Google Taps Blue Origin's New Glenn to Launch AI Data Centers Into Orbit
Google has signed a launch services agreement with Blue Origin to send rack-sized TPU data center satellites into sun-synchronous orbit by early 2029, betting orbital solar power can fuel AI growth.

Why it matters
- Google signed a launch services agreement with Blue Origin on November 19, 2026 for Project Suncatcher, with the first launch scheduled for early 2029.
- Each satellite carries roughly one data center rack of capacity, including multiple Google TPUs, launched on New Glenn rockets into sun-synchronous orbit.
- Google's first AI server satellite, launched in October to low-Earth orbit, can only operate for 15 minutes at a time.
- A single Falcon 9 launch costs about $74 million and can barely carry a small server room's hardware.
- Tracked objects in low-Earth orbit have grown from under 2,000 to over 47,000, raising Kessler syndrome concerns.
Google has signed a launch services agreement with Jeff Bezos' Blue Origin to put satellite data centers into orbit, with the first launch of its Project Suncatcher experiment scheduled for early 2029. Each satellite will carry Google Tensor Processing Units — the company's proprietary AI chips — plus data center servers, and will hold roughly the capacity of a single data center rack. Blue Origin's heavy-lift New Glenn rockets will carry the hardware into sun-synchronous orbit.
Blue Origin confirmed the deal on November 19, 2026, describing Project Suncatcher as Google's research moonshot exploring how an interconnected network of solar-powered satellites equipped with TPUs could scale machine learning in space.
The stakes are enormous, and Google knows it. Data centers on Earth are notorious resource hogs. They drain electricity grids, consume land, and demand vast cooling systems. They also make noise — endless industrial hums and high-pitched whirs, a daily reality for people living near data center facilities. Google calls the project a "long-term moonshot" to test whether AI infrastructure can operate in space by "tapping into the abundant solar power in orbit." The vacuum of space, the thinking goes, might simply be a better fit for the technology than a warehouse in Virginia.
What does the deal actually involve?
The launch services agreement covers the first Project Suncatcher deployments. New Glenn, Blue Origin's heavy-lift rocket, will carry several TPUs and data center servers per mission. The satellites will fly in sun-synchronous orbit — a path that keeps them in continuous sunlight, which matters when your power plan is solar.
Timeline matters here too. The first launch isn't scheduled until early 2029. That gives Google roughly two years of engineering work before anyone learns whether the concept functions at all outside a lab.
Google already has one data point. The company deployed its first AI server satellite back in October. That early hardware went into low-Earth orbit and can only operate for 15 minutes at a time, a constraint that illustrates just how far the project is from its goal of a functioning orbital data center network.
Can AI infrastructure actually survive in space?
The honest answer is that nobody knows yet. The obstacles are concrete and serious.
Cooling comes first. Standard heat dissipation doesn't work in space because there's no air to carry heat away. Waste heat has to move through pipes and panels instead of conventional cooling systems. Scaling those systems up from one experimental satellite to a network that behaves like a giant data center is, by most accounts, incredibly difficult.
Radiation comes second. It's everywhere in orbit, and it causes what engineers call a "bit flip" — flipping a binary digit from a one to a zero, or vice versa. For binary-coded systems, which is to say essentially all computing hardware, random bit flips are a major reliability problem. Servers running long AI training workloads in orbit would face this constantly.
"Expanding from one satellite to a vast network of them that operate like a giant data center will take years and enormous funds," Brandon Lucia, a professor of electrical and computer engineering at Carnegie Mellon University, told The New York Times. "If you do this on a large scale, there are additional engineering problems."
Then there's the launch math. Sending hardware to space is extremely expensive. A single Falcon 9 launch costs around $74 million, and even that rocket's maximum payload can barely cover the hardware for a small server room — let alone anything approaching a hyperscale data center. New Glenn exists precisely to carry more mass per launch, but the economics of lofting thousands of racks remain unproven.
Who else is racing to move compute off-planet?
Google is far from alone in trying to kick the data center problem into outer space. Elon Musk's SpaceX has been experimenting with the idea, and Musk's version may involve putting a gigantic catapult on the moon — really. Startups are working on it too. A company called Starcloud is among those developing orbital data center technology.
The competitive logic is straightforward. AI companies face a binding constraint on Earth: power. Training and running large models requires electricity at a scale that utilities struggle to supply, and data center construction is colliding with local opposition over noise, land use, and grid strain. If compute could run on uninterrupted orbital sunlight, the power constraint loosens dramatically. Whoever solves the engineering first would hold a genuinely different kind of AI infrastructure.
What's the risk back on Earth?
There's also a downside for the rest of us, and scientists have worried about it for decades. It's called Kessler syndrome. In that theoretical scenario, the density of objects in low orbit gets so high, and collisions become so commonplace, that they trigger a chain reaction of debris. That debris doesn't stay up there harmlessly. Earth is on the receiving end.
The theory was first posited when just under 2,000 objects circled in low-Earth orbit. That number has since climbed past 47,000. Elon Musk has recently floated the idea of adding millions of new satellites to the mix. Mega-constellations of orbital data centers would add yet more mass, and more collision candidates, to an already crowded region of space.
Why the 2029 date matters
Early 2029 is the number to watch. That's when New Glenn is scheduled to carry the first Project Suncatcher hardware toward sun-synchronous orbit, and when Google's moonshot gets its first real orbital test at rack scale. Everything before that — the cooling designs, the radiation hardening, the network architecture linking satellites into something that behaves like a data center — happens on the ground.
The project's success is not a given. The October satellite runs 15 minutes at a time. Cooling, radiation, launch costs, and orbital debris each pose problems without proven solutions at scale. But if Google and Blue Origin can make a network of solar-powered TPU satellites work, the payoff would be AI infrastructure that draws on unlimited sunlight and never competes with a terrestrial grid — a result that would reshape how the industry thinks about where compute physically lives.
Original: blog.google
More from Elena Vasquez
Show full bio
Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.
795 articles
Related articles
- Google books Blue Origin for 2029 orbital data center launch
- Google's First Suncatcher Orbital Data Center Test Launches October 1
- Google Books New Glenn to Launch Its Orbital AI Data Centers in 2029
- Google's Suncatcher sends a fridge-sized AI satellite to orbit on October 1
- Google Will Launch Its First AI Test Satellite on Oct. 1