Project Suncatcher’s Orbiting TPU Delivers Its First AI-Generated Haiku

Google said on October 30, 2026, that its Project Suncatcher prototype satellite returned its first AI response from orbit: a haiku generated by an onboard Gemma 4 open model running on a data-center-class Tensor Processing Unit (TPU).

Travis Beals, Senior Director of Paradigms of Intelligence at Google, announced the milestone in a Google Research post, describing it as the mission’s “Hello World” milestone and what the company characterized as the first-ever AI response from a data center class TPU in orbit. During a ground-station pass, the team uplinked its first prompt to the satellite’s onboard Gemma 4 model, asking it to generate a haiku about being in space. As the satellite passed Svalbard, Norway, the downlinked poem confirmed success: “Blue marble below / Data streams in silent grace / Starlight feeds my core.”

Following the initial output, the team tested space-themed limericks and telemetry prompts that reported the technical status of the payload systems.

Commissioning After Launch

The satellite launched on October 1, 2026, aboard the Transporter-18 rideshare mission with SpaceX and was built in partnership with Planet, according to Google’s launch announcement. The team confirmed contact with the spacecraft and said it was operating as expected.

In the weeks after launch, the team completed a commissioning checklist that included powering up the host and TPU trays and successfully running early machine learning tests, Beals wrote. The mission is designed to gather in-orbit data on how Google’s TPUs handle the physical stress of spaceflight and the radiation and thermal extremes of space, according to the launch announcement. A peer-reviewed paper detailing the research behind the mission is now available in the journal Joule.

A Moonshot Announced in 2025

Google announced Project Suncatcher on November 4, 2025, describing it as a long-term research moonshot exploring whether an interconnected network of solar-powered satellites equipped with its TPU AI chips could one day scale machine learning in space. The company’s initial research, shared in a preprint paper alongside the announcement, described its approach to satellite constellation design, control, and communication, along with initial learnings from radiation testing of Google TPUs. The plan then called for a learning mission in partnership with Planet to launch two prototype satellites by early 2027 to test the hardware in orbit.

In a behind-the-scenes engineering post published September 24, 2026, Google said that in low Earth orbit, satellites can access near-constant sunlight, generating up to eight times more solar power than on Earth. The company said it could eventually be possible to link together multiple constellations of satellites, allowing them to manage larger AI workloads while in orbit.

Ground Testing and the 2027 Milestone

Before the launch, the team tested whether the hardware could survive the trip to orbit. A rocket ride into low Earth orbit lasts about 10 minutes, with sustained acceleration loads up to 10 times the force of gravity, and individual components such as the TPU chips can experience forces of 50 to 100 g, according to the engineering post. The team conducted vibration testing by intensely shaking the satellite on all three axes to mimic the frequencies of a rocket launch.

For radiation, the team tested TPUs in a proton beam facility at UC Davis’s Crocker Nuclear Laboratory while running AI workloads, monitoring how errors such as a bitflip would affect those workloads. Google reported that initial results showed its Trillium TPUs can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission.

Cooling is another challenge the team described. TPUs generate a large amount of heat in a small area, and in the vacuum of space there is no airflow, so heat can only be diffused via radiators. The team is pursuing a combination of heat pipes and radiators to cool the chips and has so far tested the technology in a thermal vacuum chamber that simulates both the thermal and vacuum environment in space.

Future satellite designs would each carry dozens of TPU chips while orbiting Earth in clusters, communicating via lasers that must operate at very high bandwidth over extremely short distances — a pointing challenge the post compares to hitting a coin-size target from miles away while both points are in motion. Google plans to test that interconnect work in 2027, when it puts two satellites in orbit.

Next, Beals wrote, the team will run a series of tests to evaluate how the TPUs perform under thermal stress and radiation exposure in orbit. Google said verifying that its TPUs can reliably execute workloads in space paves the way for the project’s next milestone: twin interconnected satellites in 2027.