Project Suncatcher by Google put its first prototype satellite into orbit on Thursday. The spacecraft lifted off from Vandenberg Space Force Base in California. It flew aboard SpaceX’s Transporter-18 rideshare mission. Google built the craft with Planet, an aerospace and satellite imaging firm. Soon after launch, the company confirmed contact with the satellite. This is the first time Google has flown its AI chips in space.
What the satellite carries
Four Trillium Tensor Processing Units sit inside the craft. These are the Google TPUs the company already runs in data centers on the ground. In orbit, the chips meet launch forces, radiation and wide temperature swings. The test measures how well they hold up. Cooling is the other question, and it gets harder in a vacuum where no air moves heat away.
Heat sets a hard limit. Each chip can run a version of Google’s Gemma model for about 15 minutes at a time. After a run, the satellite cools before the next one. Over the coming weeks, engineers will track performance and wear.
Why Project Suncatcher by Google looks to space
Power is the draw. Google says a solar panel in the right orbit can produce up to eight times the energy of the same panel on Earth. The satellite flies in a sun-synchronous orbit, where sunlight stays near constant. That path removes the need for heavy onboard batteries. The prototype pulls about one kilowatt from its solar panels. Demand for AI compute keeps climbing, and that raises power use on the ground. Google frames the project as one answer to that pressure.
Testing the chips on the ground
Before the flight, Google ran hardware tests on Earth. The team exposed a Trillium chip to a proton beam to mimic the radiation of low Earth orbit. It showed no critical damage, even at doses above what the orbit would deliver. That result moved the project toward a real test in space.
Project Suncatcher points toward much larger systems. Google envisions clusters of solar-powered satellites, each carrying dozens of chips. The units would connect through lasers, forming space-based AI infrastructure that behaves like a data center in orbit. One published design describes a cluster of 81 satellites flying within hundreds of meters of each other. The aim is to shift some AI workloads off the planet and build AI data centers in space, run by the Sun rather than the grid.
Travis Beals, senior director and lead of Project Suncatcher by Google, described the prototype as a minimal first test of the hardware.
Google is not the only player. A startup called Starcloud flew a spacecraft last November with an Nvidia chip and ran a version of Google’s Gemini model from orbit. SpaceX has said it plans to deploy orbital compute satellites as early as 2028. More firms now view space as a way around the power limits of data centers on the ground.
What comes next
The effort stays early. Project Suncatcher by Google started in November 2025 as a long-term research program. Full orbital systems remain years out, if they arrive at all. Cost is one barrier. Launch prices and heat control both need to improve before the model scales. For now, a single satellite will run its tests in orbit and send the results back to Earth.
What is Project Suncatcher by Google?
Project Suncatcher by Google is a long-term research program that studies whether AI computing can run in space. Google announced it in November 2025. The idea is to fly satellites fitted with the company’s AI chips and power them with solar energy. Google calls it a moonshot, meaning a far-reaching bet that may take years to pay off. The first step is the prototype satellite that reached orbit this week. It carries four Trillium Tensor Processing Units, the same type of chip Google uses in its ground data centers. The satellite tests how the chips handle launch, radiation, and heat. Google built it with Planet, a satellite firm. Over time, the company hopes to link many such satellites into a single system. For now, the project remains an experiment. It is not a working data center, and consumer products will not change because of it soon.
What does the first Suncatcher satellite do in orbit?
The first satellite runs a set of basic tests. Its main job is to show whether Google TPUs can work in orbit at all. The chips face strong forces during launch, then radiation and large temperature swings in space. Engineers will watch how the hardware holds up over several weeks. Cooling is a key test. In a vacuum, there is no air to carry heat away, so the satellite must move heat to radiators and release it. Because of that limit, each chip runs a version of Google’s Gemma model for about 15 minutes at a time. The satellite then cools before the next run. It draws roughly one kilowatt of power from its solar panels. The craft also checks whether the chips can answer simple AI queries from orbit. The results will guide whether Google moves ahead with larger satellites later on.
Why does Google want to run AI in space?
Google points to power as the main reason. AI systems need large amounts of energy, and demand keeps rising. That strain falls on power grids, land, and water on the ground. Space offers a different source. Google says a solar panel placed in the right orbit can make up to eight times the energy of the same panel on Earth. A satellite in a sun-synchronous orbit sees near-constant sunlight, so it does not need heavy batteries. In theory, that makes solar-powered satellites a steady way to run AI work. The company wants to learn whether moving some compute off the planet can ease pressure on earthbound data centers. It also sees room to grow, since space has more space than crowded sites on the ground. The test does not prove the model works. It only checks the first question, which is survival of the hardware.
When could space-based AI data centers become real?
No firm date exists yet. Project Suncatcher by Google is early-stage research, and large orbital systems remain years away. Google has shared a design for clusters of satellites that link through lasers and act like a data center in orbit. One plan describes 81 satellites flying close together. Reaching that scale depends on two things. First, launch costs must fall, since sending mass to orbit is still expensive. Second, heat control must improve, because dense AI chips in a vacuum need large radiators to shed warmth. Other firms are chasing the same goal. SpaceX has said it may deploy orbital compute satellites as early as 2028, and a startup called Starcloud has already flown a test chip. Whether any of these reach full scale is not yet clear. For now, the field is made up of small tests, not working AI data centers in space.
