Google is set to establish an artificial intelligence data center in space, with initial test equipment scheduled for launch into orbit in early 2027.

The company’s scientists and engineers are confident about deploying a densely clustered array of around 80 solar-powered satellites at approximately 400 miles above the Earth, each outfitted with robust processors to cater to the escalating AI demands.

Google highlights the rapid decline in space launch costs, suggesting that by the mid-2030s, operating space-based data centers may become as affordable as their terrestrial counterparts. The study was made public on Tuesday.

Utilizing satellites could significantly lessen the impact on land and water resources that are currently required for cooling ground-based data centers.

Once operational in orbit, the data center will harness solar energy and aim to achieve up to eight times the productivity of grounded facilities. However, launching a single rocket into orbit emits hundreds of tons of CO.₂.

Astronomers have expressed concerns about the burgeoning number of satellites in low-Earth orbit, describing it as “akin to a bug on a windshield” when observing the cosmos.

The envisioned data centers under Project Suncatcher would use optical links for data transmission, primarily leveraging light or laser beams.

Major technology firms aiming for swift advancements in AI are projected to invest $3 trillion (£2.3 trillion) in data centers worldwide, ranging from India to Texas and Lincolnshire to Brazil. This surge in spending raises alarms regarding the carbon footprint if sustainable energy solutions are not sourced for these facilities.

“In the future, space might be the ideal environment for advancing AI computing,” stated Google.

“In light of this, our new research initiative, Project Suncatcher, envisions a compact array of solar-powered satellites utilizing Google TPUs and linked through free-space optical connections. This strategy has significant scaling potential and minimal impact on terrestrial resources.”

TPUs are specialized processors designed for AI model training and routine use. Free-space optical connections enable wireless communication.

Elon Musk, who oversees satellite internet provider Starlink and the SpaceX rocket program, announced last week that his company would begin expanding efforts to develop data centers in space.

Nvidia AI chips are anticipated to be launched into space later this month in collaboration with startup Starcloud.

“Space provides virtually limitless low-cost renewable energy,” commented Philip Johnston, co-founder of the startup. “The environmental cost occurs only at launch, and over the lifespan of the data center, there’s a tenfold reduction in carbon dioxide compared to ground-based power.”

Google aims to deploy two prototype satellites by early 2027, referring to the research findings as “the first milestone toward scalable space-based AI.”

However, the company cautions that “substantial engineering challenges persist, including thermal management, high-bandwidth ground communications, and the reliability of systems in orbit.”