MOUNTAIN VIEW — Google launched a prototype orbital compute satellite today aboard a SpaceX rocket, marking the first deployment of its Tensor Processing Unit into space. The satellite, built by Planet Labs, will test the TPU's ability to operate with one kilowatt of continuous power, manage cooling, and execute models in orbit.
The initial mission will activate the TPU in 15-minute bursts to avoid straining power and thermal systems. Travis Beals, Google's executive leading Project Suncatcher, said ground testing cannot replicate real-world space conditions.
Next year, Google and Planet Labs plan a second mission with two satellites linked by laser communications—a critical capability for parallel processing at scale.
Project Suncatcher envisions an orbital data center of 81 satellites flying in close formation. Beals emphasized that bandwidth and latency between TPUs matter for multi-rack workloads, and Google is planning five years ahead.
Scaling this concept requires launch economics that do not yet exist. According to a Google white paper published in the peer-reviewed journal Joule, SpaceX's Starship would need to complete 1,800 launches over the next decade—averaging 180 per year at 200 metric tons per mission.
That cadence represents a structural gap: Starship has never flown more than five times in a single year. Google, a major SpaceX investor, is betting the company can improve dramatically. The paper lays out Google's assumptions about declining launch costs but is not an economic feasibility study.
For Google's orbital AI vision to work, the math has to change first.
