Google's First Orbital Computing Prototype Satellite Successfully Launched
Google's Project Suncatcher prototype satellite was successfully launched into orbit on October 1, aboard SpaceX's Transporter-18 rideshare mission, launched from Vandenberg Space Force Base in California by a Falcon 9 rocket, with the booster successfully landing back. Alphabet CEO Sundar Pichai confirmed the successful launch of this prototype satellite in collaboration with Planet and thanked SpaceX.
The satellite, modified from an existing platform by Planet Labs, is about the size of a refrigerator and contains four Google-developed tensor processors, marking the first time this generation of advanced chips has been sent into space. Travis Beals, head of Google Research, stated on the official blog that the team has established contact with the satellite, and its operation is as expected. In the coming weeks, data will be collected on the chip's performance under launch stress, radiation, and extreme temperatures.
This satellite is only for minimal validation. It is powered by solar energy at about 1 kilowatt, with the chips operating in 15-minute cycles followed by shutdowns for cooling to avoid overwhelming the power supply and thermal control. According to reports from U.S. public broadcasting, the satellite runs an open-source weight quantization model called Gemma, handling simple queries. Ground tests have included thermal vacuum, vibration, and proton beam radiation tests, but Beals noted that no ground test can replace real orbital conditions.
The roadmap is divided into two phases. Next year, two more satellites specifically modified for computing power are planned for launch to test inter-satellite connections and heavier payloads. TechCrunch cited the target configuration as 81 satellites in close formation, processing in parallel to meet the bandwidth and latency requirements of multi-rack loads. The long-term vision is for thousands of satellites to form an orbital cluster, powered by the sun to run models and send results back to Earth. A peer-reviewed paper has been published in Joule.
Pichai mentioned at the developer conference in May that the demand for AI services exceeds supply, with capital expenditures expected to be between $180 billion and $190 billion this year, over six times that of 2022. He had previously forecast in November 2025 that the Trillium generation chip would not be damaged by radiation simulated in a particle accelerator, with the milestone being the launch of two prototype satellites with Planet in early 2027. This time, the chips are integrated into Planet's existing satellites, advancing the experiment.
The buyers are cloud vendors lacking power and land, while the sellers provide capacity, satellite platforms, and thermal management in orbit. This is a research-driven task, not a commercial rollout. Funding extends from ground data center capital expenditures to launches and satellite integration, with SpaceX securing capacity and Planet receiving platform orders. Companies that can provide reusable rockets and ready-made satellites benefit, while those relying solely on ground power expansion for data centers face pressure. With 1 kilowatt and 15-minute cycles, commercially viable orbital computing power is still far off.
Source: Public Information
ABAB AI Insight
Google's venture into orbital computing does not start with chips. Quantum computing and autonomous driving are viewed by Pichai as part of the same category of moonshot projects: sending variables that cannot be closed on the ground into a real environment. The particle accelerator test in November 2025 only proved that the Trillium chip can withstand simulated radiation, while thermal management and orbital reliability were already listed as unresolved issues. The prototype satellite, the size of a refrigerator, moves these two unresolved issues from the lab to Transporter-18.
The capital path remains anchored on the ground. Pichai has set this year's capital expenditure range at $180 billion to $190 billion, primarily focused on ground clusters and self-developed chips. The orbital project utilizes Planet's existing platform and SpaceX's rideshare capacity, without being separately categorized as a new capital expenditure item. The motivation is to use a minimal flight to verify whether 1 kilowatt, 4 chips, and 15-minute cycles can survive, before deciding whether to invest further in the two dedicated satellites planned for 2027.
A comparable approach is SpaceX's Starlink, which first occupied orbital communications rather than selling computing power. The current position is one of validation, not expansion: 4 chips are insufficient for even one ground rack, and the configurations of 81 satellites and thousands in clusters remain theoretical. External solutions like Aetherflux are also discussing space solar power, but Google is betting on whether its self-developed chips can run its models in orbit, rather than on a new energy narrative.
Structurally, this represents a reconstruction of the industry chain. Ground data centers are constrained by electricity, water, and approvals, while orbital solutions replace electricity with solar power and thermal management with vacuum, simultaneously introducing new bottlenecks in launch frequency, inter-satellite lasers, and radiation lifespan. Pricing power will not shift simply because a prototype satellite is in orbit. Only when the cost of capacity drops to a point where it can be priced by clusters rather than experiments will the marginal production of computing power potentially move off the ground.
ABAB News · Cognitive Law
The ground lacks not chips, but electricity, land, and approvals.
The prototype satellite proves it is alive, but does not prove it can compute profitably.
Before new production sites emerge, old bottlenecks merely change addresses.