Google is about to find out whether its AI chips can survive the trip to space. In a blog post published Wednesday, the company revealed that Project Suncatcher, its long-running moonshot to put machine learning infrastructure in orbit, is scheduled to embark on its first test: a prototype satellite carrying Google Tensor Processing Units (TPUs) will launch aboard SpaceX's upcoming Transporter-18 rideshare mission, developed in partnership with satellite imaging company Planet.
According to Google, the team is "putting our first TPUs in orbit next week," marking the moment the two-year-old research project moves from laboratory simulations to the real thing. For more context on this story, see our ongoing AI news.
Why Google Wants AI Compute in Space
Project Suncatcher was announced last year as a long-term research effort exploring whether space could one day host scalable machine learning infrastructure. The core argument is about power. In low Earth orbit, satellites can access near-constant sunlight, generating up to eight times more solar power than an equivalent array on Earth, according to Google's figures.
The end goal, the company says, is to link together multiple constellations of satellites so they can manage larger AI workloads while in orbit. Google frames the ambition in the same category as its early research into autonomous driving and quantum computing: years of experimentation before practical systems emerge.
But before constellations come basics, and the first question the prototype must answer is blunt: can Google's AI hardware operate in space at all?
Surviving the Launch: Vibration and G-Forces
The journey to orbit is violent. A rocket ride to low Earth orbit lasts about 10 minutes, during which the spacecraft endures intense vibration and sustained acceleration loads of up to 10 times the force of gravity. Individual components such as TPU chips can experience even greater forces, in the range of 50 to 100 g, according to the blog post.
Google's engineers conducted vibration testing by intensely shaking the satellite on all three axes to mimic the frequencies of a rocket launch. The team noted that such tests rarely go as planned, and said it was pleasantly surprised that the hardware held up.
Radiation: The Proton Beam Test
Once in space, radiation becomes the primary threat. Solar events and cosmic rays can wreak havoc on electronics, corrupting data and damaging silicon. To gauge the risk, Google tested its TPUs at a proton beam facility at UC Davis's Crocker Nuclear Laboratory while the chips were actively running AI workloads, closely monitoring how errors such as bit flips affected the workloads.
The initial results were encouraging. Google reports that its Trillium TPUs held up remarkably well and can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission. Still, the company acknowledges that some things can only be tested in space, which is precisely what the upcoming orbital demonstration is for.
Cooling Chips With No Air
Heat is the third unsolved problem. TPUs generate a large amount of heat in a small area, and on Earth data centers rely on airflow to carry it away. In the vacuum of space, there is no air. The only way to diffuse heat is through radiators, which Google describes as requiring a totally different approach to cooling electronics.
The team is working on a combination of heat pipes and radiators to cool the chips, and has tested the technology in a thermal vacuum chamber that simulates both the thermal and vacuum environment of space. How the system performs in actual orbit will inform future designs.
Lasers, Constellations and the 2027 Milestone
The prototype is only the first step. According to Google's post, future designs will carry dozens of TPU chips per satellite, flying in clusters around the Earth. To maintain the bandwidth necessary to process AI, the satellites will need to be interconnected, with high-bandwidth laser links doing the work that data center fiber does on the ground.
Google says it is working toward its next milestone in 2027, with the data from the Transporter-18 mission informing future launches. Travis Beals, Senior Director of Paradigms of Intelligence at Google Research, is leading the public explanation of the effort, and the company has published a new video series digging into the science behind the mission and the engineering challenges that remain.
A Race Beyond Earth's Data Centers
Google is not alone in looking to the sky. The idea of orbital data centers has drawn interest across the tech industry as terrestrial power grids strain under the demands of AI computation, and launch costs keep falling. What distinguishes Project Suncatcher is that it is no longer conceptual: within days, working TPU hardware will either survive in orbit or it will not.
The experiment is deliberately small scale. One satellite, one rideshare slot, and a list of questions that can only be answered 500 kilometers above the Earth. But if the TPUs come through their first weeks in space unscathed, Google will have taken the first verifiable step toward putting a meaningful share of the world's AI compute where the solar power never stops.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
