A prototype orbital data center built by Google's Suncatcher project has successfully established contact with ground teams following its launch earlier this month. The satellite rode into orbit aboard a SpaceX Falcon 9 rocket alongside more than 100 other spacecraft, and has since completed its initial commissioning phase.

The spacecraft carries four of Google's proprietary Tensor Processing Units (TPUs) and is capable of generating up to one kilowatt of electrical power. Its purpose is to demonstrate the viability of running large-scale AI inference workloads from orbit, with the longer-term goal of establishing networks of closely spaced satellites that can coordinate on computing tasks. The hardware comes with a notable limitation, however - the power requirements of each TPU mean only one unit can be active at any given time, and only for roughly 20-minute windows. That constraint adds further complexity to what is already a challenging communications environment between space and ground.

A Haiku From Orbit

After roughly three weeks of subsystem checks and commissioning work, the satellite was declared ready for its first operational test. Engineers on the ground sent the spacecraft its initial prompt this week, asking the onboard AI model - Google DeepMind's Gemma 4 - to compose a haiku about the experience of being in space. The model responded with the following:

Blue marble below

Data streams in silent grace

Starlight feeds my core.

Beyond the creative exercise, engineers also prompted the satellite to provide a report on the orbital conditions surrounding the spacecraft. A company spokesperson confirmed that the satellite is currently operating within normal parameters.

Radiation Testing Is the Real Work

The more technically critical phase of the mission involves evaluating how the TPU chips hold up under the conditions unique to space. Specifically, the test campaign will assess the chips' behavior under thermal stress and within the radiation environment that exists beyond Earth's atmosphere. Google has already subjected its TPUs to radiation testing using particle accelerators on the ground, but real-world exposure in orbit is a different matter.

Radiation poses a particular risk to computing hardware through what are known as "bit flips" - events where a stored binary value in computer memory is spontaneously altered. These flips can silently corrupt data in ways that standard monitoring systems may not catch, making them especially problematic for mission-critical workloads.

Ground-based simulations suggested that radiation would have a relatively minor effect on inference tasks while posing a more significant challenge to model training. The current mission will allow the Google team to validate or revise those projections using actual in-orbit data rather than modeled approximations.

Next Steps: Two More Satellites in 2027

If the onboard TPUs demonstrate reliable performance throughout this mission, Google plans to move forward with launching two additional satellites equipped with more advanced versions of the chips in 2027. Those follow-on spacecraft are designed to support greater power output and will also test inter-satellite optical communication links - a capability that sets the Suncatcher project apart from most competing visions for space-based computing infrastructure.

Among the various companies and projects exploring orbital data centers, Suncatcher is one of the few designing its architecture around high-speed laser links between satellites that would allow coherent distributed workloads to be shared across multiple spacecraft. SpaceX has also indicated plans to use laser links in its own space-based data center effort, but the expected orbital spacing between its satellites would likely prevent effective processing coordination between them. Other companies in the space computing sector, including Starcloud and Cowboy Space, are focused on running inference workloads on individual satellites rather than distributed networks.

Intertwined Ambitions: Google, SpaceX, and Starship

The trajectories of Google and SpaceX in this domain are closely connected on multiple levels. Alphabet, Google's parent company, holds a notable investment stake in SpaceX. More practically, both companies' orbital data center ambitions ultimately depend on SpaceX's Starship rocket - the heavy-lift vehicle still under active development - becoming fully and rapidly reusable at scale.

A peer-reviewed study published earlier this month by Suncatcher researchers estimated that SpaceX would need to fly Starship approximately 1,800 times before achieving the kind of cost reductions driven by accumulated operational experience that the company previously realized with its earlier rocket platforms. That figure underscores how much of the economic case for space-based computing infrastructure rests on assumptions about future launch economics that have not yet been proven out.

A Long-Term Bet Under Near-Term Pressure

Suncatcher fits the profile of what Google has historically referred to as a moonshot - a long-horizon infrastructure project designed to be commercially viable in the 2030s rather than in the immediate future. The project's leadership is focused on building the foundational architecture for what orbital computing could look like a decade from now, rather than chasing near-term revenue.

That long-range focus, however, sits within a broader corporate context that is under increasing pressure. Alphabet's stock dropped sharply this week following an earnings report that revealed a troubling dynamic: the company's efforts to embed AI capabilities into its core search products are beginning to weigh on advertising revenue growth. How the company navigates those near-term financial pressures - and the resource allocation decisions that follow - may ultimately shape how much runway a project like Suncatcher continues to receive.

Editor's note: The Suncatcher prototype satellite carries four TPUs in total, but the system is designed to operate only one at a time due to power constraints.