Google Tests Running AI in Space With New Satellite

Google Tests Running AI in Space With New Satellite

Google has launched a refrigerator-sized satellite carrying four Tensor Processing Units (TPUs) as part of a research effort called Project Suncatcher. The goal is straightforward but ambitious: test whether AI computing can actually work in space.

What Is the Satellite Actually Doing?

The satellite will run a version of Google’s Gemma AI model, but only for about 15 minutes at a time. That short window isn’t a software limitation, it’s because of heat limits. Cooling electronics in space is a fundamentally different challenge than cooling them on Earth, and Google is using this mission to understand just how much that limits AI workloads.

Beyond heat, Google is also testing how its chips hold up against:

  • Launch stress
  • Radiation
  • Extreme temperatures in orbit

These are the kinds of conditions that don’t exist in a typical data center, so this mission is essentially a stress test for hardware that was never originally designed with space in mind.

Why Is Google Interested in Space Data Centers?

The bigger vision here is genuinely bold. Google eventually wants to build clusters of satellites that work together as orbital data centers, powered entirely by solar energy.

Here’s how that might work in practice. Each future satellite could carry dozens of TPUs and communicate with other satellites and Earth using lasers instead of traditional radio signals. Google plans to launch two more satellites next year specifically to test these laser-based connections between spacecraft.

Google Tests Running AI in Space With New Satellite

How This Prototype Came Together

The current satellite was developed in partnership with satellite company Planet and launched aboard a SpaceX rocket. The mission is expected to run for about a year, giving Google a reasonably long testing window to gather data on chip performance and durability.

The Real Challenges Ahead

This isn’t a simple engineering problem, and Google is upfront about that.

Cooling is described as one of the biggest hurdles. Traditional cooling methods, whether air-based or water-based, simply don’t work in the vacuum of space. To work around this, Google is testing pipes and radiators designed to pull heat away from the chips without relying on either method.

Beyond cooling, there are other practical issues:

  • Repairs would be far harder than in a normal, Earth-based data center. You can’t exactly send a technician to fix a broken chip in orbit.
  • Launches remain expensive, which affects how quickly or cheaply this kind of infrastructure could scale.

Google has also been clear about managing expectations here. The company says space-based data centers are unlikely to become cheaper than terrestrial facilities within the next five years. That’s a notably honest admission for a company showcasing new technology, and it suggests this is very much a long-term bet rather than a near-term product.

Google Isn’t the Only One Exploring This

A few other companies are working on similar ideas, suggesting this isn’t just a one-off experiment but a genuine emerging area of interest in the tech industry.

  • Starcloud previously launched an Nvidia H100 chip into space
  • SpaceX has said it could begin deploying orbital AI compute satellites as early as 2028

How Google Frames This Project

Google itself describes Project Suncatcher as a long-term “moonshot”, language the company has used before for ambitious, experimental research projects that may take years to pay off, if they pay off at all. Importantly, Google isn’t positioning this as something expected to replace Earth-based data centers anytime soon.

Why This Matters

AI’s growing computing demands have made data center capacity, power consumption, and cooling genuine bottlenecks for companies like Google. While space-based computing sounds almost like science fiction, the underlying motivation is practical: solar energy in orbit is essentially unlimited and doesn’t compete with terrestrial power grids, something increasingly strained by AI’s energy demands here on Earth.

Whether this ever becomes commercially viable is still an open question, and Google’s own five-year cost estimate suggests patience will be required. But with multiple companies now testing similar concepts, this is clearly becoming a real area of research rather than just a one-off publicity experiment.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *