Google is taking a very literal approach to the question of where the next generation of AI infrastructure should live: space.

The company is preparing to send Google Tensor Processing Units (TPUs) into low Earth orbit as part of Project Suncatcher, an ambitious research effort exploring whether large-scale AI computing could eventually be built around networks of solar-powered satellites. The first orbital test is scheduled to launch next week aboard SpaceX’s Transporter-18 rideshare mission, in partnership with satellite company Planet.

This isn’t an orbital Google data center just yet. The immediate goal is much more basic—and arguably more important. Google wants to find out whether its AI hardware can actually survive and operate in space.

The company announced Project Suncatcher in November 2025 as a research moonshot built around a simple idea: put AI processors on satellites, power them with sunlight and connect the satellites using high-bandwidth optical links. Instead of concentrating enormous amounts of computing equipment in buildings on Earth, future AI infrastructure could potentially be distributed across constellations in orbit.

The appeal is largely about power.

In the right low Earth orbit, solar panels can receive sunlight far more consistently than panels on the ground. Google says a solar panel in orbit can generate up to eight times more power than one on Earth, while also spending much less time in darkness. That could make space an interesting location for the enormous energy requirements associated with AI computing.

But moving a data center into orbit isn’t as simple as bolting a TPU onto a satellite.

The first Suncatcher mission is designed to expose Google’s hardware to conditions that can’t be perfectly replicated on Earth. During launch, spacecraft experience intense vibration and acceleration. Google says a rocket journey into low Earth orbit lasts roughly 10 minutes, with the spacecraft experiencing loads of up to 10 times Earth’s gravity. Individual components such as TPU chips can experience substantially higher forces, potentially reaching 50 to 100 g.

Then there’s radiation.

Space exposes electronics to high-energy particles that can cause errors and damage hardware over time. Google has already performed radiation testing on its TPUs on Earth, but the company says an actual orbital mission is necessary to understand how the hardware behaves in the real environment.

Heat may be an even stranger problem.

On Earth, data centers rely heavily on airflow and conventional cooling systems to move heat away from processors. In the vacuum of space, there’s no surrounding air to carry that heat away. Google’s engineers are therefore experimenting with a combination of heat pipes and radiators to transfer heat from the AI chips and ultimately radiate it into space. The cooling system has already been tested inside a thermal-vacuum chamber designed to simulate the conditions the hardware will encounter in orbit.

If those challenges can be solved, Google envisions something considerably larger than a single experimental satellite.

The company’s earlier research describes a future system made up of compact satellites equipped with TPUs and connected through free-space optical communications. Rather than building one gigantic computer in orbit, computing could be distributed across a constellation of satellites working together as a single system.

That architecture could also change how AI infrastructure is built. On Earth, expanding AI capacity means finding suitable land, securing huge amounts of electricity, building cooling systems and connecting everything to increasingly demanding power grids. Space offers a completely different set of resources—and an entirely new set of problems.

Google is not suggesting that terrestrial data centers are about to disappear. In fact, the company is approaching Suncatcher as a long-term research project. The first mission is about collecting data and discovering what breaks before anyone attempts to scale the concept.

That distinction matters because an orbital data center would have to overcome more than radiation and thermal management. Launching hardware into orbit remains expensive, communications have to work across moving satellites, and the computing hardware itself evolves much faster than many spacecraft are designed to operate. Space infrastructure also has to contend with orbital congestion and the risk posed by debris.

Still, Google sees enough potential to keep pushing.

The company’s initial research points toward satellite clusters capable of handling increasingly large machine-learning workloads. The prototype mission with Planet is the next step toward that goal, while Google says it is working toward another milestone in 2027.

And Google isn’t alone in looking upward.

Space-based computing has become an increasingly interesting idea as AI companies face enormous demand for electricity and computing capacity. SpaceX and other companies are also exploring orbital data centers, with the broader industry looking at the possibility of using near-continuous solar power to support energy-intensive workloads.

For now, though, Project Suncatcher is less about launching a data center than answering a much smaller question: can Google’s AI chips survive the trip?

If the answer is yes, Google will have one of the most important pieces of information it needs to decide whether its much bigger vision is technically realistic.

The next Google data center, in other words, isn’t orbiting Earth yet.

But Google is getting ready to find out what happens when its AI hardware does.