Google to launch Suncatcher AI data center into space on October 1, taking aim at SpaceX’s orbital AI push

Google is about to send its AI infrastructure somewhere data centers have never seriously operated before: orbit.

On October 1, the search giant plans to launch the first experimental satellite for Project Suncatcher, its effort to test whether AI computing infrastructure can eventually operate in space. The refrigerator-sized spacecraft will carry four Google Tensor Processing Units, or TPUs, and run Gemini workloads hundreds of miles above Earth.

The mission is tiny compared with a terrestrial AI data center, where thousands of accelerators can work together. Yet the implications are much larger. Google is testing a future in which AI compute could move beyond electrical grids, land constraints and water-intensive infrastructure on Earth.

The October 1 launch announcement is in Google’s September 24, 2026 Project Suncatcher update, titled “Behind Project Suncatcher, our moonshot to put AI in space.” Google says the first mission is meant to answer a simple question: whether its AI hardware can survive and operate in space.

The test, developed with Planet and scheduled to fly aboard SpaceX’s Transporter-18 rideshare mission, will collect real-world data on how Google’s TPUs respond to launch stresses, radiation and extreme temperature swings in orbit.

“Can our AI hardware operate in space?” Google wrote in announcing the mission.

The company said the early flight will help inform its next major Suncatcher milestone, still planned for 2027, as engineers work through the technical challenges of running AI computing hardware beyond Earth.

The race to move AI infrastructure off Earth

It puts Google into an emerging space-computing race that already includes Elon Musk’s SpaceX, which has outlined plans for orbital AI data centers and filed with the Federal Communications Commission for a satellite network that could eventually reach enormous scale. SpaceX has discussed an initial AI1 satellite architecture capable of roughly 120 kilowatts of sustained compute, far beyond Google’s first experiment.

One twist: Google’s first orbital AI experiment is getting there aboard a SpaceX rocket.

Google first unveiled Project Suncatcher in November 2025, pitching the research effort as a way to explore whether networks of solar-powered satellites carrying its TPU chips could eventually support large-scale AI computing in orbit.

At the time, Google said it planned to launch two prototype satellites with Planet in early 2027. The October 1 mission moves that first hardware test forward by several months.

Google Project Suncatcher

The Suncatcher prototype is scheduled to ride into low Earth orbit on SpaceX’s Transporter-18 mission aboard a Falcon 9. Google worked with satellite-imaging company Planet on the spacecraft, choosing to install its AI hardware aboard an existing satellite rather than wait for the custom satellites originally planned for 2027.

“Our next step is a learning mission in partnership with Planet to launch two prototype satellites by early 2027 that will test our hardware in orbit, laying the groundwork for a future era of massively-scaled computation in space,” Google said in a blog post.

Four TPUs, one kilowatt and a very big experiment

The satellite, called MVP, carries four of Google’s custom AI accelerators. Its solar panels are expected to provide roughly one kilowatt of electricity, a tiny amount by modern data center standards. The goal is less about useful commercial compute and more about finding out whether conventional AI chips can survive and operate beyond Earth.

Google has already subjected its TPUs to vibration and radiation testing on the ground. According to the company, components can experience forces reaching 50 to 100 g during launch. Google tested Trillium TPUs with proton beams at UC Davis and found that the chips survived radiation exposure exceeding what they would be expected to receive during a five-year space mission.

Orbit introduces another problem that may prove harder: heat.

Space is cold, yet cooling electronics there is surprisingly difficult. No air carries heat away from processors. Google is using heat pipes and radiators to move heat from the TPUs and release it into space. For this first mission, the cooling system has a major limitation. Google expects the chips to run Gemini workloads for roughly 15 minutes at a time before shutting down so the thermal system can recover.

That limitation gets to the heart of the orbital data center challenge. Putting AI chips in space is one problem. Keeping large numbers of them running continuously is another.

Testing AI chips to survive in space

Project Suncatcher was announced in November 2025 as a research effort built around constellations of solar-powered satellites equipped with Google TPUs and linked through high-speed optical connections. Google says solar panels in favorable low Earth orbit could generate up to eight times as much energy as comparable panels on Earth.

Future Suncatcher satellites could carry dozens of TPUs and communicate through laser links, creating distributed computing clusters in orbit. Two more prototype satellites remain planned for 2027.

Google is far from alone in chasing the idea. SpaceX has made orbital computing part of its longer-term AI strategy and has described satellites carrying large amounts of compute, solar arrays, radiators and optical links. The company’s existing experience launching and managing thousands of Starlink satellites gives it an infrastructure advantage few competitors can match.

Orbital data centers still face daunting economics and engineering questions, from launch costs and radiation to heat rejection and satellite replacement. The U.S. Government Accountability Office has identified cooling, orbital congestion, and collision risks as major barriers to the concept.

Google’s October 1 mission will not answer all of them.

It may answer the first question that matters: can the same AI chips filling data centers on Earth work reliably once they leave it?