Google Project Suncatcher advanced the notion of space-based AI infrastructure last week with the launch of a refrigerator-sized satellite carrying four AI chips into low Earth orbit on a SpaceX rocket. The early-stage experiment is designed to explore whether technology that powers energy-intensive data centers on Earth can operate reliably in space.
Google Project Suncatcher and orbital AI ambitions
The satellite, built with Planet Labs, lifted off on Oct. 1 and is functioning as expected, according to a company blog post. Google said the mission will assess how its custom Tensor Processing Unit chips withstand launch vibrations as well as the radiation and temperature extremes of space. The spacecraft is expected to run experiments in orbit for roughly a year, according to reports.
Powered by solar panels, the spacecraft relies on a network of heat pipes and radiators to manage thermal loads, since space offers no easy way to shed heat with water or fans. Given cooling limits, the chips will operate in 15-minute bursts before pausing to cool, according to reports.
Although preliminary, the launch marks a notable milestone for orbital computing, which advocates say could address soaring AI demand without drawing on scarce terrestrial resources such as power, water, and land. As companies invest heavily in new ground-based facilities and face pushback from communities and policymakers, shifting some AI workloads to orbit has become more compelling.
Interest in orbital data centers surged in early 2026 when SpaceX filed plans with U.S. regulators to deploy as many as 1 million data-center satellites. Elon Musk said in January that space would soon be the lowest-cost location to run AI.
Blue Origin, led by Jeff Bezos, followed in March with a proposal for a 51,600-satellite constellation called Project Sunrise.
Google began pursuing the concept even earlier. In November 2025 it unveiled Google Project Suncatcher to study how a network of solar-powered satellites, equipped with TPU chips, might harness continuous sunlight for AI computing.
The latest mission advances that work, though experts caution that large hurdles remain. Generating sufficient power would require massive solar arrays and heavy supporting hardware that are costly to lift to orbit. Despite steep reductions in launch costs, with estimates falling about 95% since 2008 to roughly $1,000 per kilogram according to industry research, the price tag for full-scale infrastructure remains formidable.
Radiation effects on computing hardware are another concern. Google said ground testing showed its chips can tolerate a total ionizing dose higher than expected over a five-year mission, yet it acknowledged that solar storms and cosmic rays still pose risks.
Maintenance and repairs present additional challenges, since servicing equipment in space is far more complex than on Earth.
Other companies are also experimenting. In November, startup Starcloud sent a satellite equipped with an Nvidia H100 chip into orbit and demonstrated it could run Google’s lightweight AI model Gemma.
Another entrant, Cowboy Space Corp., founded by Robinhood cofounder Baiju Bhatt, launched alongside Google last week. Its Reason-1 satellite plans to test beaming solar energy to Earth using a high-powered laser. Bhatt said the company has established contact with the spacecraft and aims to demonstrate its technology in the coming weeks. He added that Cowboy Space ultimately wants to build a vehicle to deliver data center payloads to orbit.
Google’s Oct. 1 launch is a starting point for broader ambitions. The company plans two additional satellites in 2027 to test laser communications between spacecraft, a task it likened to striking a coin-size target from miles away while both endpoints are moving.
Even so, a full-scale shift to orbital data centers may take time. James Manyika, Google’s senior vice president for research, said progress could mirror other long-horizon efforts, noting that some past initiatives took more than a decade before yielding visible results.
As companies push the limits of AI capability and infrastructure, from cloud platforms to orbital systems, industry leaders such as Nikesh Arora are similarly exploring how far advanced computing can be scaled securely on Earth.