Recommended Articles
Tech & AI

Google Ran an AI Chip in Orbit — and It Wrote a Haiku About Space

By Angeled · November 3, 2026 · Tech & AI
Google Ran an AI Chip in Orbit — and It Wrote a Haiku About Space
Google Ran an AI Chip in Orbit — and It Wrote a Haiku About Space

Google has taken a data-center-class AI chip to space and brought back a poem. Project Suncatcher, a prototype satellite launched October 1, 2026, aboard a SpaceX Transporter-18 rideshare mission with Planet, has successfully run inference on a Tensor Processing Unit (TPU) in low Earth orbit, according to reporting by The Register.

The experiment is small in scale and deliberately framed as long-term research. But it points to a question that will only grow louder as AI demand strains terrestrial power grids: what if some computing moves off the planet entirely?

The experiment

The Suncatcher prototype carries four Trillium TPUs, Google's data-center-class AI accelerators, powered by roughly one kilowatt of solar energy. That power budget is the key constraint: it limits continuous operation to about one TPU for roughly 20 minutes at a time. This is not a data center in space; it is a proof of concept that a data-center chip can survive and function in orbit at all.

For its demonstration, the team uplinked a prompt asking the onboard Gemma 4 open model to write a haiku about space. The satellite processed the prompt in orbit and downlinked the result over Svalbard:

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

The poem itself is a novelty. What matters is the round trip: a prompt sent up, inference performed on an orbiting TPU, and a result sent back down.

Why orbit?

The case for orbital computing rests on two resources that are abundant in space and constrained on Earth: sunlight and cooling. Solar energy is available almost continuously in orbit, unimpeded by weather or night in certain orbits, and the cold vacuum of space offers a vast heat sink. On Earth, by contrast, AI data centers face growing constraints — power availability, grid interconnection queues, water for cooling, and local opposition to new facilities.

Google frames Suncatcher as research into whether orbital compute could one day supplement terrestrial data centers, not replace them. That framing is important. The technical hurdles are formidable: radiation degrades electronics, thermal management in vacuum is difficult, and launching and maintaining hardware in orbit is vastly more expensive than building on land.

What comes next

Google's next steps for the prototype are thermal and radiation stress tests — essentially, finding out how the hardware holds up under the harsh conditions of space over time. Those tests will determine whether orbital TPUs are a curiosity or a credible long-term direction.

The project also raises practical questions that go beyond engineering. Data sovereignty, latency, launch costs, and the regulatory environment for space-based computing all remain unresolved. A satellite data center would need to answer who governs the data it processes, how quickly it can respond to users on the ground, and whether the economics ever beat building another facility in a desert with cheap solar power.

Why it matters for founders and investors

For founders, Suncatcher is a reminder that infrastructure assumptions are not permanent. The defining constraint of the AI era so far has been access to power and compute; any technology that changes that equation — even at the margins — reshapes who can build what. Orbital compute is years from commercial relevance, but the research direction signals where the industry's biggest players believe the bottlenecks lie.

For investors, the project is best read as an option, not a product line. Google can afford to explore orbital AI while its terrestrial business scales. The value of Suncatcher today is not revenue but information: learning, cheaply and early, whether space is a viable frontier for the industry's most expensive resource — compute itself.

Read Next