Google to orbit AI chips in SpaceX test flight

A prototype satellite equipped with Google’s tensor processing units (TPUs) is scheduled for deployment into low Earth orbit aboard SpaceX’s Transporter-18 mission. This launch marks the initial phase of Project Suncatcher, an initiative designed to assess whether current AI hardware can withstand the harsh conditions of space. The primary objective is not immediate operational capability but rather establishing the functional viability of TPUs in an orbital environment.

Project Suncatcher goals and timeline

Unveiled in November 2025, the broader concept envisions solar-powered AI infrastructure hosted by constellations of satellites linked via optical connections. While previous plans targeted two orbital prototypes by early 2027, this specific test advances the hardware validation stage. The project aims to determine if high-performance processors can operate reliably enough to support future space-based data centers.

Engineering challenges in orbit

The journey to low Earth orbit presents significant mechanical stresses, with acceleration reaching up to 10 g during the roughly 10-minute ascent. Individual components may endure loads between 50 g and 100 g. To address these risks, Google conducted three-axis vibration testing in laboratory settings prior to the launch.

Radiation exposure represents another critical factor. Trillium TPUs were subjected to proton bombardment at the UC Davis Crocker Nuclear Laboratory while running AI workloads. According to the company, the chips tolerated total ionizing-radiation doses exceeding those expected over a five-year orbital mission.

Thermal management and power efficiency

Cooling systems face unique constraints in a vacuum, where convection cannot remove heat from spacecraft. Google is evaluating heat pipes and radiators designed to transfer thermal energy away from the TPUs and dissipate it into space. This remains one of the key engineering questions the flight intends to resolve.

Solar energy offers substantial advantages for this architecture. Estimates suggest that a solar panel in a suitable orbit could generate up to eight times more energy than its terrestrial equivalent. However, continuous power generation does not eliminate the complexity of managing processor temperatures in such an environment.

Future constellation architecture

The proposed end-state design involves placing dozens of TPUs on each satellite, with multiple spacecraft forming a compute cluster. A subsequent milestone planned for 2027 involves launching two satellites capable of communicating through high-speed laser links. Maintaining precise pointing between rapidly moving spacecraft to sustain these connections constitutes a major technical hurdle before an orbital AI cluster becomes practical.

Transporter-18 serves as a foundational test for these premises rather than a deployment of conventional data center infrastructure. The mission seeks to verify if powerful AI processors can operate with sufficient reliability to justify the development of complex orbital networks.