ahead to decide whether the dream of sun-powered AI factories in space survives contact with economics.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
