Google Project Suncatcher (2026): Blue Origin Will Launch Its AI Data Center Satellites in 2029

Jump to comments

Blue Origin's New Glenn rocket rising past its launch tower at sunrise, with pink exhaust clouds billowing at the base of the pad and a water tower and crane nearby
Blue Origin’s New Glenn rocket lifts off from Space Launch Complex 36 at Cape Canaveral on its third flight, April 19, 2026.
Photo: Gwendolyn Kurzen/U.S. Space Force, 2026, via Wikimedia Commons (Public domain)

Google has booked a whole Blue Origin rocket for the next big step of Project Suncatcher, its long-shot plan to run AI computers on solar-powered satellites. The two companies announced a launch services agreement on Thursday, Nov. 19, 2026: a New Glenn heavy-lift rocket will carry a set of Google satellites, each packing roughly the computing capacity of a data center rack, into a sun-synchronous orbit in the first quarter of 2029.

The short answer: This is a launch contract, not a data center. Nothing new goes up for more than two years, Google has not said how many satellites will fly or what the deal costs, and the rocket it picked has been grounded since a launch-pad explosion in May. What the deal does show is that Google is moving Suncatcher from a single fridge-sized test satellite, which has carried Google’s TPU chips in orbit since Oct. 1, toward a dedicated mission sized for rack-scale hardware. It is the clearest commitment yet from a big cloud company to test whether orbiting AI computers make sense, and it landed the same day SpaceX’s lenders were reported to be worried about Nvidia chips ending up in orbit.

Below: what was announced, how Suncatcher got here, why New Glenn matters and what it still has to prove, the engineering problems the 2029 satellites have to solve, and how Google’s plan compares with SpaceX’s and Starcloud’s.

What Google and Blue Origin announced

Blue Origin’s release says the two companies "have entered into a launch services agreement" and that the mission will be Google’s first dedicated heavy-lift launch for Project Suncatcher. Google’s own post, by Travis Beals, the senior director who leads the project, calls it "a crucial milestone on our long-term roadmap" and says the launch "builds on our ongoing work with other satellite and launch partners."

What flies: Google satellites carrying its Tensor Processing Units, or TPUs, the AI chips it designs for its own data centers. Each satellite will have "roughly the capacity of a data center rack," in Google’s words. Reuters reported that "several" satellites will ride on the rocket; Via Satellite noted that neither company gave a number.

Where: sun-synchronous orbit, a near-polar orbit that keeps a satellite in the same relationship to the Sun all year. Google’s research has favored a dawn-dusk version of it, which rides the line between day and night so the solar panels almost never see Earth’s shadow.

When: the first quarter of 2029.

What wasn’t said: the price, the satellite count, the total power on board, which TPU generation will fly and who is building the spacecraft. Google did not say why it chose Blue Origin, either. Its parent, Alphabet, owns a SpaceX stake that CNBC valued at more than $82 billion in October, and SpaceX launched Suncatcher’s first satellite.

Blue Origin’s senior vice president for New Glenn, Jordan Charles, said the project "represents the ambitious missions New Glenn was built to support," pointing to the rocket’s "heavy-lift capacity and large payload volume."

From research paper to rocket: the Suncatcher timeline

Nov. 4, 2025: Google unveils Project Suncatcher as a "research moonshot" and publishes a preprint describing clusters of TPU-carrying satellites linked by lasers. The plan then: a learning mission with the satellite company Planet to "launch two prototype satellites by early 2027."

Oct. 1, 2026: The first prototype, built with Planet, reaches orbit on SpaceX’s Transporter-18 rideshare from Vandenberg, California. Google says the team has made contact and the satellite is "operating as expected." NPR reported that it carries four TPUs, runs a version of Google’s open Gemma model for 15 minutes at a time because of heat limits, and flies in a standard orbit on batteries, not the sun-synchronous orbit planned for later satellites. Google says its peer-reviewed paper is now out in the journal Joule.

Oct. 30, 2026: Google reports its "Hello World" moment, which it calls "the first-ever AI response from a data center class TPU in orbit." During a ground-station pass, the team asked the onboard Gemma 4 model for a haiku about space and got one back as the satellite passed Svalbard, Norway.

Nov. 19, 2026: The New Glenn launch agreement is announced.

2027: Google’s next milestone is "twin interconnected satellites," a test of the laser links that would let satellites share AI work.

Q1 2029: The first dedicated heavy-lift mission, on New Glenn.

The schedule has already shifted once. The original plan was two prototypes by early 2027. Instead one prototype flew in 2026, the pair follows in 2027, and the rack-scale satellites come after.

