How will Google deploy AI data centres in space? Here's the full plan

Google is preparing its first orbital test of Trillium TPUs under Project Suncatcher. The experiment will show whether space-based, solar-powered AI computing can scale despite radiation, heat and connectivity challenges.

In Short

• Planet partnership mission will assess TPU performance during radiation and spaceflight stress
• Low Earth orbit offers near-constant sunlight for far greater solar generation
• Earth-based AI facilities face pressure from power, land and cooling needs

Google is preparing to take artificial intelligence beyond Earth, with plans to deploy AI data centres in space that could eventually use solar-powered satellites to process massive amounts of data in orbit.

The technology giant's Project Suncatcher, announced last year, is moving towards its first orbital test, which will examine whether Google's Tensor Processing Units (TPUs), specialised chips designed to run AI workloads, can operate in the harsh environment of space.

The initial mission, developed in partnership with satellite company Planet, is scheduled to launch aboard SpaceX's Transporter-18 rideshare mission. It will test Google's Trillium TPUs in orbit, gathering data on their performance under radiation, extreme temperatures and the physical stresses of spaceflight.

Google's long-term vision is to connect multiple satellites carrying AI computing hardware, creating orbital infrastructure capable of handling increasingly complex AI workloads.

WHY GOOGLE WANTS AI DATA CENTRES IN SPACE

The idea is to use the abundant solar energy available in space to power AI computing.

According to Google, satellites in low Earth orbit could access near-constant sunlight and generate up to eight times more solar power than comparable installations on Earth.

As demand for AI computing grows, data centres on Earth face challenges involving electricity consumption, land requirements and cooling infrastructure. Google is exploring whether space-based computing could eventually help address some of these constraints.

However, the company acknowledges that turning the concept into a working system will require overcoming major engineering challenges.

SURVIVING RADIATION AND EXTREME TEMPERATURES

The first challenge is ensuring that AI chips can survive launch and operate reliably in orbit.

During a rocket launch, spacecraft experience intense vibrations and acceleration forces. Individual components, including TPUs, can experience forces of up to 50 to 100 times Earth's gravity.

Google said its chips underwent vibration testing and radiation exposure experiments at the University of California, Davis' Crocker Nuclear Laboratory. Initial results showed that its Trillium TPUs could withstand radiation doses exceeding those expected during a five-year space mission.

Cooling is another major hurdle. Unlike Earth, space has no air to carry away heat from electronic equipment. Google is developing systems using heat pipes and radiators to dissipate the heat generated by AI chips.

LASER LINKS TO CONNECT SATELLITES

Google eventually wants to place dozens of TPUs aboard individual satellites operating in clusters.

These satellites would communicate through high-bandwidth laser links, allowing them to share computing workloads. Maintaining precise connections between rapidly moving satellites will be a significant technical challenge.

Google plans to test this technology by placing two satellites in orbit in 2027.

The upcoming mission is an early experiment, not a deployment of operational orbital data centres. Google says the results will help determine whether space can become a scalable location for future AI computing infrastructure.

- Ends