Google will send its first experimental satellite into orbit on October 1 as part of Project Suncatcher, an initiative to investigate whether data centers that run artificial intelligence workloads can be located in space. According to the New York Times, the mission will not deploy a full space data center but rather a small-scale experimental device.
Technical characteristics and mission profile
The refrigerator-sized satellite carries four Tensor Processing Units (TPUs). Google says the combined compute power of these four TPUs is roughly equivalent to a single data-center server and will be used to process simpler AI queries, such as short requests for the Gemini system.
The satellite's solar panels produce about one kilowatt of power — roughly the consumption of a hair dryer — which is sufficient for basic measurements and modest AI tasks. Google plans to launch the satellite into Earth orbit from the Vandenberg Space Force Base in California on a SpaceX Falcon 9 rocket. The mission is planned to be actively operational for one year, although the satellite could remain in orbit for up to six years before reentering the atmosphere and burning up.
Engineering challenges: radiation and cooling
Radiation hardening was one of the primary challenges. In February 2025, Google tested circuits at the Crocker Nuclear Laboratory in California using a cyclotron, exposing them to a dose equivalent to five years of space radiation. Radiation-induced single-event upsets (bit flips) were largely mitigated by restarting the chips, an encouraging result.
Cooling posed another significant engineering problem because conventional fans cannot be used in vacuum. Google developed a multilayer thermal-management system: the chips are mounted on a printed circuit board and the heat is conducted away through thermal interface materials, aluminum and copper layers, and finally into a radiator panel that dumps heat to space. With this design the chips can operate continuously for about 15 minutes before needing to pause to cool down.
Development background and participants
The idea for the project came from Blaise Agüera y Arcas, Google vice president and AI researcher, about three years ago after he encountered the growing energy demands of AI at a conference. James Manyika, Google’s vice president for research, was initially skeptical but later agreed to feasibility tests. In 2025, CEO Sundar Pichai and co-founder Sergey Brin approved the initiative, accelerating development.
Planet Labs, a long-time Google partner, built the satellite. Although the original schedule called for first units to launch in 2027, Google pressed for an earlier launch and integrated the chips into one of Planet Labs’ existing satellite platforms.
Costs, next phases and timeline
Google has not disclosed the project's budget. The company projects that with continued declines in launch costs, orbital data centers could approach the cost levels of terrestrial facilities by the mid-2030s. As next steps, Google plans to launch two more satellites next year and is developing a constellation of more than 80 units. That planned fleet would fly in tight formation and use direct inter-satellite communications to process AI queries. Longer-term plans include building a custom, football-field-sized satellite.
Practical significance
The upcoming launch is primarily a technology demonstration to test how radiation tolerance, cooling and power supply can work for AI chips in space. Many analyses note that realizing very large-scale space data center fleets — such as the multi-million-unit vision sometimes discussed by figures like Elon Musk — faces substantial engineering and economic hurdles and is more likely decades away than imminent.
The October 1 launch will provide Google with real-world results on core technical questions that will inform subsequent phases and longer-term strategy.
