Google has spent years putting artificial intelligence into phones, laptops, search engines, and data centers. Next week, it plans to put AI hardware somewhere considerably harder to reach… space!
Yes, the search giant is preparing to launch a prototype satellite carrying its Tensor Processing Units as part of Project Suncatcher, an experiment looking at whether large-scale AI computing could eventually operate in orbit. This is not an orbital data center yet, but Google is taking the first physical step toward finding out whether such a thing could actually work.
The prototype is expected to hitch a ride aboard SpaceX’s Transporter-18 mission. Google developed the initial mission with Planet, and the goal is fairly straightforward: put its AI hardware into low Earth orbit and see what happens. There are plenty of things that can go wrong.
Getting the chips off the ground is the first problem. A rocket launch exposes hardware to intense vibration and acceleration of up to 10 times Earth’s gravity, while individual components can experience forces reaching 50 to 100 g. Google says it shook its satellite across all three axes during testing to simulate those conditions, and the hardware survived.
Radiation is another concern because Earth’s atmosphere provides protection that electronics do not get once they head into space. Solar activity and cosmic rays can interfere with chips, potentially causing errors such as bit flips, which could obviously become a problem when those chips are busy processing AI workloads.
Google tested its Trillium TPUs using a proton beam at UC Davis’s Crocker Nuclear Laboratory while simultaneously running AI workloads. According to the company, the chips survived a total ionizing radiation dose greater than what they would be expected to encounter during a five-year mission.
Cooling AI hardware presents another big challenge. It is already difficult enough inside conventional data centers, but putting those chips in a vacuum removes airflow entirely, meaning Google cannot simply scale up the cooling systems it uses on Earth.
Project Suncatcher is experimenting with heat pipes and radiators instead. Google has already tested the technology inside a thermal vacuum chamber designed to simulate conditions in space, but the upcoming mission will show how the cooling system behaves in the real thing.
So why bother putting AI hardware in space at all? Energy is a big reason, as Google says satellites in low Earth orbit can receive nearly continuous sunlight and potentially generate up to eight times as much solar power as systems on Earth.
If the economics and engineering eventually make sense, orbital computing infrastructure could tap into that energy without consuming the land and electricity required by increasingly enormous terrestrial AI data centers. With AI companies scrambling for more power and compute, it isn’t difficult to understand why Google would at least explore the possibility.
Google’s ambitions go well beyond a single satellite. Future Project Suncatcher designs could place dozens of TPUs aboard each satellite and connect groups of them using lasers, allowing the hardware to work together on larger AI workloads.
Those connections would need extremely high bandwidth while maintaining precise alignment between objects traveling around Earth at enormous speeds. Google compares the challenge to hitting a coin-sized target from miles away while both points are moving.
The company plans to take another crack at that problem in 2027 by launching two satellites and testing communication between them. Eventually, the idea is to connect multiple satellite constellations capable of sharing much larger AI workloads.
It sounds like science fiction, but next week’s launch should provide Google with something far more useful than simulations: actual data from AI chips operating in space. That should help determine which parts of Project Suncatcher are realistic and which need considerably more work.
Of course, proving that a TPU can survive orbit is very different from proving that orbital AI infrastructure makes sense. Launch costs, repairs, hardware replacement, cooling, networking, and the sheer complexity of maintaining computers hundreds of miles above Earth could make the idea impractical at scale.
For now, Project Suncatcher remains exactly what Google calls it: a moonshot. When the first TPUs head into orbit next week, however, Google will finally get a chance to find out whether putting serious AI compute in space is brilliant, ridiculous, or perhaps a little bit of both.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz
