Google brings orbital AI testing online with first TPU satellite

Traders Union
All News
Business & Wealth
Companies
Google orbital AI satellite test

Highlights

Google Suncatcher’s prototype satellite successfully began AI testing in orbit, operating a DeepMind Gemma 4 model after a three-week commissioning process.
The satellite’s single TPU can run for only 20 minutes at a time due to 1 kilowatt power constraints, complicating real-time space-to-ground inference workload tests.
If current tests succeed, Google plans two satellites with advanced chips and optical links in 2027, supporting its Starcloud distributed space computing strategy amid ongoing launch cost pressures.

Orbital test campaign begins

As reported by TechCrunch, Google Suncatcher’s prototype satellite is now communicating with the ground after a roughly three-week commissioning process that checked its onboard subsystems. Engineers this week sent the first prompt to the craft’s pre-loaded Google DeepMind Gemma 4 model, which responded with a haiku about being in space.

The satellite generates up to 1 kilowatt of electricity and carries one Google TPU, but the chip’s power demands mean it can operate for only about 20 minutes at a time. That constraint adds complexity to space-to-ground communications as Google tests whether larger inference workloads can eventually be handled by groups of satellites flying in close formation.

Google says the spacecraft is operating nominally and is also being used to report on orbital conditions around the satellite. The ongoing campaign is examining how the processor performs under thermal stress and in the space radiation environment, where bit flips can corrupt stored data without triggering standard monitoring tools.

Next steps and commercial implications

Ground testing previously suggested radiation would have limited effects on inference tasks but could pose a bigger challenge for AI training workloads. The current mission is intended to provide real-world data on those effects in orbit rather than relying on simulation or particle-accelerator testing on Earth.

If the onboard TPU proves capable, Google plans to launch two more satellites with more advanced chips in 2027. Those spacecraft are expected to support higher power generation and inter-satellite optical links, a key part of the company’s Starcloud vision for distributed computing in space.

Google’s approach stands out because it aims to connect satellites with high-speed links so they can support coordinated workloads, while other companies including Starcloud and Cowboy Space are also pursuing inference in orbit on individual satellites. The long-term economics remain tied to launch costs and vehicle reusability, and that creates a close link to SpaceX and its still-developing Starship system as Alphabet weighs longer-range infrastructure bets against nearer-term AI pressure on its core advertising business.

Our earlier article on the U.S. AI data center boom explained that many planned projects are running into credibility questions as costs rise, construction timelines slip, and power availability tightens. It also highlighted how higher financing costs and investor skepticism can derail buildouts, while grid constraints and local opposition add another layer of friction for expanding compute capacity.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.