Story
October 2, 2026
Google Put AI in Orbit—Now It Needs 1,800 Starship Launches
Google’s first TPU-carrying satellite has reached orbit, turning Project Suncatcher from a paper moonshot into a live experiment. But the company’s vision of orbital AI data centers still hinges on vastly cheaper launches and hardware that can survive space.
Before Thursday’s launch, Alphabet cast Project Suncatcher as a long-term test of whether space could host scalable machine-learning infrastructure. Its case is straightforward: satellites in low Earth orbit could draw near-constant sunlight and eventually work together on larger AI jobs, avoiding some of the power and land pressures confronting terrestrial data centers. But even the company acknowledged that it did not yet know how its chips would perform in orbit.1
Then Falcon 9 lifted off from California on the Transporter-18 rideshare mission, carrying a Planet-built solar-powered prototype fitted with Google TPUs. The mission put four TPUs into orbit, marking the first hardware step for Suncatcher.
2 Google chief executive Sundar Pichai called the result evidence of progress — and a reminder of the unfinished work: “Today’s successful launch with our partner @planet prototype satellite shows how far we’ve come...and how far there is to go.”
3
The next phase is deliberately modest. Once commissioned, the satellite is expected to run its TPU in 15-minute bursts, testing continuous power, cooling and performance under real orbital conditions. Project leader Travis Beals put the rationale bluntly: “There’s no test that’s completely as good as the real thing.”4
That experiment also exposes the gulf between a prototype and an orbital data center. Google envisions clusters of 81 satellites processing in parallel, but its analysis suggests cutting launch costs to roughly $200 a kilogram by 2035 could require Starship to deliver 370,000 tons to orbit — about 1,800 launches in a decade. Cooling, radiation tolerance, debris and constrained launch capacity remain equally stubborn constraints.
SpaceX has promoted “supercompute in space” as a nearer-term frontier, while Google is framing Suncatcher as a research moonshot. For now, both visions rest on the same unresolved wager: rockets must become dramatically more frequent and cheaper before the sky can become AI’s next data-center campus.