According to The New York Times, Alphabet is preparing to advance the infrastructure testing of AI to low Earth orbit. The company plans to launch a test satellite equipped with custom computing hardware to verify the feasibility of deploying AI computing power in a real space environment.

First verification of TPU in orbit operation

This project is named Project Suncatcher. At this stage, instead of directly building a commercial orbital data center, we are using precursor satellites to carry out basic AI inference tasks to observe the performance of the chips under conditions of radiation, vacuum, and temperature changes.

The report mentions that the experimental satellite will carry a Tensor Processing Units customized by Google, which is also known as TPU. The focus of the mission is not to increase short-term business revenue, but to confirm whether the specialized AI chip can operate stably in orbital environments.

• Testing the chip's radiation resistance
• Verification of passive heat dissipation in a vacuum environment
• Evaluating the Support of Solar Power Supply for Computing Power

Planet Labs Participates in communication testing

To support distributed orbital computing, Alphabet also plans to collaborate with satellite imagery company Planet Labs to verify the optical laser communication links between satellites. Such high-bandwidth connections are considered an important foundation for future orbital computing networks.

If the relevant tests are successful, the computing nodes in space will not only undertake standalone tasks but may also form a connected computing power system. However, reports also point out that there is still a long way to go before large-scale commercial deployment, as launch costs and orbital maintenance remain practical limitations.

Aiming at the bottleneck of ground computing power

Alphabet This move reflects that large technology companies are seeking new solutions beyond traditional on-ground data centers. As the training and inference demands of AI continue to grow, power supply, heat dissipation, and site selection are becoming major constraints for the expansion of computing power.

In this context, low Earth orbit is seen as a frontier testing ground: on one hand, it allows for the utilization of solar energy; on the other hand, it enables avoidance of some of the pressures associated with terrestrial energy sources and infrastructure. This testing is still in its early stages, but it has already shown that large technology companies are pushing the AI computing power competition to even further limits.