Google Is Sending AI Chips Into Orbit on a SpaceX Rocket

iEXExchanger
Google Is Sending AI Chips Into Orbit on a SpaceX Rocket

On October 1, SpaceX will launch a Google satellite carrying four Trillium TPU chips — the first real flight test of Project Suncatcher, which explores whether AI computing could run on solar power in orbit.

On October 1, a SpaceX rocket lifting off from Vandenberg will carry an unusual passenger among dozens of satellites sharing the ride: a prototype loaded with four of Google's own Trillium chips, the same silicon the company uses to train its models on the ground. The mission, part of the Transporter-18 rideshare, marks the first real flight test of Project Suncatcher, an effort Google first unveiled last November.

The pitch is blunt: data centers on Earth are running out of power, water and land to expand, while a satellite in the right orbit catches near-constant sunlight and can pull up to eight times more energy per square meter than a panel bolted to the ground. Less downtime means fewer batteries, which means a lighter, simpler spacecraft. Google built the test satellite together with Planet, the imaging company that has spent years perfecting small, cheap orbital platforms.

Before launch, the chips went through a punishing rehearsal: ten minutes of acceleration reaching 10g at liftoff, with some components briefly hitting 50 to 100g; a proton beam at UC Davis that delivered more radiation than five years in orbit would; and a vacuum chamber to see whether heat pipes could shed warmth with no air to carry it away. Google says the hardware came through intact.

What comes next is about coordination, not just survival. In 2027, Google plans to fly two satellites and test a laser link between them — the kind of precise alignment a real compute cluster would need if dozens of TPU-carrying satellites are ever going to act as one machine instead of many separate boxes in formation. SpaceX is chasing a similar bet through its own SpaceXAI unit, also aiming for 2027, so orbital computing already has more than one company racing toward it.

Whether the solar-power advantage actually outweighs the cost of launching, cooling and maintaining chips in a vacuum is still an open question — Google itself calls this a research moonshot, not a product. The company has bet big on moonshots before and lost some, Loon's stratospheric balloons among them. For now, this is one satellite and a year of data.

Questions and answers

Frequently asked questions about this article

What is Project Suncatcher?

It's a Google research effort exploring whether AI computing infrastructure could run on satellites powered by solar energy in orbit.

When is the satellite launching?

A prototype carrying four Trillium TPU chips is set to launch on October 1, 2026, aboard a SpaceX rocket as part of the Transporter-18 rideshare mission from Vandenberg.

Why does Google want to compute in space?

Satellites in the right orbit get near-constant sunlight and can generate up to eight times more solar power per square meter than panels on the ground, which could ease the power crunch facing AI data centers.

Is anyone else pursuing orbital computing?

SpaceX is developing a similar concept through its SpaceXAI unit and also aims to launch its first orbital compute satellites in 2027.

Are there doubts about whether this makes economic sense?

Yes — Google itself calls Suncatcher a research moonshot, not a product, and it's unclear whether the solar-power gains will outweigh the cost of launching, cooling and maintaining chips in a vacuum.