Google’s AI Chips Are Live in Orbit. Now Comes the Hard Part.
Starship reached orbit, Google’s chips went live in space, two power experiments hitched a ride – and Amazon warned that Earth is getting harder to build on.
Listen to the audio version of this article (generated by AI).
On Monday, Sept. 28, SpaceX (SPCX) sent Starship into orbit for the first time.
Three days later, Google (GOOGL) confirmed that its first Project Suncatcher prototype was online and operating in space.
The Falcon 9 mission that carried Google’s prototype also hauled two more unusual payloads. One is designed to beam power from one satellite to another. The other will test whether solar energy collected in orbit can be sent back to Earth.
Then, on Friday, Amazon (AMZN) warned that more than 100 communities are considering moratoriums on new data centers.
That is a lot of movement in four days.
We have spent months writing about why orbital computing could eventually make sense. We have studied the cost curve, the engineering challenges, the SpaceX–Nvidia strategy, and the companies building around it.
This week, the story picked up its pace.
Orbital Data Centers Are Moving From Theory to Hardware
When we broke down Google’s Project Suncatcher roadmap a few days ago, the whole thing was still on paper.
Google now has hardware in orbit.
The company says its prototype satellite, built with Planet Labs (PL), successfully established contact after launch and is operating as expected. Over the coming weeks, Google will test how its Tensor Processing Units (TPUs) handle the physical stress of spaceflight, radiation, and extreme thermal conditions.
This is still a small research mission.
One test satellite is a long way from an orbital data center. It will not train the next Gemini model or replace one of Google’s giant terrestrial computing campuses.
Its job is more basic.
Does the hardware survive launch? How does it perform under radiation? Can the system manage the enormous swings in temperature? Which parts fail first?
Those questions are difficult to answer inside a laboratory. Eventually, the equipment has to fly – and now it is.
This will give Google real operating data from real hardware in the environment where a future system would have to work.
The results will shape the next design. They will also begin revealing which components can move forward largely unchanged and which ones require a complete rethink.
That’s how a science project turns into a parts list.
Orbital Data Centers Will Need Their Own Power Infrastructure
Getting a chip to orbit is one thing. Running it is another.
A powerful AI satellite also needs electricity, and the amount required can change dramatically from one moment to the next.
Today, each spacecraft generally brings its own solar arrays and batteries. Engineers size that equipment around the satellite’s expected peak workload, which adds mass and limits what the machine can do.
A shared power network could change that equation.
Star Catcher Wants to Sell Power as a Service in Orbit
Star Catcher’s new Protostar satellite is designed to harvest energy, locate another free-flying spacecraft, and transmit power to it. After commissioning, Protostar plans to deploy a CubeSat and beam energy into ordinary solar panels attached to that second spacecraft.
If the experiment works, Star Catcher says it would mark the first optical power transfer between two untethered spacecraft.
In other words, one satellite would act like a power station for another.
That could eventually allow computing satellites to draw extra electricity during demanding workloads without carrying every panel and battery themselves. Power generation could sit in one part of a constellation while compute sits somewhere else.
And customers aren’t waiting for the demo.
Star Catcher has announced 10 power-purchase agreements with space companies, more than 40 letters of intent, and a $60 million strategic-funding award from the U.S. Space Force. Those agreements may not guarantee a profitable orbital grid; but they show that spacecraft operators are already thinking seriously about buying power as a service.
The Transporter-18 mission carried another energy experiment, too.
Reason-1, built by Cowboy Space, is designed to collect solar energy in orbit and test sending it toward Earth. Founder Baiju Bhatt has described the long-term idea as something like Starlink for power: an orbital network that could deliver electricity to areas without a dependable terrestrial grid.
Both concepts are still highly experimental.
Power can be lost during transmission. Hardware has to be extremely precise. Regulators will surely have plenty to say about beams traveling between spacecraft or toward Earth. And the economics are nowhere near proven.
But those questions are now being answered in orbit instead of argued in conference rooms.
