Google tests Suncatcher, taking its AI chips into space

Google is launching the first satellite of Suncatcher, the 'moonshot' project it has been working on for years to test AI chips in space. According to Ars Technica, the satellite, called MVP, will go into orbit on 1 October on SpaceX's Transporter-18 rideshare mission. The satellite is the size of a refrigerator and carries four of Google's own TPU AI chips.

Why a data centre in orbit?

According to Google's own explanation, satellites in low Earth orbit can receive almost uninterrupted sunlight, which means up to eight times as much solar energy generation as panels on the ground. In future, the company aims to link multiple satellite clusters with high-bandwidth laser connections to process larger AI workloads in orbit.

What tests will the satellite face in space?

According to Ars Technica, a rocket launch lasts about 10 minutes, during which the vehicle is subjected to ten times the force of gravity and some components, such as chips, to between fifty and a hundred times. Google simulated these vibrations on the ground by shaking the satellite along three axes and found that the hardware held up.

For the radiation test, according to Google, the team tested the TPUs while running at a proton beam facility at the Crocker Nuclear Laboratory at UC Davis. The Trillium TPUs withstood more than the total radiation dose they would receive during a five-year space mission.

Cooling is the project's biggest hurdle

Because there is no airflow in space, heat can only be dissipated through radiators. As reported by Ars Technica, Google's solution uses a flexible 'thermal interface material' that connects the chips to a radiator via aluminium and copper heat pipes. On the test flight, the TPUs will only be able to run in short bursts of 15 minutes before being paused to let the radiators catch up.

How will the satellites connect to each other?

According to Google, future satellite designs will carry dozens of TPU chips and orbit the Earth in clusters. To maintain bandwidth, each satellite must constantly know its own position and its location relative to its neighbours; this communication will be handled by lasers. Google likens the required precision to 'hitting a coin-sized target from kilometres away while both are moving'. Because current laser communication systems are designed for long distances at low bandwidth, this demands far greater precision.

The idea of orbital data centres is not unique to Google. As reported by Ars Technica, figures such as Elon Musk and Jeff Bezos have also put forward AI infrastructure in orbit as an alternative to ground-based data centres, which have become contentious because of power and cooling problems. Google's experiment is one of the first concrete steps to move this debate from the lab to real space data.

What does this mean for businesses?

Suncatcher is not yet a product but a multi-year research programme; Google itself says it will take years for it to turn from a 'project' into a 'product'. For companies investing in data centres or planning cloud capacity, the key message is that big technology companies are looking for alternatives to energy-constrained AI infrastructure, a search that could affect capacity pricing and investment priorities in the medium term.

In the short term, technology and infrastructure teams need not wait for orbital data centres; what really needs watching is which direction the big cloud companies take in siting data centres because of energy constraints.

• Orbital data centres are not yet a commercial product; managing expectations matters ahead of the second test Google has planned for 2027.
• The search for data centres that are not constrained by energy is growing; this trend should be monitored in capacity and investment decisions.
• Google's combination of its own chip with an off-the-shelf satellite bus (Planet Labs) is an example of an approach that prioritises speed.