Google and Blue Origin Schedule 2029 Launch of TPU Compute Satellites on New Glenn

Google and Blue Origin have announced that Blue Origin's New Glenn rocket will launch satellites carrying Google's Tensor Processing Units (TPUs, the custom AI accelerator chips Google designs for its own data centers) into space in the first quarter of 2029. The launch is part of Project Suncatcher, Google's effort to put AI compute hardware in orbit. The Verge reported the schedule, and Blue Origin's own announcement describes New Glenn as the vehicle selected for the program.
The mission is planned as the first dedicated heavy-lift launch for Project Suncatcher. The satellites on that flight are expected to carry computing capacity at roughly the scale of a data center rack, and they are slated for a sun-synchronous orbit. In that orbit, the spacecraft's orbital plane stays at a fixed angle to the Sun, so each pass takes it over a given point on the ground at roughly the same local solar time. That gives the solar arrays a fairly consistent illumination profile, which matters for a power-hungry payload.
The New Glenn flight is the next step in a program that has already reached orbit in a smaller form. Google deployed its first AI server satellite in October, and its TPUs booted successfully once in space. That satellite launched aboard SpaceX's Falcon 9 as part of a rideshare with several other payloads. After reaching orbit, it used its on-board Gemma 4 model to transmit an AI-generated haiku to Earth.
The contrast between those two flights is the useful detail. A rideshare places a satellite on someone else's schedule, stacked among payloads with their own requirements. A dedicated heavy-lift launch means the vehicle is organized around the Suncatcher hardware, so the orbital configuration can be set for that hardware rather than for a shared manifest. The 2029 date gives the two companies roughly two years to build and integrate the satellites on the announced timeline.
The broader context here is why anyone would move compute off the ground at all. Terrestrial data centers face constraints on power delivery, cooling water, and land, and the orbital case rests on different trade-offs. Rejecting heat in a vacuum is not the same engineering problem as rejecting it through a chilled-water loop, and the cost of reaching orbit has to be weighed against the cost of building and powering capacity on the ground. The announcement settles none of those trade-offs, and the companies have not published a cost figure or an operating-life figure.
In my view, the most useful way to read the announcement is as a test of hardware and operations rather than a finished service. The October satellite showed that TPUs can power up in orbit and run a model there. The 2029 launch will have to show that a rack-scale compute payload can be built, launched, and operated for a useful period in sun-synchronous orbit. Those are the milestones the schedule actually commits to, and they are the ones worth tracking.
The optimistic reading is that a heavy-lift vehicle could eventually carry compute to places where power and cooling are cheaper or easier to obtain, and that a satellite running a model can return results to Earth within minutes. The haiku was a small demonstration, but it covered the full path from inference in orbit to a ground receiver. Whether orbital compute becomes a practical part of the infrastructure stack will depend on what the 2029 flight delivers, and the schedule gives the industry a concrete date against which to measure it.


