Google has successfully booted up an artificial intelligence processor aboard a satellite in orbit and generated text directly from space. According to reports from TechCrunch and The Verge, ground engineers sent a prompt to Google’s experimental Project Suncatcher satellite, asking an on-board AI model to compose a haiku about being in space.
The spacecraft, running Google DeepMind’s lightweight Gemma 4 model, responded: “Blue marble below / Data streams in silent grace / Starlight feeds my core.”
While generating a three-line poem might sound modest, the test represents a milestone for edge computing—the practice of processing data directly where it is gathered rather than beaming raw information back to data centers on Earth.
How the orbital hardware operates
The Project Suncatcher prototype launched earlier in October 2026 aboard a SpaceX Falcon 9 rocket alongside more than 100 other spacecraft, according to TechCrunch. After a three-week setup period to check its basic systems, engineers began testing its processing capabilities.
The satellite carries four of Google’s custom Tensor Processing Units (TPUs) and generates up to one kilowatt of electricity. However, because AI chips consume significant amounts of power, the satellite can only keep a single TPU powered on for about 20 minutes at a time.
In addition to asking for space-themed poems, engineers instructed the satellite to transmit telemetry reports on its orbital surroundings. The team is now conducting tests to evaluate how the processor handles extreme thermal changes and space radiation. In space, radiation can trigger “bit flips,” where radiation alters stored values in computer memory and corrupts data without setting off standard system alerts. Ground testing indicated bit flips mainly affect model training rather than inference—the process of generating responses from a trained model—but orbital tests will verify how the hardware behaves in real conditions.
If the current prototype passes these checks, Google plans to launch two interconnected satellites in 2027. Those spacecraft will feature higher power capabilities and test high-speed optical laser links between satellites in orbit.
Why tech companies are heading to space
Building computing infrastructure in orbit is part of a broader search for alternative ways to handle mounting AI workloads. Traditional data centers require massive amounts of land, water, and electricity, leading to local friction in several regions as data centres face local voter backlash over energy demands.
Running AI models directly on satellites could allow space hardware to process imaging and sensor data in orbit, sending concise summaries down to Earth instead of massive raw data files. Google is not alone in exploring orbital servers; companies including SpaceX, Starcloud, and Cowboy Space are pursuing similar concepts, according to TechCrunch.
Google’s strategy relies on linking multiple satellites together via optical lasers so they can share processing tasks across a single network. Achieving that scale will depend on low launch costs. Alphabet, Google’s parent company, holds an investment stake in SpaceX and plans to rely on SpaceX’s heavy-lift Starship rocket. A Suncatcher research paper published earlier this month estimated that SpaceX would need to launch Starship 1,800 times to achieve the necessary cost reductions for large-scale orbital computing.
Can you use this today?
No. Project Suncatcher is an experimental internal research project, not a commercial service. Google has stated that Suncatcher is focused on infrastructure designed to become economical in the 2030s.
While everyday users cannot interact with the AI satellite in orbit, developers on Earth can test versions of Google’s Gemma models locally, while general consumers can access standard Google AI services through web browsers and mobile apps for free.