For years, the biggest constraint on artificial intelligence has been what happens on the ground. AI models need enormous amounts of computing power, and that means data centres need vast supplies of electricity, cooling systems and physical space. As demand for AI continues to grow, technology companies are beginning to look somewhere very different for the infrastructure they will need next that is above earth.
Google is preparing to test whether some of that infrastructure could eventually operate in orbit. On October 1, the company plans to send a prototype satellite into low Earth orbit aboard SpaceX’s Transporter-18 rideshare mission. Built in partnership with satellite company Planet, the satellite will carry four Google Tensor Processing Units, or TPUs, into space for the first time.
The experiment is part of Project Suncatcher, Google’s long-term research effort to explore whether orbital satellites could eventually work together as AI data centres. Google is not putting a data centre in space. It is trying to answer a much more basic question, can its AI hardware survive there?
Tensor Processing Unit (TPU) is a custom application-specific integrated circuit (ASIC) designed by Google Cloud to accelerate artificial intelligence and machine learning workloads. Google CEO Sundar Pichai in a post on X said, “Can our TPUs survive and operate in space? Well, we’re going to find out. Project Suncatcher is hitching a ride aboard @SpaceX’s Transporter-18 mission, testing a prototype satellite built in partnership with @planet. One small step for TPUs….”
Why put AI in space?
The attraction begins with energy. AI data centres on earth consume huge amounts of electricity. In space, satellites operating in certain low Earth orbits can receive sunlight for much longer periods, with some orbits offering almost continuous exposure to the sun. Google estimates that satellites in low Earth orbit could generate up to eight times more solar power than comparable systems on Earth.
The company’s longer-term vision is to use that energy to power satellites carrying dozens of TPU chips. Those satellites could then be linked together into clusters, allowing them to process increasingly large AI workloads.
Google Senior Director, Paradigms of Intelligence Travis Beals explained the idea as an attempt to work backwards from a seemingly unrealistic proposition. “This idea, that’s crazy, there must be a reason it won’t work. And so the first thing we tried to do is to find that reason. It turns out that it is possible. At the heart of it, Project Suncatcher is a simple idea. How can we efficiently scale AI infrastructure through orbital data centres by leveraging the abundant energy that we can get from the sun?”
Google considered several types of orbits before focusing on the dawn-dusk orbit, where solar panels can receive sunlight for almost 24 hours a day. “When we first thought about this idea, we had looked at a variety of orbits, even geosynchronous orbit. But the dawn-dusk orbit is really optimal for what we’re wanting to do, because the solar panels in that orbit will get almost 24/7 sun,” Beals explained.
First, the chips have to survive
Getting an AI chip into orbit is considerably harder than getting it to work inside a terrestrial data centre. The journey itself can be brutal. A rocket reaches low Earth orbit in roughly 10 minutes, subjecting the spacecraft to intense vibration and acceleration. Google says the spacecraft can experience forces of up to 10 times Earth’s gravity, while individual components can face forces as high as 50 to 100 g. Google tested its satellite by shaking it across three axes to reproduce the conditions expected during launch. The company said the hardware survived the tests.
Radiation is another concern. Without Earth’s atmosphere providing the same level of protection, electronics in orbit are exposed to solar events and cosmic rays. These can cause errors in computing, including bit flips, where individual bits of data change unexpectedly.
Google exposed its Trillium TPUs to proton beams at the University of California, Davis’ Crocker Nuclear Laboratory while running AI workloads. The company said the chips survived a radiation dose greater than what they would be expected to encounter during a five-year space mission. The real test, however, will begin once the chips are actually in orbit.
The space race for compute
Google’s experiment comes as other technology companies explore the same idea. SpaceX and Starcloud are pursuing orbital computing projects, with the broader goal of moving some AI infrastructure away from Earth. Starcloud sent an Nvidia H100 GPU into orbit in November 2025, while SpaceX has also discussed putting AI computing hardware in space.
Musk has offered perhaps the most sweeping prediction about the direction of the industry. “The amount of compute in space will obviously round up to 100% of all compute,” he said while sharing Google’s video. In a September 13 post on X, Musk also said, “I am highly confident that SpaceX will be launching Nvidia VR NLV72 AI computers in space next year.”
The commercial case, however, remains far from settled. Launch costs, satellite manufacturing, radiation protection, cooling and communications all add layers of complexity that terrestrial data centres do not face. That makes Google’s first Suncatcher mission less about proving that orbital data centres are ready and more about finding out what could stop them from working.
“Big breakthroughs happen when you work backwards from an end goal. In our case, it’s to ensure AI’s profound benefits in key areas, from healthcare to scientific discovery, can reach everyone, far into the future,” Google said. The company described the October mission as a first step in a much longer experiment. “This first launch is about seeing what works, identifying points of failure, and applying those findings to future missions,” Google said.
The problem of heat
Even if the hardware survives the launch and radiation, there is another problem that looks surprisingly ordinary: heat. AI chips produce substantial amounts of heat. On Earth, data centres rely on airflow and other cooling systems to prevent processors from overheating. In the vacuum of space, there is no air to move that heat away.
Google is testing heat pipes and radiators instead. “We’re working on a number of different approaches for this, including a combination of heat pipes and radiators to cool the chips in space,” the company said. The system has already been tested inside a thermal vacuum chamber designed to replicate the conditions of space. The October mission will show how the technology performs outside the laboratory.
Building an orbital AI network
A single satellite would have limited computing power. Google’s larger ambition depends on putting many of them together. Future versions could carry dozens of TPUs and operate in clusters. Those satellites would need to communicate with each other at extremely high speeds, which is where laser communication comes in.
The challenge is that both satellites would be moving while trying to maintain a highly precise connection. Existing space-based laser systems are generally designed for lower bandwidth over longer distances. Google’s proposed network would need very high bandwidth across much shorter distances.
Google plans to put two satellites into orbit in 2027 to test those connections. “At its smallest level, you’ll have an individual satellite that will have dozens of TPU chips. And then, to scale to solve larger problems, we’ll have many such satellites linked together and flying in clusters…The sun is the ultimate energy source in our solar system. Once you start looking at solar power and think, “How can I scale that? How can I make that as simple and efficient as possible?”, doing it in space, in the right orbit, becomes the natural answer,” Beals said.