Google will launch its AI chips into orbit for the first time next week as its Project Suncatcher space data center program takes another step forward.

The hyperscaler said it will conduct its first test in orbit, launching a prototype satellite to evaluate how its Tensor Processing Unit (TPU) AI hardware performs in space.

Details of Project Suncatcher were first revealed last year, with Google stating it would have a two-satellite constellation up and running by 2027. It is part of a wave of space data center projects being undertaken by some of the biggest names in tech, as companies explore whether off-planet compute can provide additional capacity to run AI systems. The likes of Elon Musk’s SpaceX and Jeff Bezos-backed Blue Origin have revealed this year that they are seeking regulatory approval of orbital compute clusters.

This initial Project Suncatcher deployment will be launched into space on the upcoming Transporter-18 rideshare SpaceX mission, which will transport craft from several different vendors into low Earth orbit (LEO), and is scheduled to take off on October 1. Google’s system has been developed in partnership with satellite firm Planet Labs.

Google’s Project Suncatcher: Shaking and radiating

The initial test craft, which Google has named MVP, per a report from the New York Times, will feature just four TPUs, which will be powered by solar panels supplying them with 1kW.

In a blog post about Project Suncatcher, Travis Beals, Google’s senior director for paradigms of intelligence, revealed some of the challenges the team has been trying to solve to ready its compute infrastructure for space.

To try and combat the vibrations the chips will experience when being blasted into space, as well as high g-force in LEO, the Suncatcher team conducted vibration testing by intensely shaking the satellite on all three axes to mimic the frequencies of a rocket launch. “Tests like this rarely go as planned, so we were pleasantly surprised that the hardware held up to the force,” Beals said.

The TPUs were also tested in a proton beam facility at UC Davis’s Crocker Nuclear Laboratory while running AI workloads, to try and simulate how the hardware would perform when facing the high levels of radiation it will encounter outside the Earth’s atmosphere. This radiation can play havoc with electronic systems.

“During the test, we monitored closely to see how errors, like a bitflip, would affect our workloads,” Beals said. “Initial results have shown that our Trillium TPUs hold up remarkably well, and can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission.”

Cooling and connectivity

The cold conditions found in space are often cited as a potential benefit for orbital data centers, but the lack of atmosphere also means there is no airflow to transfer heat away from the components.

Beals said Google is working on a number of different approaches to cooling its satellites, including “a combination of heat pipes and radiators to cool the chips.”

He said: “So far, our team has tested the technology in a thermal vacuum chamber that simulates both the thermal and vacuum environment in space. We’ll see how our new TPU cooling system works in space and refine our designs as we learn more.”

Google has said previously that it envisions 1km (0.6 mile) arrays of 81-satellite compute clusters. These will be connected via laser links, Beals said. Constellations like SpaceX's Starlink network already use laser-based communication.

“The technology in space already exists, but most state-of-the-art systems are optimized for low bandwidth across large distances, whereas our lasers need to operate at very high bandwidth over extremely short distances,” Beals explained.

“Maintaining the necessary connection requires extraordinary precision, similar to hitting a coin-size target from miles away while both points are in motion. We’ll test our work on this in 2027 when we put two satellites in orbit.”