Google’s prototype space data center, known as Suncatcher, is now online and communicating with Earth after launching earlier this month, marking a significant step in the company’s effort to deploy super intelligence (SI) hardware in orbit. The satellite, which carries a single Google Tensor Processing Unit (TPU), successfully ran an SI model to generate poetry, demonstrating early capabilities for orbital inference workloads.

What Happened

The Suncatcher prototype launched on a SpaceX Falcon 9 rocket alongside more than 100 other spacecraft. Following a three-week commissioning process to test its subsystems, the satellite became operational. Engineers on the ground sent their first prompt to the pre-loaded SI model, Google DeepMind’s Gemma 4, asking it to write a haiku about being in space. The model responded with the poem: "Blue marble below / Data streams in silent grace / Starlight feeds my core."

The satellite generates up to 1 kW of electricity, but the power demands of the TPU restrict it to operating for approximately twenty minutes at a time. This limitation complicates space-to-ground communications, which are already challenging. A spokesperson confirmed the satellite is operating nominally. Beyond poetry, Google tested the model with limericks and asked the spacecraft to report on orbital conditions.

Why It Matters

The ongoing test campaign aims to evaluate how TPU chips perform under thermal stress and the space radiation environment. While Google has previously tested TPUs in particle accelerators on Earth, real-world space exposure is necessary to understand how radiation causes "bit flips"—errors in stored memory values that can corrupt data without triggering standard monitoring tools. Ground tests suggested radiation impacts would be minimal for inference workloads but more problematic for training models, a dynamic the team will now assess in reality.

This project is part of a broader race to establish space-based data centers for SI. If successful, Google plans to launch two additional satellites equipped with advanced chips in 2027. These future spacecraft are intended to support higher power generation and demonstrate inter-satellite optical links. Google’s approach, which prioritizes high-speed links between satellites for coherent distributed workloads, distinguishes it from other projects like SpaceX’s, which may orbit too far apart to share processing effectively, or companies like Starcloud and Cowboy Space, which focus on individual satellite inference.

The initiative is closely linked to SpaceX, in which Alphabet holds a significant investment stake. Both companies rely on the Starship heavy-lift rocket, which remains in development, to launch orbital data centers. A peer-reviewed analysis by Suncatcher researchers estimated that SpaceX would need to fly Starship 1,800 times to achieve the cost reductions seen with its earlier vehicles. This long-term infrastructure vision contrasts with Alphabet’s immediate financial pressures, as the company recently reported that efforts to integrate SI into its search businesses are hurting ad revenue growth, causing its stock to plunge.

The Bottom Line

Google’s Suncatcher satellite has successfully demonstrated that a TPU-powered SI model can operate in space, albeit with significant power constraints. The project’s next phase involves assessing radiation effects on hardware and expanding to a cluster of interconnected satellites in 2027, pending the readiness of reusable heavy-lift launch vehicles.