Google has decided that the next place to run its AI is not a bigger building on Earth. It is space.
On Thursday, November 19, Blue Origin announced that Google has picked its New Glenn rocket to launch a set of AI computing satellites in the first quarter of 2029. The satellites belong to Project Suncatcher, Google’s long running experiment in running machine learning hardware in orbit, powered by sunlight. It is the first time Google has booked a dedicated heavy lift flight for the project, and it is the clearest sign yet that the company is serious about testing whether the cloud can leave the ground.
The deal is a research step, not a product launch. But it says a lot about where the AI industry thinks its real bottleneck is. It is not chips. It is power.
What Google just agreed to
The agreement is a launch services deal between Google and Blue Origin, the rocket company founded by Jeff Bezos. Blue Origin’s New Glenn rocket will carry several Project Suncatcher satellites into sun-synchronous orbit in early 2029.
Each satellite carries Google’s own tensor processing units, the chips the company designs in house to run AI programs. Google says each satellite holds computing capacity roughly equal to one rack of servers in a conventional data center. The satellites are meant to be linked to each other by free-space optical links, beams of light that carry data between spacecraft instead of cables.
“Project Suncatcher is a long-term moonshot to scale AI infrastructure in space by tapping into the abundant solar power in orbit,” said Travis Beals, senior director of Project Suncatcher at Google. “This dedicated launch with Blue Origin represents a crucial milestone on our roadmap as we expand the boundaries of what’s possible for the future of compute.”
For Blue Origin, the deal is a significant win. New Glenn is its heavy lift rocket, able to lift 45 metric tons to low Earth orbit inside a seven meter nose cone. Jordan Charles, senior vice president of New Glenn at Blue Origin, said the rocket’s “heavy-lift capacity and large payload volume give customers the flexibility to bring new capabilities to orbit.”
Why space, why now
The logic of Suncatcher starts with a problem every AI company now faces: data centers need enormous amounts of electricity, land, and cooling water, and communities are pushing back.
Anthropic’s own expansion has run into local opposition, including a fight over a facility under construction in West Virginia. Voters in several places this month pushed back against new data center projects. The industry’s answer so far has been to build bigger and buy more power. Google is asking whether the power problem can be sidestepped entirely.
In orbit, sunlight is constant and free. A satellite in sun-synchronous orbit stays in daylight, so its solar panels never go dark. There is no land to buy, no grid connection to wait for, and cooling in space works by radiating heat away rather than pumping water through a building. The pitch is simple: if the most expensive part of AI is the electricity, put the computers where electricity falls from the sky.
Google described Suncatcher on its research blog as “a long-term research moonshot exploring whether space could one day host scalable machine learning infrastructure.” The word “could” is doing a lot of work there. Nobody has shown that an orbital data center makes economic sense yet. Launching hardware is expensive, fixing it in orbit is nearly impossible, and a satellite that fails is gone. The 2029 flight is designed to answer whether the idea survives contact with reality.
The road so far
Suncatcher has moved in careful steps, and each one has worked so far.
Google published the system concept on November 4, 2025: compact constellations of solar powered satellites carrying TPUs, connected by optical links. On October 1, 2026, Google and Planet Labs launched the first prototype, a refrigerator sized satellite carrying four TPUs, aboard a SpaceX Transporter-18 mission. Google confirmed contact with the satellite and said it was operating as expected, gathering data on how its TPUs handle the physical stress of launch and the radiation and thermal extremes of space.
On October 30, Google reported that the prototype had powered up its TPU hardware and generated a response using an onboard Gemma 4 model. That is a small result with a large meaning: a Google AI chip has now run a model in orbit and answered back.
The next step is a two-satellite experiment planned for 2027, to test whether the optical links between satellites work well enough for them to act as one distributed system. The 2029 New Glenn flight would then scale the test up to several satellites at once. Google has not announced that the 2029 mission will deploy an operational orbital data center. It is still research, and Google is careful to say so.
