Google and Blue Origin prepare AI servers in orbit
Google plans to launch its Suncatcher satellites with Blue Origin in 2029. The project targets AI computations in low Earth orbit, but cooling, chips, and laser links still need validation.
Google announced on November 19, 2026, an agreement with Blue Origin to launch, in the first quarter of 2029, the first dedicated satellites of Project Suncatcher aboard New Glenn. These devices will carry Tensor Processing Units, Google's acceleration chips, to a sun-synchronous orbit. The challenge is concrete: to verify whether AI computations can truly be performed in space, at the scale of a data center rack, without the whole system going into orbit... figuratively this time.
Un test de passage à l’échelle, pas encore un centre de données spatial
Project Suncatcher, presented by Google as a research avenue, is moving beyond the isolated prototype stage. According to both sources, the Blue Origin launch is intended to serve as the first heavy mission dedicated to the program, following a first test satellite deployed in October 2026 during a shared flight on Falcon 9. This first device allowed starting its TPUs, running machine learning tests, and even producing a haiku generated by the Gemma 4 model, which Google describes as the first response from a data center-class TPU in orbit.
The next milestone is more ambitious. Google then mentions thermal and radiation stress tests, followed by a planned 2027 stage with two interconnected satellites. Only after that will the idea of a computing constellation gain more substance. At this stage, it is mainly a very serious technical demonstration, not a cloud that has risen too early.
Pourquoi l’orbite basse intéresse Google
The energy bet relies on low Earth orbit, where satellites benefit from almost continuous sunlight. Google claims they could produce up to eight times more solar energy there than on the ground. This abundance changes the game for power-hungry computing loads, provided the rest is mastered: heat, communications, and component stability.
Each Suncatcher satellite is designed to provide computing capacity comparable to that of a data center rack. Future versions could carry dozens of TPUs and exchange data via lasers to link several satellites in clusters. The principle is appealing on paper, but it depends on extreme positioning precision between moving devices. Without this, the space network quickly resembles a video meeting that drops out.
New Glenn apporte la charge utile, mais pas les réponses techniques
Blue Origin mainly brings payload capacity to the table. New Glenn is announced with a capacity of 45 tons in low Earth orbit and more than 13 tons to geostationary transfer orbit, as well as a 7-meter diameter fairing, twice as large as that of smaller launchers. Its first stage is designed for at least 25 flights. For Google, this volume and payload margin make multi-satellite deployments possible, where a shared flight was only sufficient for a proof of concept.
But the equation remains incomplete. In the vacuum of space, chips cannot rely on air to dissipate heat. Google says it has tested heat pipes and radiators in a thermal vacuum chamber. There is also the question of semiconductor reliability under radiation, often more punishing than a Monday morning update bug. Until these components are validated, the computer in orbit remains a promise, not an infrastructure.
Ce que cette alliance change pour l’IA spatiale
The partnership between Google and Blue Origin mainly counts as an industrial signal. It shows that the idea of computing centers in orbit is no longer solely a concept or a laboratory mock-up. A schedule exists, a launcher is identified, and a technical path is drawn between prototype, satellite duo, then larger deployments. For now, nothing guarantees that the architecture will scale. However, the question of space computing is now posed with dated milestones, precise physical constraints, and actors capable of dedicating significant resources.
If the program succeeds, it could interest the AI infrastructure market, launch operators, and more broadly, actors seeking to circumvent the energy pressure of large terrestrial data centers. If it fails, it will at least have clarified what is still missing: robust chips, reliable thermal dissipation, and stable laser links between satellites in formation. In both cases, the next step will count.
By the numbers
- November 19, 2026 — announcement of the Google-Blue Origin agreement.
- First quarter 2029 — targeted window for the dedicated launch.
- 45 tons — announced capacity of New Glenn in low Earth orbit.
- 7 meters — diameter of New Glenn's fairing.
- 8 times — potential solar gain claimed for low Earth orbit.