Google’s first Project Suncatcher prototype satellite reached orbit after launching on SpaceX’s Transporter-18 rideshare mission on October 1, 2026. In its launch post, Google says the satellite was built in partnership with Planet and that “our team has confirmed contact with the satellite and it is operating as expected.”
Four weeks later the TPUs were doing AI work. On October 30, 2026, Google reported that the satellite had finished its commissioning checklist, including powering up the host and TPU trays and running early machine-learning tests, and had returned “the first-ever AI response from a data center class TPU in orbit”: a haiku written by the onboard Gemma 4 open model and downlinked during a pass over Svalbard, Norway. Radiation and thermal tests in orbit come next; a pair of interconnected satellites is planned for 2027.
Last checked November 16, 2026 against Google’s and Planet’s own posts. Google has published no Suncatcher update after October 30.
Project Suncatcher is Google’s research effort to find out whether machine-learning compute can run on satellites powered by sunlight. Google calls it “the first step in a long-term research moonshot exploring whether space could one day host scalable machine learning infrastructure.” The prototype carries Google’s Tensor Processing Units (TPUs), the AI accelerators Google designs for its own data centers.
On this page
Project Suncatcher timeline
| Date | What happened | Source |
|---|---|---|
| November 4, 2025 | Google announces Project Suncatcher with a preprint on constellation design and TPU radiation tests; Planet says it will build and operate two prototype satellites, targeting launch by early 2027 | Google, Planet |
| September 24, 2026 | Google publishes its pre-launch test results and a video series on the mission | |
| October 1, 2026 | First prototype launches on Transporter-18; contact confirmed; the peer-reviewed paper appears in Joule | |
| Week of October 30, 2026 | Commissioning complete; first AI response from orbit (a Gemma 4 haiku), then limericks and telemetry prompts on payload status | |
| 2027 (planned) | Two interconnected satellites to test laser links between spacecraft |
What the satellite has done in orbit
Google’s October 30 post, signed by Travis Beals, Senior Director of Paradigms of Intelligence, describes the first test as a “Hello World” milestone. During a ground-station pass the team uplinked a prompt asking Gemma 4 for a haiku about being in space, and the poem came back on the next downlink. Google then ran space-themed limericks and prompts that asked the model to report the technical status of the payload systems.
That result shows the TPU hardware survived launch and can run an inference workload in orbit. It is not yet the radiation and thermal data the mission was launched to collect: Google says those tests come next, and that showing its TPUs “can reliably execute workloads in space” is what clears the way for the 2027 pair of satellites. Google has not published error rates, temperatures or throughput from orbit.
What Google tested on the ground
Google laid out what it had already checked before launch in a September 24, 2026 explainer:
| Question | What Google reported before launch |
|---|---|
| Radiation | Trillium TPUs were run under AI workloads in a proton beam at UC Davis’s Crocker Nuclear Laboratory. “Our Trillium TPUs hold up remarkably well, and can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission.” |
| Launch loads | The ride to low Earth orbit lasts about 10 minutes, with sustained loads up to 10 g on the spacecraft and 50 to 100 g on individual components; the satellite held up when shaken on all three axes |
| Cooling | No airflow in a vacuum, so heat leaves only through radiators; a combination of heat pipes and radiators was tested in a thermal vacuum chamber, with refinement ongoing |
| Power | In low Earth orbit, satellites can generate “up to eight times more solar power than on Earth” |
The October launch post adds that the underlying research is now a peer-reviewed paper in the journal Joule.
Where the design is meant to go
The single prototype is far from the system Google has sketched. In the November 2025 research post behind the preprint, and in the September 2026 explainer, Google describes the end goal with these figures:
| Element | What Google has published |
|---|---|
| Satellites | Future satellites would each carry dozens of TPU chips and fly in tight clusters |
| Example cluster | An illustrative 81-satellite constellation at a mean altitude of 650 km, cluster radius 1 km, with neighboring satellites about 100 to 200 m apart |
| Links between satellites | Lasers at very high bandwidth over very short distances; a bench-scale demonstrator reached 800 Gbps each way (1.6 Tbps total) with one transceiver pair |
| Radiation test | Trillium, Google’s v6e Cloud TPU, tested in a 67 MeV proton beam |
| Launch cost | Google’s analysis projects launch prices could fall below $200/kg by the mid-2030s, the point at which it says launch and operation could become roughly comparable to a ground data center’s reported energy costs |
The launch-cost line is Google’s projection, not a measured result. The 2027 mission is the first planned test of the laser links: Google compares holding that connection to “hitting a coin-size target from miles away while both points are in motion.”
What comes next
When Planet announced its role on November 4, 2025, it said it would build and operate two prototype satellites for Google, targeting launch by early 2027, to test the TPUs in space and “the ability of two such spacecraft to work in concert.” Google’s later posts split that into steps: the single prototype launched on October 1, 2026, and Google’s October 30 post names “twin interconnected satellites in 2027” as the next milestone.
On the ground, Google’s TPU line keeps moving. On November 9, 2026, it announced two new eighth-generation systems, which we cover in TPU 8i Max and TPU 8tx. The Google TPU explainer covers the TPU generations and who can use them, and their specs sit in the AI chips table. Announced ground-based AI compute projects are in the AI data centers tracker.





