Google is sending its first AI data center to space

Google is preparing to put its AI chips somewhere no data center has gone before: low Earth orbit.

The company’s Project Suncatcher prototype is scheduled to launch on October 1, 2026, aboard a SpaceX Falcon 9. The goal is simple to state and hard to pull off: find out whether Google’s custom AI hardware can survive, and actually work, in space.

Calling it a data center is a little generous. The satellite is roughly the size of a refrigerator and carries about as much computing muscle as a single server rack slot back on Earth. It starts with four chips.

What’s actually going up

The prototype goes by the name MVP. It will ride to orbit on SpaceX’s Transporter-18 mission, lifting off from Vandenberg Space Force Base in California.

Google built the spacecraft together with Planet Labs, the satellite company. Inside are four Google Tensor Processing Units, or TPUs. These are the custom chips Google designed specifically to run machine-learning workloads.

Combined, those four TPUs deliver compute equivalent to one terrestrial data center server.

Power comes from solar panels producing approximately 1 kW.

The operating plan reflects those limits. The satellite will run short AI inference jobs in brief bursts. Then it will lean on onboard cooling systems to shed the heat those jobs generate before going again.

Inference is the part of AI where a trained model actually answers questions or processes data. Training builds the brain, and inference puts it to work. MVP is focused on the second, lighter task.

Why space is a brutal place for chips

Radiation can flip bits, corrupt calculations, and degrade chips over time. Temperature swings between sunlight and shadow add thermal stress on top of that.

Google has already done some homework here. Ground testing showed its TPUs could withstand simulated radiation doses equivalent to more than five years in orbit.

The satellite is designed to operate for about one year. It could potentially stay in orbit for up to six years before it is deorbited. Google frames the whole effort as exploratory, meaning the main deliverable is data about what works and what breaks.

Cooling deserves special attention. On Earth, data centers rely on air and water to move heat away from servers. In the vacuum of space, there is no air to carry heat off, which is why the bursty operating schedule and onboard thermal management matter so much.

The bigger picture behind Suncatcher

Google first announced Project Suncatcher in 2025. The original plan called for a test launch in 2027. The company has since pulled that timeline forward to this October flight.

The underlying pitch is about energy. Running large machine-learning workloads on Earth requires enormous amounts of electricity, and power availability has become a real constraint on AI growth.

In low Earth orbit, solar energy can be up to eight times more abundant than it is on the ground. Panels in the right orbit avoid clouds, nighttime, and the atmosphere filtering out sunlight.

MVP is meant as the first piece of a larger plan. Google intends to launch two more satellites in 2027 to test high-bandwidth laser links between spacecraft.

What this means

For Google, Suncatcher is a long-horizon bet on two fronts at once: AI and space technology. The company is effectively testing whether part of AI’s power problem can be solved by leaving the planet rather than fighting over grid capacity on it.

The near-term stakes are technical, not commercial. If the TPUs perform as ground testing suggests, Google gains real-world evidence that its chips can handle orbit. If they falter, the company learns that lesson on a single refrigerator-sized prototype rather than an expensive constellation.

The milestones to watch are clear. First comes a successful October 1 launch and the satellite’s health in its early months. Then come the 2027 laser-link tests, which will show whether orbiting chips can work together rather than just survive alone.