“Our TPUs are headed to space,” wrote Sudar Pichai, chief executive of Alphabet Inc, Google’s parent, while launching Project Suncatcher—an experiment to run artificial intelligence data centres in space.
Google is launching two solar-powered satellites early 2027, each carrying four tensor processing units (TPUs)—the company’s proprietary AI chips which could run AI model training using power directly from the Sun.
The project is part of Google’s Moonshot philosophy, a radical idea which pushes the boundaries of space and engineering and often takes decades. For instance, Google envisioned autonomous vehicles 15 years ago which eventually became Waymo and has today completed 10 million trips without a single hazard.
Similarly, a space-based data centre could be a reality by 2030.
The advantages are radical—constant sunlight, minimal atmospheric loss, no nightfall and vacuum-based cooling that could cut energy costs by up to 40% versus Earth-based data centres, the company said in a blog on Wednesday.
In the right orbit, solar panels could be up to eight times more productive than on Earth, long durations of sunlight remove the need for heavy batteries, and cooling in space could significantly reduce energy and infrastructure costs.
Google’s math suggests that by 2030, the cost per kilowatt-year of operating space-based AI infrastructure could match that of terrestrial facilities.
After major technology breakthroughs in quantum computing and autonomous driving, Google is now headed for off-planet computing, cementing its position as a deeptech research powerhouse. Google brought home seven Nobel Prizes and surpassed IBM in 2025 to become the No.2 company housing Nobel laureates, behind Bell Labs.
However, there are major engineering hurdles to overcome.
These include building ultra-fast optical links to connect satellites at terabit-per-second speeds, maintaining precise formations with satellites flying just hundreds of meters apart, and ensuring that compute hardware like TPUs can survive harsh space conditions and radiation over multi-year missions.
Google plans to begin with a “learning mission” in cooperation with commercial partner Planet Labs, launching two prototype satellites early in 2027 to validate the hardware and optical communication systems in orbit.
The vision is shared by billionaire Elon Musk who stated that his company, SpaceX, and their next-gen V3 Starlink satellites could serve as a foundation for space-based data centres. Other startups like Starcloud, Axiom Space and Lonestar Data Holdings are also working in this direction.
Google is launching two solar-powered satellites early 2027, each carrying four tensor processing units (TPUs)—the company’s proprietary AI chips which could run AI model training using power directly from the Sun.
The project is part of Google’s Moonshot philosophy, a radical idea which pushes the boundaries of space and engineering and often takes decades. For instance, Google envisioned autonomous vehicles 15 years ago which eventually became Waymo and has today completed 10 million trips without a single hazard.
Similarly, a space-based data centre could be a reality by 2030.
The advantages are radical—constant sunlight, minimal atmospheric loss, no nightfall and vacuum-based cooling that could cut energy costs by up to 40% versus Earth-based data centres, the company said in a blog on Wednesday.
In the right orbit, solar panels could be up to eight times more productive than on Earth, long durations of sunlight remove the need for heavy batteries, and cooling in space could significantly reduce energy and infrastructure costs.
Google’s math suggests that by 2030, the cost per kilowatt-year of operating space-based AI infrastructure could match that of terrestrial facilities.
After major technology breakthroughs in quantum computing and autonomous driving, Google is now headed for off-planet computing, cementing its position as a deeptech research powerhouse. Google brought home seven Nobel Prizes and surpassed IBM in 2025 to become the No.2 company housing Nobel laureates, behind Bell Labs.
However, there are major engineering hurdles to overcome.
These include building ultra-fast optical links to connect satellites at terabit-per-second speeds, maintaining precise formations with satellites flying just hundreds of meters apart, and ensuring that compute hardware like TPUs can survive harsh space conditions and radiation over multi-year missions.
Google plans to begin with a “learning mission” in cooperation with commercial partner Planet Labs, launching two prototype satellites early in 2027 to validate the hardware and optical communication systems in orbit.
The vision is shared by billionaire Elon Musk who stated that his company, SpaceX, and their next-gen V3 Starlink satellites could serve as a foundation for space-based data centres. Other startups like Starcloud, Axiom Space and Lonestar Data Holdings are also working in this direction.
