MetaPatchAI home
AI

Google's Space TPU Just Wrote Its First Haiku: Project Suncatcher Says "Hello World" From Orbit

Analysis by Stephen Pierce · Published , 1:26 PM ET

Oct 30, 2026: A Google TPU in orbit has answered its first prompt. In a post on The Keyword dated Oct 30, 2026, Google says its Project Suncatcher prototype satellite reached its "Hello World" milestone, which the company calls "the first-ever AI response from a data center class TPU in orbit." The prompt asked for a haiku about being in space, and the open model on board, Gemma 4, wrote one.

What happened

The post is by Travis Beals, Senior Director, Paradigms of Intelligence at Google. It describes the steps like this:

  • Oct 1, 2026: Google launched the Project Suncatcher prototype satellite.
  • The following weeks: the team worked through a commissioning checklist: powering up the host and TPU trays and successfully running early machine learning tests.
  • This week: during a ground-station pass, the team uplinked its first prompt to the onboard Gemma 4 open model, asking for a haiku about being in space.
  • The result came down as the satellite passed by Svalbard, Norway.

This is the poem Google published:

Blue marble below Data streams in silent grace Starlight feeds my core.

The post adds that after this first output, the team also "tested space-themed limericks and telemetry prompts reporting the technical status of the payload systems."

What's next for Suncatcher

Google lists two next steps:

  1. Stress testing in orbit. The team will run "a series of tests to evaluate how TPUs perform under thermal stress and radiation exposure in orbit."
  2. Twin satellites in 2027. Google says showing that its TPUs can run workloads reliably in space "paves the way for our next milestone of twin interconnected satellites in 2027."

Beals: "Our chips have officially earned their space wings."

Quick background: what Project Suncatcher is

On Oct 1, 2026, Google's launch post called Suncatcher "a long-term research moonshot exploring whether space could one day host scalable machine learning infrastructure." That post also said:

  • The satellite was built in partnership with Planet.
  • It launched on SpaceX's Transporter-18 rideshare mission.
  • The team planned to collect in-orbit data on how the TPUs handle "the physical stress of spaceflight and the radiation and thermal extremes of space."
  • The research behind the mission has a peer-reviewed paper in Joule.

In its own Oct 1 press release, Planet said the satellite will run "the first-ever test of Google's Tensor Processing Units (TPUs) in space."

Third-party reporting (not in Google's posts): NPR reported that the prototype carries four TPUs and runs Gemma in roughly 15-minute bursts because of heat limits. NPR also issued a correction: this prototype is in a standard orbit and uses batteries.

Why it matters

The case for AI compute in orbit comes down to solar power. The hard problems are radiation and getting rid of heat in a vacuum. A haiku doesn't settle any of that. It does show the whole chain working: uplink, inference on an orbiting data-center-class TPU, and downlink. The radiation and thermal tests Google announced next will tell us more. Google hasn't published any numbers yet.

What's confirmed / not confirmed

Confirmed (Google, The Keyword, Oct 30, 2026, Travis Beals): - The Project Suncatcher prototype launched Oct 1, 2026, and finished commissioning: host and TPU trays powered up, early ML tests run. - "This week" the onboard Gemma 4 model answered its first uplinked prompt, a haiku, which was downlinked during a pass near Svalbard, Norway. - Google calls this "the first-ever AI response from a data center class TPU in orbit." - Next come thermal-stress and radiation tests, then twin interconnected satellites in 2027.

Confirmed (Google Oct 1 post and Planet's Oct 1 press release): the satellite was built with Planet, launched on SpaceX Transporter-18, and a peer-reviewed paper is in Joule.

Not confirmed / not disclosed: - The exact day of the first response (Google only says "this week"). - Any performance, error-rate, radiation or thermal data. None has been published yet. - NPR's four-TPU and 15-minute-burst details are third-party reporting, not stated in Google's posts. - Whether "first-ever" holds across the whole industry. That's Google's own claim, and it's limited to "data center class" TPUs.

Sources

Related

Up: AI