Artificial intelligenceNews

Google tests AI chips in space: what the Project Suncatcher satellite will check

Google is running the first orbital test of Project Suncatcher. Why the satellite carries four TPUs and why this is not a space data center yet.

A satellite against the Earth for a story about Google’s Project Suncatcher

At a glance

  • Google is sending a satellite with four TPU accelerators into orbit
  • Engineers will check heating, power use and the chips’ radiation tolerance
  • It’s a research test, not a cloud service from space

Google is running the first orbital test of Project Suncatcher, a research program in which the company is trying to find out whether some AI computing could one day move to space. A small satellite carrying four of Google’s own TPU accelerators is due to head to orbit today.

This is not a “space data center” in the usual sense, nor a new cloud service. The experiment is meant to check how specialized chips behave in a vacuum, under radiation and with difficult heat dissipation. The results will decide whether the idea is worth developing further.

What exactly Google is testing

The prototype carries four TPUs, the processors Google designed for machine learning workloads. The team will monitor the chips’ stability, power consumption, temperature and the effects of space radiation. In ground-based data centers these issues are handled with air conditioning, backup power and hardware maintenance. On a satellite, any failure is much harder to fix.

Another goal is to see how realistic the architecture itself is. Suncatcher is built around compact satellites, solar power and optical links between spacecraft. In theory, such a network could operate where sunlight is available almost all the time. In practice, it will have to overcome problems with cooling, launch costs, communication delays and the safe retirement of satellites.

Why move data centers to orbit at all

  • Energy. Solar panels in space can get more sunlight than on the Earth’s surface.
  • Growing AI compute. Training and running large models takes ever more electricity and cooling.
  • New limits on the ground. Building large data centers increasingly runs into constraints on grid access, water and land.

Orbit doesn’t eliminate resource use, though; it changes what the resources are spent on. A satellite has to be designed, launched, kept in contact with Earth and safely decommissioned. That’s why Google’s near-term goal is not to replace data centers on Earth but to get real engineering data.

What happens after the launch

First, the satellite will go through commissioning: communications, power and the main systems must confirm normal operation. Then engineers will start comparing the data with ground-based calculations. Heating and the TPUs’ radiation tolerance will be especially important. If the experiment succeeds, the project will have grounds for further, more complex missions with several spacecraft working together.

FAQ

Is this a real data center in orbit?

No. It’s a small test satellite with computing chips that helps assess whether the idea is viable.

Why put TPUs in space?

They are designed for machine learning tasks. The experiment will show whether such accelerators can work reliably in orbital conditions.

When will everyday AI services run from space?

No specific timeline has been given. The project is at an early research stage.

Was this article helpful?

Be the first to rate

Author

Vasyl Vasyliev

Has been working on AppMaxx since 2019, writing reviews of smartphones, laptops and other gadgets, game and movie roundups, and how-tos on Windows and apps.

All articles by this author