Google launches first orbital AI chip as it plots massive space data center ambitions

Google sent one of its Tensor Processing Units into space today aboard a SpaceX rocket, marking the company's first crewed test of advanced silicon designed to operate in orbit. The milestone represents a significant step toward the company's ambitious vision of building AI data centers floating above Earth.

Testing in the final frontier#

The satellite, constructed by Planet Labs, will validate whether Google's TPU—its answer to Nvidia's dominant GPUs—can actually function reliably beyond the atmosphere. The test involves powering the chip continuously, managing its heat dissipation, and running computational models to detect any anomalies.

"There's no test that's completely as good as the real thing," explained Travis Beals, who leads the initiative. The satellite will cycle the TPU on and off in 15-minute intervals to prevent overheating and power depletion. Google and Planet Labs plan to launch a more specialized demonstration mission next year featuring two purpose-built satellites that will communicate via laser links.

The massive infrastructure challenge ahead#

Google's long-term ambition, called Project Suncatcher, envisions a cluster of 81 satellites operating in coordinated formation to handle parallel AI workloads. Unlike competing space ventures, this effort targets infrastructure needs five or more years into the future, not immediate applications.

The company released a peer-reviewed analysis published in Joule that lays out the economic requirements for making orbital compute viable. A critical finding: achieving cost targets of roughly $200 per kilogram by 2035 would demand Starship to complete approximately 1,800 launches within a decade—translating to roughly 180 flights annually.

The launch cadence problem#

That projection rests on SpaceX delivering 200 metric tons to orbit per mission, allowing the company to accumulate the operational experience needed to drive down prices further. Historically, SpaceX has achieved roughly 20% annual cost reductions since the Falcon 1 era, according to Google's analysis.

Yet the math is daunting: SpaceX has never flown Starship more than five times in a single year. While Elon Musk has publicly suggested hourly launch cadences by 2029, those claims remain speculative. The gap between Google's infrastructure requirements and SpaceX's current proven capabilities represents one of the starkest hurdles facing the space AI sector.