Google's "computing satellite" will launch next week! It will be launched on a SpaceX rocket and will need to shut down for cooling every 15 minutes.

Google is about to launch its self-developed AI chip into low Earth orbit for the first time, exploring ways to extend AI computing power from ground-based data centers to space.

Next week, an experimental satellite jointly developed by Google and satellite imaging company Planet will be launched into space aboard SpaceX's Transporter-18 rideshare mission. The satellite carries four Google-designed Trillium TPU chips and will run Google Gemini models in orbit, receiving and responding to commands from the ground.

This marks the first in-orbit test of Google's Project Suncatcher. Google's long-term goal is to deploy an AI computing cluster consisting of multiple satellites in Earth orbit to obtain more solar energy and alleviate the power and land constraints faced by ground-based data centers.

However, this mission is still in the early stages of technology validation. James Manyika, Google's senior vice president of research, stated that the current focus is on verifying whether the chip and related hardware can withstand launch, radiation, and the thermal environment of space, rather than immediately launching a commercial system. A more practical issue is that even heat dissipation for operating AI chips in space has not yet been fully resolved—the satellite must be shut down for cooling after approximately 15 minutes of full-power operation each time.

Why is AI computing power targeting space?

One of the core reasons driving Google's exploration of space-based data centers is the ever-increasing energy demands of AI infrastructure. As data centers continue to expand, power supply, land, and grid capacity face increasing constraints, while low Earth orbit can access more abundant solar energy, theoretically generating up to eight times more electricity than on the ground.

This idea was first proposed by Blaise Agüera y Arcas, Google's vice president. Three years ago, he formulated this concept at a closed-door meeting discussing the energy bottlenecks of AI, and then reported it to CEO Sundar Pichai and co-founder Sergey Brin, after which the project received support and was launched.

Google estimates that as rocket launch costs continue to decline, the overall cost of operating data centers in space could approach that of terrestrial data centers by the mid-2030s. However, to achieve this goal, the fundamental engineering challenges of running chips in orbit must first be solved.

To send chips into space, they must first overcome three hurdles.

The first hurdle is the launch impact. During rocket liftoff, the satellite needs to withstand approximately 10 times the force of gravity, and the internal chip components may even experience impacts of 50 to 100 times the force of gravity. The Google team conducted a full three-axis vibration test on the satellite, and none of the fixing screws loosened after the test.

The second hurdle is space radiation. Once outside the protective atmosphere, cosmic rays and high-energy particles can penetrate electronic components, causing calculation errors. Google sent its Trillium TPU to the Crocker Nuclear Laboratory at UC Davis, where it was bombarded with proton beams to simulate the cumulative radiation dose of a five-year orbital mission while running AI workloads. Tests showed that the chip could withstand this dose, and calculation errors could be recovered through a system reboot.

The third challenge, and currently the most difficult, is heat dissipation. In a vacuum environment, there is no air convection, so the heat generated by the chip can only be dissipated through conduction and radiation. Google designed a purely conductive heat dissipation structure that transfers heat to the metal layer, and then the heat is dissipated into the vacuum through radiating plates.

However, this approach still limits the chip's continuous operating capability: the satellite can only operate at full power for about 15 minutes at a time, after which it must be shut down to cool down. For data centers that need to continuously process AI tasks, this means that thermal management remains a critical aspect that must be addressed for large-scale deployment.

The race to create space-based data centers heats up: Google plans to validate two satellites by 2027.

The experimental satellite launched this time is roughly the size of a household refrigerator. Its deployed solar panels provide only about 1 kilowatt of power, and it is designed to operate for one year before burning up in the atmosphere. Compared to a complete data center, it is more like an orbital test platform for verifying key technologies.

Google's long-term vision is to deploy a computing cluster consisting of multiple satellites. The current plan is to have 81 satellites fly in formation at an altitude of about 650 kilometers and within a radius of 1 kilometer, sharing computing power and data through ultra-high bandwidth inter-satellite laser links.

This presents new challenges for laser communication. Traditional space-based laser communication is primarily designed for long-distance data transmission, while Google envisions satellites that need to maintain precise alignment over distances of several kilometers and at high speeds. Google likens this challenge to "hitting a moving coin at high speed."

The first on-orbit test of the technology is scheduled for 2027, when Google will launch two satellites simultaneously to verify the laser networking capabilities between the satellites.

Google isn't the only company betting in this direction. According to The Next Web, Nvidia-backed startup Starcloud launched a satellite equipped with an H100 chip into orbit last November; SpaceX plans to launch its first Starmind AI satellites equipped with Nvidia chips as early as the end of 2027.

However, significant obstacles remain before space-based AI infrastructure can be widely deployed. Currently, approximately 44,870 objects have been tracked in low Earth orbit, most of which are debris, with the orbital space around 650 kilometers being particularly crowded. Meanwhile, launch costs, on-orbit engineering, and the capacity for large-scale satellite manufacturing will all influence whether this concept can truly move from technological experimentation to commercial infrastructure.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.