Google is about to find out whether AI data centers can run in space. On October 1, 2026, the company plans to launch its first Project Suncatcher satellite — a prototype spacecraft carrying four of Google’s Trillium AI chips — into orbit to test how the hardware handles launch stress, radiation, and the freezing vacuum beyond the atmosphere.
The mission, announced by Google on September 24, 2026, is the first real-world test of an idea the company first floated late last year: that the future of AI computing might not be on the ground at all. During the flight, the satellite’s chips will run Google’s Gemini models in short bursts while engineers on Earth watch to see what works — and what breaks.
So what is Google Project Suncatcher, and why does the search giant want to put artificial intelligence in orbit? Here is everything we know about the mission, the technology, and the stakes.
What is Project Suncatcher?
Project Suncatcher is Google’s research program to explore whether AI workloads can run in space. The idea is simple in concept and brutally hard in execution: satellites enjoy almost constant sunlight and the natural deep cold of space — two things that data centers on Earth spend enormous amounts of electricity and money on.
Powering AI is becoming one of the tech industry’s biggest bottlenecks. Training and running large AI models consumes huge amounts of electricity and generates enormous heat. Google’s bet is that orbit could eventually solve both problems at once: solar panels provide round-the-clock energy, and space itself is the ultimate cooling system.
The first flight: four TPUs on a Planet-built satellite
The October 1 test is deliberately small. According to Google and subsequent coverage, a single prototype spacecraft — built in partnership with satellite imaging company Planet — will carry four Trillium TPUs, Google’s latest AI accelerator chips, into orbit aboard a SpaceX Transporter-18 rideshare launch.
The spacecraft carries about 1 kilowatt of solar power, and the computing will happen in carefully rationed doses: roughly 15-minute bursts of AI processing, followed by cooldown periods. The goal is not to train a model in space, but to answer a more basic question — can these chips survive and function at all up there?
The hard problems: radiation, vibration, and heat
Space is a hostile place for delicate silicon. Google says it has put the hardware through punishing ground tests to prepare:
- Launch stress: the satellite will experience forces of up to 10g during launch, with individual components absorbing 50 to 100g — far beyond anything a normal data center ever faces.
- Radiation: the Trillium TPUs have been tested to survive a total ionizing radiation dose greater than what a five-year mission in orbit would deliver, according to Google.
- Heat management: ironically, space computing has a heating problem, not just a cooling one — with no air to carry heat away, the chips rely on heat pipes and radiators to shed warmth, which is why the compute runs in short bursts.
The mission will also measure how the chips perform under constant vibration and thermal swings, data that will decide whether a larger orbital system is even worth building.
What comes next: laser-linked satellites in 2027
If the prototype survives its test, Google plans a bigger milestone in 2027: two satellites linked by laser communication, forming the first real prototype of an orbital compute network. Laser links — the same technology SpaceX uses for its Starlink constellation — would let space-based machines talk to each other at high speed without going through ground stations.
That 2027 test would be the first step from a single experiment toward an actual system: a constellation of satellites that could share AI workloads between them, thousands of kilometers above Earth.
Why Google wants AI in space
The timing is no accident. Demand for AI computing is exploding faster than anyone can build power plants and cooling systems for it on the ground. Google, Microsoft, and Anthropic are all racing to secure computing capacity — just this week, Anthropic signed a $11.6 billion cloud-infrastructure deal with Akamai to feed its growing appetite for compute. Space represents a radical long-term answer to the same problem: virtually unlimited solar energy and free cooling.
There is also a strategic angle. Whoever learns to run AI in orbit first gains an edge in everything from satellite data processing to global communications. Google is not alone in thinking about it, but Project Suncatcher is the most concrete orbital AI-compute experiment announced so far.
What to watch next
October 1 is the first real milestone: if the Transporter-18 launch goes smoothly and the TPUs boot up and run Gemini workloads in orbit, Google will have proven that AI chips can at least survive space. The harder questions — reliability over months, real workloads, and whether any of this is economical — come after. For another angle on the AI industry’s biggest moves this week, see Microsoft’s Copilot overhaul and the Claude Opus 5.5 launch.
For now, Project Suncatcher is exactly what Google says it is: an experiment. But it is an experiment that could quietly redraw the map of where computing happens.
