Google Project Suncatcher: TPUs in Orbit
On October 1, 2026, Google put four of its AI chips into orbit. Project Suncatcher is an experiment to see whether solar-powered satellites could one day run AI workloads, and its first prototype is testing the basics: heat, radiation and reliability. We explain what flew, what limits it faces, and how to judge the next wave of space data-centre claims.
⚡ Quick facts
- Launched: October 1, 2026, from Vandenberg on SpaceX's Transporter-18 rideshare, built with Planet
- Payload: four Trillium TPUs, roughly one server's worth, powered by about 1 kW of solar
- Main limit: heat: compute runs in bursts of about 15 minutes before the radiators must shed it
- Radiation: Trillium survived ground tests beyond a five-year low-Earth-orbit dose; the HBM memory was more sensitive
- Next step: a two-satellite mission with Planet in early 2027 to test optical links between satellites
What Google actually launched
Project Suncatcher is Google's research effort to find out whether AI computing could one day run on solar-powered satellites instead of in data centres on the ground. The first prototype launched on October 1, 2026, from Vandenberg Space Force Base on SpaceX's Transporter-18 rideshare mission. Google built it with the satellite company Planet. It carries four Trillium TPUs, Google's sixth-generation AI chips, which is about one server's worth of compute, powered by roughly 1 kW of solar panels. According to Science Times, contact was confirmed the same evening and the satellite is operating as expected.
Why heat is the real test
In space there is no air to carry heat away, so a satellite can only cool itself by radiating heat into space. Tech Times reports that the prototype can run its TPUs flat out for only about 15 minutes before the radiators must catch up. That makes this a test of thermal design as much as of the chips: if a small satellite cannot shed the heat of four TPUs, a full orbital data centre cannot work either.
Radiation and memory
Before launch, Google tested Trillium chips in a proton beam at the UC Davis Crocker Nuclear Laboratory. As InfoQ reported when Google first described the project, the chips survived more than the radiation dose expected over five years in low Earth orbit. The high-bandwidth memory (HBM) next to the chips was the more sensitive part. Hardware Busters puts it simply: for this mission, surviving the trip and running reliably is the whole test. Reported design life is about a year, though the satellite could stay up longer before it re-enters the atmosphere.
What comes next
The bigger question is whether satellites can work together like one computer. That needs fast laser links between them, which this prototype does not test. Google and Planet plan a two-satellite mission in early 2027 to try optical inter-satellite links, in a dawn-dusk sun-synchronous orbit at about 650 km, where solar panels stay in near-constant sunlight. Google's long-term concept is clusters of about 80 satellites flying close together. Futurum Group frames this as an early lead in an emerging race, but nothing here is a commercial service yet.
Our take: a checklist for judging space data-centre claims
Suncatcher will not be the last “AI in orbit” headline. When the next one appears, ask:
- How long can it compute? Continuous operation matters more than peak chip count. Suncatcher's 15-minute bursts show where today's limit is.
- Can the satellites talk to each other? Without fast optical links, each satellite is an isolated small server, not a data centre.
- What about memory errors? Radiation tolerance of the chip alone is not enough; Google's own tests found HBM was the weak spot.
- How long does the hardware last? A satellite that lasts a year must be replaced far more often than a server on the ground.
- Is it a service or an experiment? Suncatcher is clearly an experiment. Treat any claim of paid orbital compute in the next few years with caution.
Sources
- Tech Times: Google's Trillium TPUs reach orbit; radiators must beat a 15-minute heat limit
- Hardware Busters: Project Suncatcher puts four Google TPUs in orbit on October 1
- Science Times: Google's Project Suncatcher launches first prototype satellite
- InfoQ: Google's Project Suncatcher explores AI compute in space
- Futurum Group: Project Suncatcher prepares to launch TPUs
Frequently asked questions
What is Google Project Suncatcher?
It is a Google research project testing whether AI computing could run on solar-powered satellites. The first prototype carries four Trillium TPUs and was built with Planet.
When did the Suncatcher prototype launch?
It launched on October 1, 2026, from Vandenberg Space Force Base on SpaceX's Transporter-18 rideshare. Contact was confirmed the same evening.
How much computing power is on the satellite?
Four Trillium TPUs, about one server's worth, powered by roughly 1 kW of solar panels.
Why can the TPUs only run for about 15 minutes at a time?
In space heat can only be shed by radiating it away. Reports say the radiators need to catch up after about 15 minutes of full compute.
Can TPUs survive radiation in orbit?
In proton-beam tests at UC Davis, Trillium chips survived more than a five-year low-Earth-orbit dose. The high-bandwidth memory was more sensitive.
What is the next Suncatcher mission?
Google and Planet plan a two-satellite mission in early 2027 to test optical links between satellites, in a sun-synchronous orbit at about 650 km.