Starship Just Took a Big Step Toward Orbital Freight
None of this matters if you can’t get the stuff up there – and “the stuff” is heavy. An orbital AI system isn’t just chips. It’s radiators the size of billboards, batteries, shielding, solar arrays, laser links – tons of hardware that has to survive launch and arrive in exactly the right orbit, again and again.
Which is why the week’s quietest big deal happened on Sept. 28.
That’s when SpaceX’s 14th Starship test became the vehicle’s first true orbital mission. It also deployed 26 next-generation Starlink V3 satellites, giving the rocket its first meaningful orbital payload.
Reaching Orbit Is Only the First Test
That doesn’t mean the mission was perfect, though. One booster engine shut down during ascent. After additional engine trouble, SpaceX shortened the planned flight from about 10 hours to roughly three. The upper stage ended its mission with a fiery Pacific splashdown.
Rapid reusability is questionable. So is the long-term launch cost.
But Starship completed two jobs orbital infrastructure desperately needs:
It reached orbit, and it released a large payload.
Previous flights proved that the vehicle could survive parts of the journey. Flight 14 showed it has the potential to become a freight system.
Musk is now targeting a Starship cadence of one or two launches per week next year. He also says SpaceX and Tesla are aiming for 200 gigawatts of annual solar-production capacity, partly because solar panels can generate much more consistently in space than they can on Earth.
Starship still has to demonstrate reflight, orbital refueling, safe recovery, and dependable operations before a weekly cadence becomes realistic.
Take the targets with a grain of salt. Musk’s timelines always need some. But even discounted, they sketch the size of the machine he’s assembling: industrial-scale solar production, heavy launches every week, and enormous amounts of computing hardware operating beyond Earth.
Earth Is Adding Urgency
The case for orbital AI does not depend on terrestrial data centers disappearing. They will remain the center of the computing world for years.
But adding new capacity on Earth is becoming more expensive, political, and time-consuming.
Amazon just committed more than $1 billion over five years to communities that host its data centers. The company says the money will support education, workforce training, energy affordability, and water and energy conservation. It also pledged to stop using nondisclosure agreements with government agencies around new projects.
This comes as communities are raising concerns over electricity prices, water consumption, construction, noise, local benefits, and the lack of transparency around some projects. AWS chief Matt Garman says more than 100 data-center moratoriums are now being considered across the United States.
Amazon is spending money to keep the terrestrial buildout moving.
Google is now collecting data from the first hardware designed to explore an alternative.
Orbit doesn’t get cheaper just because a town says no to another server farm. But every “no” makes an alternative worth a little more.
A computing platform that can add capacity without competing for a local grid connection, a water permit, or a zoning approval could eventually solve a very expensive problem.
That future is still years away. But the pressure creating it is already here.
The Bottom Line: Orbital Data Centers Have Entered the Prototype Phase
A few months ago, orbital computing was mostly an argument about future economics.
Then SpaceX chose Nvidia hardware for Starmind. Google put Project Suncatcher on a launch schedule. Startups began raising money around orbital power.
Now Starship has reached orbit with a meaningful payload.
Google’s first prototype is alive in space.
And two power experiments are beginning the work required to test an orbital energy network.
The economics remain unresolved, the engineering difficult.
Many of these experiments will probably disappoint. Some may fail completely.
But a concept becomes a prototype. A prototype creates supplier requirements. Successful tests turn into larger programs, manufacturing schedules, and orders.
And that buildout sits at the center of what I call XPANSE.
It started as a thesis. This week, pieces of it started phoning home from orbit.
Musk himself has floated the idea that early investors in this project could see 1,000-fold gains. I won’t promise that; nobody honestly can. What I can do is show you the map: the three moves I’d make today to get positioned, and the one stock I believe sits closest to the center of the buildout. Its name and ticker are in my new briefing, free – no strings attached.
The first pieces of the orbital grid are already overhead.
Watch the XPANSE briefing – and grab the free pick – right here.
