[Space. Photo Credit to Pixabay]

On September 24th, Google announced the scheduled launch of its first outer space AI hardware as a part of Project Suncatcher, aiming to test whether large-scale AI servers could be sustained in space.

According to Google Research, Project Suncatcher, is “a moonshot exploring a new frontier: equipping solar-powered satellite constellations with TPUs and free-space optical links to one day scale machine learning compute in space.”

As Project Suncatcher’s inaugural step, with the support of Planet Labs and SpaceX, Google plans to fly a prototype satellite into low Earth orbit on October 1st.

This satellite will carry four specialized computer chips called Tensor Processing Units, or TPUs—the equivalent of a single AI server, rather than a full operational data center.

It will collect in-orbit data and identify potential obstacles regarding AI hardware's ability to withstand radiation and extreme temperatures in outer space.

The mission serves as an early test of what Google aims to do in the future: put many of these AI servers into orbit and create a giant operational data center powered by solar panels.

Google—and many other tech companies—aim to transition AI data centers into space due to the limitations of keeping them on Earth.

One concern is the possibility of running out of electrical power in the future.

Furthermore, Google vice president Blaise Agüera y Arcas voiced concerns about the environmental impact of electricity overuse.

“If you look at the exponent — how this is going to go over the next decade or two — it starts to look alarming,” he said.

Therefore, companies plan to position these data centers into a “sun-synchronous orbit” that keeps the satellites constantly in direct line with the Sun.

This orbit ensures the Sun can shine continually onto solar panels on these satellites, consistently powering the computer servers.

Elon Musk, founder of SpaceX, predicted that “A.I. in space” will be the most cost-effective way of powering the technology “not more than five years from now.”

However, deploying hundreds of AI servers into orbit presents its own set of challenges.

First of all, though companies believe they can push down costs, it won’t be cheap.

Additionally, when computer servers are launched into outer space at 17,000 miles an hour, the violent vibrations and G-forces can damage the hardware.

Moreover, radiation from cosmic rays in space can alter data and software in ways that current technology cannot always fix.

“You get a lot of weird particles out in space — high-energy radiation that we are just not exposed to on Earth,” said Brandon Lucia, a professor of electrical and computer Engineering at Carnegie Mellon University. 

“Sometimes, you get a random high-energy particle strike that is like someone throwing a dart at the insides of your computer chip,” he said.

Finally, Google must address the issue of cooling the hardware.

On Earth, cold air or liquid coolant is typically used to cool down overheated machines through a process called convection.

However, convection doesn’t work in a vacuum, meaning Google must build a specialized thermal management system that can radiate heat away from the hardware.

Google admitted they do not yet know how to make this level of technology function at scale, nor how to make it economically viable.

Some researchers believe that deploying a massive number of data centers in space could take decades.

Project Suncatcher’s next phase is scheduled for 2027, when two additional satellites will be launched.