Google is testing whether solar-powered satellites carrying TPUs can process AI workloads in orbit.
A prototype launched on October 1, 2026, returned the first AI response from a space-based TPU on October 30.
Google plans a two-satellite test in 2027 to evaluate distributed AI computing and inter-satellite laser links.
Google is testing whether the next generation of artificial intelligence infrastructure could operate not on Earth, but in orbit. Under Project Suncatcher, the company is developing solar-powered satellites carrying its Tensor Processing Units (TPUs) that could one day work together as space-based data centres for AI workloads.
The idea remains experimental. But Google already sent a prototype satellite into orbit, run an AI prompt on a TPU in space, and planned a two-satellite test for 2027.
Project Suncatcher is a Google Research “moonshot” announced on November 4, 2025. Its goal is to explore whether networks of solar-powered satellites equipped with TPUs could harness sunlight in space to support large-scale machine learning.
Google’s long-term vision is not a single orbital server farm, but a constellation of interconnected satellites flying in coordinated formations. The company has described potential clusters of 81 satellites arranged across roughly a kilometre, each contributing computing power to a shared AI system.
The company sees space as a possible answer to the growing constraints of terrestrial AI infrastructure. Data centres require land, electricity, cooling and grid capacity, while demand for AI computing continues to rise. Google says satellites in low-Earth orbit could access near-constant sunlight and potentially generate up to eight times more solar power than on Earth.
Each satellite would carry TPUs—Google’s custom AI accelerator chips—and solar panels to generate electricity. In orbit, the chips would process AI tasks such as running models or handling inference workloads.
The satellites would communicate using free-space optical links, effectively laser connections between spacecraft. Google says these links are intended to provide the high bandwidth needed to distribute AI computations across multiple satellites.
The 2027 mission is designed to test that central concept: whether two satellites can divide an AI workload and coordinate their processing while in orbit. Google has said the test will examine whether AI computations can be distributed across multiple spacecraft.
Earth-based data centres are becoming larger and more power-hungry as companies train and deploy increasingly capable AI models. They also face limits around land availability, electricity supply and cooling, since servers generate substantial heat and often depend on water or energy-intensive cooling systems.
Space offers a different set of conditions. In low-Earth orbit, satellites can receive sunlight for much of their orbit, reducing reliance on terrestrial power grids. Google argues that this could make orbital computing attractive if launch costs and hardware reliability improve enough.
Still, Google has not claimed that space data centres are commercially viable today. Amanda Peterson Corio, Google’s global head of energy and power, has said space-based data centres could be up to 100 times more expensive than building equivalent capacity on Earth.
Also Read: TeraWulf Secures Google's Financial Support for Anthropic AI Infrastructure
Solar power is central to the project. Google says satellites in low-Earth orbit can access near-constant sunlight and may generate up to eight times more solar power than comparable installations on Earth.
That advantage matters because AI chips consume significant electricity. A larger orbital system would need enough solar generation not only to run TPUs, but also to support communications, satellite operations and thermal management.
The current prototype’s 1 kW capacity illustrates the gap between today’s experiment and a future orbital data centre. Google’s next satellites are expected to have higher power-generation capacity.
The project faces several major hurdles.
Launch costs: Deploying large numbers of satellites remains expensive. Google researchers have suggested that launch costs would need to fall to about $200 per kilogram for orbital computing to become broadly economical, a level it associates with the mid-2030s.
Radiation: Space radiation can disrupt electronics and corrupt data. Google is testing how its TPUs withstand radiation and the physical stresses of spaceflight.
Heat management: There is no air in space for conventional cooling. Satellites would need radiators and other systems to release waste heat into space.
Communications: Distributing AI work across satellites requires fast, reliable laser links. The 2027 mission will test whether this is practical.
Maintenance: Unlike terrestrial data centres, satellites cannot be easily repaired or upgraded once launched.
Orbital debris: Large constellations would add to congestion in low-Earth orbit, raising concerns about collisions and space debris.
Project Suncatcher fits into Google’s broader effort to secure computing capacity for future AI models. The company already relies heavily on TPUs across its cloud and AI systems, and Suncatcher extends that strategy by asking whether the same chips could operate beyond Earth.
Google has described the initiative as groundwork for “massively-scaled computation in space,” rather than a near-term replacement for its terrestrial data centres.
Google launched its first Suncatcher prototype on October 1, 2026, in partnership with satellite company Planet, using SpaceX’s rideshare service. The spacecraft entered sun-synchronous orbit and carries four TPUs.
On October 30, Google said the satellite had returned its first AI response from orbit, calling it the first response from a “data centre class TPU” in space. The company said the result clears the way for twin interconnected satellites in 2027.
Those two satellites will test optical links and distributed AI computing. If successful, they would still represent only an early research milestone—not an operational space data centre.
For now, Project Suncatcher is best understood as a long-term bet: Google is testing whether space can become a viable location for AI infrastructure, while acknowledging that cost, reliability and scale remain unresolved.
Also Read: Google’s Tamba Hints at DMs, Group Chats and Creator Communities on YouTube
Project Suncatcher is Google’s research initiative exploring solar-powered satellites equipped with TPUs that could eventually form interconnected orbital data centres for large-scale AI computing.
Satellites would carry TPUs, generate electricity from solar panels, and use laser links to share AI workloads across multiple spacecraft operating in coordinated low-Earth-orbit formations.
Google launched a prototype on October 1, 2026, and on October 30 said it returned its first AI response using an onboard TPU in space.
Google sees space as a potential source of abundant solar energy and an alternative to Earth-based data centres facing rising electricity, land and cooling constraints.
Major challenges include launch costs, radiation exposure, heat management, satellite maintenance, reliable laser communications, orbital debris and the high expense of deploying computing hardware in space.