Why New Glenn, and the catch

New Glenn is one of the few rockets that can lift what Google is describing. Blue Origin says it can carry 45 metric tons to low Earth orbit and more than 13 tons to geostationary transfer orbit. Its payload fairing, the nose cone that holds the cargo, is 7 meters wide, which Blue Origin says gives "twice the volume" of the 5-meter fairings most rockets use. The rocket stands more than 98 meters tall, and its first stage is designed to fly at least 25 times. Volume can matter as much as mass for satellites with large solar arrays and radiators.

The catch is that New Glenn has flown only three times and is not flying now. Its third flight, carrying AST SpaceMobile’s BlueBird 7 satellite, lifted off on April 19, 2026. Then on May 28, during a hot-fire test at Launch Complex 36 in Florida, a New Glenn exploded on the pad. Blue Origin chief executive Dave Limp wrote that the company lost the pad’s lightning tower, the transporter-erector that lifts the rocket upright, and its hydraulic cylinders, and that early analysis pointed to the aft section of the first stage. The rocket was due to launch 48 satellites for Amazon’s Leo internet service a few days later.

Blue Origin is rebuilding the pad in a new way rather than restoring the old one, with a hybrid setup in which the two stages are joined horizontally and the rocket is then lifted upright on the pad. On Nov. 5, CNBC reported a memo from Limp pushing the return to flight from the end of 2026 to January 2027. Limp told CNBC his first estimate after the explosion was that New Glenn could not fly again until mid-2028.

That leaves about two years between New Glenn’s planned return and Google’s mission, enough time for Blue Origin to build a flight record if the rebuilt pad works. Launch capacity is also getting tight. Starcloud’s chief executive, Philip Johnston, told TechCrunch in August that "launch is pretty constrained right now because [SpaceX’s] Falcon 9 program is scheduled to end in 2028." Google did not say whether that played a part in its choice. If you want to watch a New Glenn launch in person, our guide to Florida rocket launch viewing spots covers the best places on the Space Coast.

A green Google TPU v4 circuit board seen from above, with four chip packages under metal cold plates connected by red, blue, yellow and green liquid-cooling hoses
A Google TPU v4 board with four liquid-cooled chip packages, an earlier generation of the AI chips Project Suncatcher is testing in orbit.
Photo: Norman P. Jouppi et al., 2023, via Wikimedia Commons (CC BY 4.0)

What the 2029 satellites have to prove

Google’s 2025 research post and preprint lay out the problems Suncatcher has to solve before orbiting AI computers can compete with ones on the ground. The rack-scale satellites are where several of them will be tested together for the first time. For the deeper physics of power and cooling, see our explainer on whether AI data centers in space can work.

Moving data between satellites fast enough

Training and serving large AI models means spreading work across many chips that talk to each other constantly. Google says links between satellites would need "tens of terabits per second" to match a ground data center, and that this needs received signal power "thousands of times higher" than usual long-range laser links. Its answer is to fly satellites very close together. A bench demonstrator has reached 800 gigabits per second each way, or 1.6 terabits per second total, over a single transceiver pair. Its example layout puts 81 satellites within a 1-kilometer radius at a mean altitude of 650 kilometers, with neighbors roughly 100 to 200 meters apart.

Surviving radiation

Google tested Trillium, its v6e Cloud TPU, in a 67 MeV proton beam. The high-bandwidth memory was the most sensitive part and started showing irregularities after a cumulative dose of 2 krad(Si), nearly three times the 750 rad(Si) Google expects over a shielded five-year mission. No hard failures were attributed to radiation dose up to 15 krad(Si) on a single chip. Tests in a particle beam are not the same as years in orbit, which is what the prototype is for.

Getting rid of the heat

In a vacuum, chips cannot be cooled by air or water flowing past. All the heat has to leave through radiator panels. Beals told NPR that the radiators are among the heaviest parts of the current prototype, and the prototype runs its AI model in 15-minute bursts because of heat. A satellite with a rack’s worth of compute will have to shed far more heat, continuously.

Making the economics work

Google’s analysis says launch prices may fall below $200 per kilogram by the mid-2030s if they keep falling along a steady learning curve. At that price, it says, launching and running a space-based data center "could become roughly comparable" to the energy costs of an equivalent ground data center, per kilowatt per year. The upside is the power source: Google says a solar panel in the right orbit can be up to eight times as productive as on Earth and produce power almost continuously.

