Google tests solar-powered AI data centers in orbit

Google has deployed an experimental satellite into orbit as part of the initial phase of its "Sun Catcher" project. The initiative aims to determine whether artificial intelligence infrastructure can operate in space by relying on solar power. The company confirmed that communication with the satellite was established successfully and that it is functioning as expected.

Satellite launch and hardware details

The mission was carried out aboard SpaceX's Transporter-18 flight in collaboration with Planet Labs. The payload includes Tensor processing units developed by Google to run AI applications. This experiment seeks to gather real-world data on the resilience of these chips against space conditions, including vibrations and acceleration during launch, as well as exposure to radiation and extreme temperatures.

The journey to low Earth orbit lasts approximately ten minutes. During this period, the satellite experiences acceleration reaching up to ten times the force of gravity. Certain components, including the processing units, may be subjected to forces equivalent to 50 or 100 times gravity. Prior to sending the chips into space, Google conducted ground vibration tests to simulate launch conditions.

Radiation and thermal challenges

Launch represents only the beginning of the testing process. Radiation outside the atmosphere can cause electronic errors within the chips. Additionally, heat dissipation poses a significant challenge because there is no airflow in space to cool devices. Consequently, Google is testing the use of heat pipes and radiators to manage the heat generated by the processing units' operations.

The core concept of Sun Catcher relies on the abundance of solar energy in low Earth orbit. Satellites in this region can access sunlight for longer durations and obtain up to eight times more solar energy than is available on Earth. Looking ahead, Google hopes to link multiple satellites together to form computing clusters capable of handling larger AI workloads.

Future connectivity plans

Achieving this goal requires solving communication issues. Google plans to test high-bandwidth laser communications between two satellites in 2027. This step will help determine if a large number of satellites can be connected at speeds sufficient to support AI workloads.

The idea remains in the research stage and faces hurdles related to launch costs, radiation, cooling, and communications. Google states that the objective of the current experiment is to identify what works and pinpoint failure points before moving toward larger designs.