The rocket has its own problems
The 2029 date depends on a rocket that has had a rough year. Blue Origin’s New Glenn program has been beset by delays, and a test explosion on May 28 at Cape Canaveral, tied to a BE-4 engine valve, damaged its Florida launchpad, which is still being repaired. The next New Glenn flight has been pushed to January 2027.
That matters because the Suncatcher schedule is now tied to Blue Origin’s recovery. If New Glenn slips again, the 2029 window slips with it. Google chose Blue Origin over SpaceX for this flight even though SpaceX carried the first Suncatcher prototype in October. The choice gives Blue Origin a major future mission as it tries to build its launch business, and it spreads Google’s bets across more than one launch provider.
There is also an irony worth noting. Blue Origin was founded by Jeff Bezos, and Amazon is both a cloud rival to Google and a major backer of Anthropic. In the AI infrastructure race, the old rivalries are tangled: the rocket carrying Google’s AI chips into orbit belongs to the founder of one of Google’s biggest competitors.
What this really tests
Strip away the space romance and Suncatcher is a bet on the economics of power. Training and running frontier AI models now consumes electricity at a scale that strains grids and provokes local opposition. Every major lab is hunting for cheaper, more available power, whether that means new nuclear plants, deals with utilities, or, in Google’s case, the sun.
Google has one advantage here that its rivals do not: it designs its own AI chips. It does not need to buy Nvidia hardware and ship it to orbit. It can put TPUs on satellites and keep the whole stack, silicon included, under its own control. That is the same vertical integration that made its data center business work, extended to space.
But the economics are unproven and the timeline is long. A test flight in 2029, with a working system years after that at best, will not solve the power crunch AI companies face today. Anthropic has committed around $900 billion to compute and infrastructure over the coming years. OpenAI’s commitments are similarly vast. Those bills are due on Earth, in the next few years, not in orbit.
The honest way to read Suncatcher is as a hedge. If orbital computing works, the company that proves it first owns a new kind of infrastructure. If it does not, Google has spent a modest amount of research money learning exactly why. The prototype cost is small next to a single data center campus.
SpaceX and a startup called Starcloud are pursuing similar ideas, which suggests the concept has moved from science fiction to an engineering question with more than one team working on it. That does not make it real yet. It makes it a race worth watching.
What to watch
Three dates matter now. First, New Glenn’s next flight in January 2027, which will show whether Blue Origin has recovered from its May explosion. Second, the two-satellite optical link test in 2027, which answers the core networking question: can moving satellites exchange enough data, reliably enough, to act as one distributed AI system. Third, the 2029 launch itself, and whether several rack-scale satellites can operate together in orbit.
If the optical links work and the satellites hold up, expect Google to talk about Suncatcher very differently, as a path to real infrastructure rather than a research project. If the links struggle, the project will likely stay what it is now: a fascinating experiment with a long horizon.
Either way, the announcement marks a shift. The AI industry spent the last three years arguing about who has the best models. It is now arguing about where the computers that run them should live. Google’s answer, at least in part, is above our heads.
Related reading: SpaceX Bets $74 Billion on AI Supercomputers | Gemini 4 Neon Is Google’s Fast, Cheap Frontier Model | America Is About to Pick Its AI National Champions
make it pop
Share card studio
More from Signals
All in topicAnthropic Opened at $2 Trillion. It Closed at $97 a Share.
Anthropic is now a public company, and the market has put its first price on a frontier AI lab. The maker of Claude opened on the…
Gemini 4 Neon Is Google’s Fast, Cheap Frontier Model
Google has launched a fourth Gemini 4 model in five weeks. Gemini 4 Neon arrived on November 18, and it is the clearest sign yet of…
Anthropic Priced Its IPO at $83. The Market Just Valued Claude at $1.5 Trillion
Anthropic priced its initial public offering at $83 a share on Wednesday, above the $70 to $78 range it set last week. The final price was…
OpenAI’s GPT-6.2 Is Cheaper, Sharper, and Built for Real Work
OpenAI shipped two new models on November 17, and the most interesting thing about them is not how smart they are. It is how cheap they…
Keep exploring