I don’t see this being something where it’s cheaper to do this in the next five years. I think it will take longer than that.|Travis Beals, Project Suncatcher lead, to NPR, Oct. 1, 2026

Outside experts are more cautious still. Brandon Lucia, an electrical and computer engineering professor at Carnegie Mellon University, told NPR that repairs and upkeep in orbit have "their cost and complexity amplified by a factor of 10, maybe a factor of 100." Beals himself expects ground and space data centers to coexist "for a long time."

Google isn’t alone: SpaceX, Starcloud and the money

SpaceX is the loudest backer of the idea. Its chief operating officer, Gwynne Shotwell, said in September that the company will deploy "supercompute in space" in 2027, CNBC reported, and NPR reported that SpaceX expects to start deploying "orbital AI compute satellites" as early as 2028. Elon Musk said earlier this year that space would be the cheapest place to train AI "within two years, maybe three at the latest." SpaceX also has its own rocket for the job: Starship, which made its first ship catch on Flight 15 last week.

The same day as Google’s announcement, money got in the way. The Financial Times reported, in a story summarized by Investing.com, that lenders behind SpaceX’s latest $43 billion debt package, called Project Saturn, sought safeguards to stop Musk from sending the Nvidia chips that secure the loans into space, because chips in orbit would be hard to repossess and sell after a default. SpaceX said the chips will be used in data centers on Earth, and it is not clear whether the restriction made it into the final documents. Bloomberg reported that the package includes a $25 billion investment-grade bond backed by SpaceX’s lease payments for the chips, priced at about 8.3%.

Starcloud, a startup whose investors include Nvidia, put an Nvidia H100, a data center GPU, into orbit on its Starcloud-1 satellite in November 2025. In August it raised a $250 million extension that valued it at $2.3 billion, and it plans to put two 8-kilowatt Starcloud-2 satellites on rideshare flights in 2027, TechCrunch reported. It has asked the Federal Communications Commission for permission to operate 88,000 spacecraft, and its biggest design is built around Starship.

Google’s approach differs in two ways. It is flying its own chips, the same family it already runs in its data centers on Earth, and it is moving in public, measured steps: one prototype, then a pair, then a rack-scale mission on a rocket it has now booked.

What we still don’t know

  • How many satellites, and how much computing in total. "Roughly a rack" each is the only figure given.
  • Which chips. Google tested Trillium in a proton beam, but it has not said which TPU generation will fly in 2029.
  • Whether the satellites will work as a cluster. The tightly packed formation and laser links are the heart of the plan, and the 2027 twin satellites are meant to test them first.
  • Who builds the spacecraft. Planet built the prototype; Google has not named a partner for the 2029 satellites.
  • Whether New Glenn is ready. Blue Origin is targeting January 2027 for its return to flight, and the 2029 date depends on Blue Origin flying regularly before then.
  • What it costs. Neither company disclosed the contract value.

Bottom line

Google’s Blue Origin contract turns Project Suncatcher from a test satellite into a scheduled heavy-lift mission, with rack-scale TPU satellites due in sun-synchronous orbit in early 2029. It is a real step, but a measured one. The satellites have not been built, the rocket is not flying yet, and Google’s own project lead says cheaper-than-Earth computing in orbit is more than five years away. The milestones to watch are New Glenn’s return to flight in January 2027 and Google’s twin-satellite laser-link test later that year. If both go well, the 2029 launch will be the first real test of whether a data center’s worth of AI chips can live on sunlight.

Further reading

Cover of Chip War: The Fight for the World's Most Critical Technology by Chris Miller
Chip War by Chris Miller

The history of the computer chip and the companies that make it, and why AI chips like Google’s TPUs have become such a prize.

Cover of Supremacy by Parmy Olson
Supremacy by Parmy Olson

How Google DeepMind and OpenAI raced to build ever bigger AI systems, and why the computing behind them keeps growing.

Charcoal Amazon Echo Dot smart speaker, a fabric-covered sphere with a blue light ring at its base, on a wooden table
Amazon Echo Dot

A small Alexa smart speaker; every question you ask it is answered by AI running in a distant data center, the kind of computing Suncatcher wants to move into orbit.

Four Apple AirTag trackers in a row, white-rimmed discs with a polished silver back and the Apple logo
Apple AirTag 4-pack

Coin-sized trackers that find keys and bags through Apple’s Find My network, one of the most popular gadgets on Amazon.

WD_BLACK SN850P 2TB NVMe M.2 drive with a black heatsink marked with the PlayStation logo
WD_BLACK SN850P 2TB NVMe SSD

A fast 2TB NVMe drive with a heatsink, licensed for the PlayStation 5; a reminder that keeping fast chips cool is a problem on the ground too.

Aglena Best Of product picks

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top
Aglena