Google plans orbital AI chip test as data centers face pushback

Google plans to send its first AI chips into orbit during the week of September 28, testing how its Trillium TPUs handle conditions in space. In a September 24 announcement, the company said the chips are scheduled to fly on SpaceX’s Transporter-18 mission with satellite firm Planet.

The flight, covered in The Rundown, is an early step in Project Suncatcher, Google’s effort to explore data centers powered by nearly constant sunlight. If the idea works at scale, it could give AI companies a way to expand computing while easing pressure on communities and local grids.

What Google is testing

Google says the test is designed to collect data on radiation and heat, following tests on the ground for launch stresses such as vibration. A separate experiment planned for 2027 would test laser links between two satellites. Future designs would carry dozens of TPUs on each spacecraft.

In its November 2025 research, Google proposes an orbit with nearly continuous sunlight and estimates that solar panels there could produce up to eight times the energy of panels on Earth. Cooling poses a separate challenge. A vacuum provides no airflow to carry heat away, so Google’s design relies on heat pipes and radiators.

TechCrunch reported in March on orbital computing proposals from SpaceX, Blue Origin, and Starcloud, including SpaceX’s plan for one million satellites.

Why it matters

Data center projects face resistance over noise, power demand, and electricity costs. In Lowell, Massachusetts, residents complained about cooling equipment noise and generator fumes. The city council voted unanimously in February 2026 to halt further data center expansion for a year, AP reported.

Power costs add to the concern. Axios reported in July that a market monitor estimated data centers accounted for $6.3 billion of $16.4 billion in PJM capacity auction costs. Researcher Jonathan Koomey cautioned in AP’s reporting that demand pressures differ by location.

For communities facing those pressures, orbital computing could offer substantial relief if it replaces capacity that would otherwise require new facilities on Earth. That could mean fewer disputes over where to build, less equipment noise, and less added demand on local grids. If companies add computing in orbit while continuing the same expansion on the ground, the benefit would be smaller. Any reduction in household bills remains uncertain.

The planned tests address hardware behavior and links between satellites. Data center operators would also need reliable service over years, fast connections to the ground, and costs that can compete with facilities on Earth.

Google’s cost model suggests launch prices could fall below $200 per kilogram by the mid-2030s. At that level, the company estimates that launch and operating costs could approach the energy costs of equivalent computing on Earth. That limited comparison leaves the full economics of a commercial system unresolved.

In a February interview with The Indian Express, months before Google’s September announcement, OpenAI CEO Sam Altman questioned whether space data centers could reach meaningful scale this decade. He cited launch economics and the difficulty of fixing failed GPUs in space.