Google Just Put AI Chips in Orbit. The Real Story Is the Power Bill on the Ground.

A SpaceX Falcon 9 lifted off from Vandenberg Space Force Base on October 1 carrying 130 payloads on the Transporter-18 rideshare mission. One of them was the first physical piece of Google's Project Suncatcher: a prototype satellite, built with Planet, carrying four of Google's Trillium-generation TPUs, the chips Google uses for AI work.

What actually launched

The satellite is there to answer three questions the ground can't answer together: whether the chips survive launch, how they hold up against radiation, and how well the heat they make can be shed in orbit.

Google has already tested part of this on Earth. Its Trillium TPUs survived a radiation dose greater than they would receive on a five-year space mission, in testing at UC Davis's Crocker Nuclear Laboratory. Launch is its own problem: Google says components see sustained acceleration up to 10 times the force of gravity, and individual chips face 50 to 100 g.

The bigger test comes in 2027, when Google plans to put two satellites in orbit together. Future designs would each carry dozens of TPUs, flying in clusters and linked by lasers that need very high bandwidth over very short distances.

Why go to space at all

That matters because energy is now one of the hard limits on AI growth. Google's research suggests launch costs could drop below $200 per kilogram by the mid-2030s, seven or eight times cheaper than today. That would put construction costs on par with some equivalent facilities on Earth.

Google itself calls Suncatcher a long-term research moonshot.

What this means for B2B leaders

• Nothing about your 2027 cloud bill changes because of this launch. Four chips in orbit are a test, not capacity. Don't plan around space compute.
• The signal is the constraint, not the rocket. When the largest cloud providers start testing orbit as a power source, it tells you how tight power and capacity are on the ground. Expect AI infrastructure to stay expensive and capacity to stay scarce for the next few years.
• Energy is becoming part of the AI vendor conversation. Ask cloud and AI vendors where their capacity comes from, how exposed their pricing is to power costs, and what their plans are for the next two years of growth.
• Watch the 2027 test. The laser links between satellites are what any real orbital cluster depends on. If two satellites can work together as one system, this moves from a science project toward something buyers might one day evaluate.

The bottom line

Project Suncatcher's first satellite won't run anyone's workloads. But the fact that Google is spending real money to find out whether AI compute belongs in orbit is the clearest sign yet that the limit on AI isn't chips or models. It's power.