Google has taken its first concrete step toward orbiting data centres, launching a test satellite equipped with artificial intelligence processors into low Earth orbit. The experimental spacecraft, part of an initiative dubbed Project Suncatcher, lifted off on 1 October aboard a SpaceX Falcon 9 rocket from Vandenberg Space Force Base.

The refrigerator-sized satellite, known as MVP, accelerates Google's timeline by utilising a spacecraft bus originally built by imaging firm Planet Labs. Inside, it houses four of Google’s standard Tensor Processing Units (TPUs)—the exact same custom hardware accelerators used to train and run models in the company's terrestrial server farms. Drawing on roughly one kilowatt of power from its solar arrays, the hardware yields a computing capacity comparable to a single data-centre server.

The long-term appeal of space-based compute infrastructure lies primarily in energy abundance. With terrestrial data centres projected to consume around three per cent of global electricity by 2030, technology giants are actively searching for ways to sidestep local grid constraints. In orbit, solar panels can generate significantly more continuous power than they do on the ground. For nations in the Asia-Pacific—including Aotearoa, where renewable energy is highly prized but grid capacity remains tightly managed—off-world computing could eventually provide access to massive AI capabilities without stressing regional power grids.

However, the harsh environment of space introduces severe engineering hurdles. The most immediate is thermal management. Space lacks the atmosphere required for the convective cooling systems standard in Earth-bound facilities. Instead, MVP must route heat through thermal-interface materials, aluminium, and copper heat pipes to a radiator panel that expels it into the vacuum. This setup currently limits the TPUs to running Gemini AI workloads in brief 15-minute bursts before the processors must shut down to shed excess heat.

The mission will also test how unmodified server hardware withstands orbital realities. Beyond surviving the intense vibrations and g-forces of a rocket launch, the TPUs face constant bombardment by cosmic radiation. High-energy particles can physically damage silicon or cause 'bit flips' that disrupt complex calculations. Determining whether standard chips can function reliably under these conditions without expensive, space-hardened redesigns is a primary objective for the coming months.

While Google hopes to launch two custom-built Suncatcher satellites by 2027 and eventually deploy large orbital constellations linked by high-speed lasers, operational space data centres remain years away. For now, the MVP mission serves as a critical proof of concept, establishing whether the fundamental building blocks of modern AI can survive off the planet.

Sources


AI disclosure: Automated systems assisted with research and drafting. Gemini independently peer-reviewed this article before publication. Sources are linked for verification.