Google Books Blue Origin's New Glenn Rocket for Orbital AI Chips in 2029
Google has booked a New Glenn rocket from Blue Origin to launch satellites carrying TPUs, its own AI chips, in Q1 2029. The idea behind Project Suncatcher is to draw power from the sun: Google estimates that solar panels in space can generate up to eight times more than on Earth. But the rocket hasn't flown since May, when it exploded during a test, and cooling and communications remain unresolved. That makes 2029 a test of the idea, not the start of a business.
AI-processed from TNW; edited by Hamidun News
AI data centers are running into electricity limits: the more models are trained and run, the more energy they consume. Google is proposing an unusual way out by moving part of the computing into space, where the sun shines almost without interruption. The idea now has a date. The company has agreed with Blue Origin that the heavy-lift New Glenn rocket will launch the first Project Suncatcher satellites in Q1 2029. On board will be TPUs, tensor processing units, Google's own AI chips.
What will actually fly
Suncatcher is Google's research project for placing AI data centers in orbit. The satellites will run on solar panels and carry TPUs. By design, each spacecraft should hold computing power roughly equal to one standard data center rack. For comparison, a typical data center has hundreds of such racks, so this is still a set of first building blocks rather than a full facility.
The satellites will fly in a sun-synchronous orbit. Its key feature is that the spacecraft stays on the sunlit side of Earth almost all the time. For solar panels, these are nearly ideal conditions. Google claims that in space they can deliver up to eight times more energy than on Earth. This is the company's own estimate, not the result of independent measurements.
The launch will be the first dedicated flight of a heavy rocket for Suncatcher. Neither Google nor Blue Origin has disclosed the contract value.
What has been done and what comes next
Planet built the project's first prototype. In October it reached orbit and ran Gemma 4, Google's open language model, which sent a haiku back to Earth. The result is largely symbolic. It shows that an AI model can in principle run on board, but it says nothing about cost or scale.
Google is now testing how its chips handle heat and radiation. The next milestone is set for 2027: the company plans to link two satellites with each other before the large New Glenn launch. The sequence is logical: first verify communications and chip reliability on small spacecraft, then send heavier ones.
Where the catch is
There are more unsolved problems than solved ones. The first is cooling. On Earth, servers are cooled with air or liquid, but in a vacuum there is no air, and heat can only be shed by radiation. For a rack of powerful chips, this is a serious engineering challenge. The second is communications. For the satellites to work as a single computing cluster, they must hold a tight formation and exchange data by laser. Any error in spacing and the link is lost.
There is also a risk tied to the rocket itself. New Glenn has not flown since May, when it exploded during a test. There is time before 2029 to investigate the failure and return to flight, but the schedule still depends on how confidently Blue Origin gets back to operation.
Google is not alone in this race. The startup Starcloud raised funding in March to build orbital data centers. Meanwhile SpaceX, Blue Origin's competitor, mentioned its own orbital data centers in its IPO documents and warned investors about the associated risks.
What this means in practice
For developers and product managers the takeaway is simple: in the coming years this is not an alternative to the cloud. No prices, latencies, or capacity figures have been announced, and the main technical questions remain open. But the fact that Google is booking a heavy rocket more than two years before launch shows how much energy has become a bottleneck for AI, and how willing large companies are to spend on the most exotic options.
Two milestones are worth watching: the two-satellite communications test in 2027 and New Glenn's return to flight. If both go smoothly, orbital computing will move from the category of a bold idea to an engineering problem with an economic case. If not, 2029 could easily slip.
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026MIRA Neural Network Renders Rocket League at 20 FPS, but Ping Hits 460 ms
- Nov 21, 2026Google built a closed copy of the internet with fake banks to train Chrome's AI agents
- Nov 21, 2026MLPerf Training Adds Nvidia Rubin and TPU Ironwood in First Appearance: 153 Systems
Key Questions Answered
What will actually fly?
Suncatcher is Google's research project for placing AI data centers in orbit. The satellites will run on solar panels and carry TPUs. By design, each spacecraft should hold computing power roughly equal to one standard data center rack. For comparison, a typical data center has hundreds of such racks, so this is still a set of first building blocks rather than a full facility. The satellites will fly in a sun-synchronous orbit.
What has been done and what comes next?
Planet built the project's first prototype. In October it reached orbit and ran Gemma 4, Google's open language model, which sent a haiku back to Earth. The result is largely symbolic. It shows that an AI model can in principle run on board, but it says nothing about cost or scale. Google is now testing how its chips handle heat and radiation. The next milestone is set for 2027: the company plans to link two satellites with each other before the large New Glenn launch.
Where the catch is?
There are more unsolved problems than solved ones. The first is cooling. On Earth, servers are cooled with air or liquid, but in a vacuum there is no air, and heat can only be shed by radiation. For a rack of powerful chips, this is a serious engineering challenge. The second is communications. For the satellites to work as a single computing cluster, they must hold a tight formation and exchange data by laser. Any error in spacing and the link is lost. There is also a risk tied to the rocket itself.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.