Google Runs AI in Space for the First Time: Suncatcher Test
Google says it has run an AI model on a satellite in space and received the first response back on Earth. The test is part of Project Suncatcher, a research project that asks whether AI systems could eventually run in space instead of relying only on data centers on Earth.
Google announced the result on October 30. These details come from Google’s own announcement, and independent experts have not yet reviewed them.
What Happened
The satellite was launched on October 1 on a SpaceX rocket and has spent the past few weeks in testing.
In the experiment:
- Google sent a request from Earth asking the onboard Gemma 4 AI model to write a short poem about being in space
- The model ran on a Tensor Processing Unit (TPU), a chip built for AI work
- As the satellite passed over Svalbard, Norway, it sent its answer back to Earth

The Poem
The AI wrote this three-line poem:
Blue marble below
Data streams in silent grace
Starlight feeds my core.
Google called the exchange its “Hello World” milestone. That is a common name in computing for a first successful test.
A Claim of a First
Google says this was the first complete AI response generated using a data-center-class TPU in orbit. That is the company’s own claim, and it is a narrow one, because it concerns a specific kind of chip. We cannot check whether anyone else has done something similar.
Google also ran more tests afterwards. They included more poems and requests for information about the satellite’s operating conditions. The company says the results confirmed that the hardware can process requests in space and send answers back.
Our Earlier Report
Earlier this month, we reported that Google had launched this satellite with four TPUs, and that it planned to run a version of Gemma for about 15 minutes at a time because of heat limits. Today’s report does not repeat those details. It says the test was a success, but it does not say how long the model ran or how many chips were used.
What This Test Proves, and What It Does Not
What it shows: the chip can run a model in orbit, take a request from Earth, and send the result back.
What it does not show: that space data centers are practical. The test was a short poem. The report does not say how fast it ran, how much power it used, or how reliable it was over time.
Why Google Is Interested in Space Computing
According to the report:
- AI data centers use large amounts of electricity for processors and cooling
- Project Suncatcher asks whether solar-powered satellites could handle some of this work
- The idea is to use sunlight in space to power AI systems and rely less on electricity made on Earth
The Challenges
Operating computers in space is hard. Satellites face extreme temperature changes, radiation, and other conditions that can damage electronics. The current satellite is built to help researchers learn whether the AI chips work reliably in those conditions.
What Comes Next
- Longer tests. Google will test how the chips perform over longer periods, with radiation and changing temperatures.
- 2027. Google plans to test two interconnected satellites. This will show how several satellites could work together on AI tasks, which is needed for larger systems.
- The company is also studying the practical problems of connecting, powering, and managing hardware in space.
No Commercial Service Yet
Google has not announced plans to offer commercial AI computing from space. The report says Project Suncatcher is still an early research effort and not a working space data center.
What We Do Not Know Yet
- How many chips were running and for how long
- How fast the model answered, and how much power it used
- Whether the satellite is still working normally
- Costs. Our earlier report said Google believed space data centers were unlikely to be cheaper than ground ones within five years, but this report does not repeat that.
