Google Is Taking AI Data Centers to Space — and the Future Could Be Closer Than We Think
As communities in the United States increasingly push back against large AI data centers being built near their homes, technology companies are exploring a very different solution: put the data centers in space.
Google’s First Step Toward Orbital Data Centers
Google is testing an ambitious project called Project Suncatcher, which explores the possibility of placing AI computing infrastructure in orbit around Earth.
The company recently launched its first experimental satellite as part of this research. The satellite traveled aboard a SpaceX Falcon 9 rocket from Vandenberg Space Force Base in California on a rideshare mission known as Transporter-18.
.webp "A SpaceX Falcon 9 rocket with a crew of four launches from pad 40 at the Cape Canaveral Space Force Station in Cape Canaveral, Fla., Thursday, Oct. 1, 2026.") |
| A SpaceX Falcon 9 rocket with a crew of four launches from pad 40 at the Cape Canaveral Space Force Station in Cape Canaveral, Fla., Thursday, Oct. 1, 2026. (AP Photo/John Raoux) |
The goal isn't to immediately build a giant AI data center above Earth.
Instead, Google wants to answer a much more basic question:
Can its AI chips actually survive and work reliably in space?
The experimental satellite is designed to test Google's Tensor Processing Units (TPUs) after they experience the intense forces of a rocket launch and the harsh radiation environment of space.
Travis Beals, Google's senior director of paradigms of intelligence, explained that the first mission is mainly about learning what works, discovering what fails, and using those lessons to improve future missions.
That approach makes sense. Before anyone builds an enormous AI computing network in orbit, they need to understand the small details that could make or break the idea.
Google plans to continue the research with two more satellites in 2027.
Why Put AI Data Centers in Space?
At first, the idea might seem completely unnecessary.
Why spend enormous amounts of money launching computers into orbit when data centers can simply be built on Earth?
One reason is the growing demand for AI computing.
Modern AI models require huge amounts of computing power. As companies build increasingly advanced AI systems, they also need more electricity, more land, more cooling infrastructure and larger data centers.
That growth is creating problems on the ground.
Large data centers can place significant demands on local electricity and water resources, while communities sometimes object to having these massive facilities built nearby.
Space could potentially offer a completely different environment.
There is plenty of sunlight available in orbit, and solar energy could potentially provide a major source of power for future computing satellites.
But getting there is the difficult part.
Space Comes With Its Own Problems
Moving an AI data center into space doesn't make the engineering challenges disappear.
It creates an entirely new list of problems.
One of the biggest concerns is radiation.
Earth-based computers are protected by the planet's atmosphere and magnetic field. Satellites don't have that same level of protection.
High-energy particles can interfere with electronic systems and cause something known as a bit flip, where stored or processed data changes unexpectedly.
Long-term exposure to radiation can also damage hardware.
Google has already tested some of its TPU technology in simulated radiation environments at the Crocker Nuclear Laboratory at the University of California, Davis.
According to Google, the chips performed remarkably well during those tests.
But there is an important limitation:
Some things simply cannot be fully understood until you actually put the hardware into space.
That's why the satellite experiments matter.
Then There’s the Heat Problem
Radiation isn't the only challenge.
There is another problem that may sound surprisingly simple: keeping the computers cool.
Anyone who has used a powerful computer knows that high-performance hardware produces heat.
On Earth, data centers can rely on air, water, cooling towers and other systems to move that heat away from computer equipment.
Space is completely different.
There is no normal atmosphere surrounding an orbital satellite. There are no air molecules available to carry heat away through ordinary convection.
That means future space-based data centers will need specialized systems to release heat.
Google is exploring technologies such as large radiators and heat pipes that could move heat away from the computing hardware and release it into space.
However, the company's first test satellite isn't designed to solve the entire cooling problem.
In fact, the prototype has a very practical limitation.
According to a Google spokesperson, the satellite will need to shut down approximately every 20 minutes to prevent overheating.
Think about that for a moment.
A computer designed to test the future of space-based AI can't continuously run because it gets too hot.
That's not necessarily a failure. It's exactly why experiments like this are important.
Before a technology can become practical, engineers have to discover its weaknesses.
The Power Challenge Could Be Even Bigger
Cooling is only one part of the problem.
AI computing requires enormous amounts of electricity.
Google's first experimental satellite is expected to provide only around 1 kilowatt of power to its TPUs, according to reporting by The New York Times.
That may sound like a lot, but it is tiny compared with what a fully operational AI computing satellite could eventually require.
Caleb Henry, director of research at Quilty Space, has noted that today's telecommunications satellites generally operate with around 10–20 kilowatts of power.
AI computing satellites could potentially need dramatically more.
A Google research paper published in the journal Joule suggests that future space-based data centers could require approximately 50–100 kilowatts, roughly comparable to the power requirements of a typical data rack on Earth.
That's a huge jump.
And it shows just how different a small experimental satellite is from a true orbital AI data center.
Google Wants the Next Satellites to Run Continuously
Google's next two planned test satellites are expected to take things further.
Unlike the first prototype, the company says these satellites should be capable of operating continuously without shutting down because of overheating.
They may also help Google test another important technology: laser-based communication between satellites.
This could become extremely important if future AI data centers are built as large networks of satellites rather than as isolated machines.
Instead of one giant computer floating above Earth, imagine dozens, hundreds or potentially thousands of satellites communicating with one another and sharing computing workloads.
That's where the idea starts becoming much more ambitious.
Space Traffic Is Another Concern
Of course, space isn't empty.
Earth's orbit is becoming increasingly crowded with satellites.
As more companies launch communications, navigation, observation and scientific satellites, the risk of collisions and orbital congestion becomes a growing concern.
Adding large numbers of AI computing satellites could make that situation even more complicated.
Any future orbital data center network would therefore have to be designed with satellite traffic, collision avoidance and space sustainability in mind.
There Is Also a Problem Nobody Can Ignore: Cost
Even if engineers successfully solve radiation, cooling, communications and power problems, one major question remains:
Will it actually be cheaper than building data centers on Earth?
This could ultimately determine whether orbital AI computing becomes a real industry or remains an interesting experiment.
AI hardware changes incredibly quickly.
New AI chips can become outdated within just a few years, while satellites are expected to remain in operation for much longer.
A June report from JLL Research pointed out this mismatch: AI chips can advance every 1–2 years, while satellites can remain operational for around 5–7 years.
That creates a difficult problem.
Imagine spending millions of dollars launching an AI satellite into orbit, only to discover a year or two later that a new generation of AI hardware is dramatically faster.
Replacing that hardware isn't as simple as opening a data center on Earth and installing a new server.
You have to get the equipment into space.
And that is expensive.
Launch Costs Matter More Than Ever
For orbital data centers to become economically attractive, the cost of launching hardware into space would likely need to fall significantly.
But launch costs are not necessarily moving in that direction.
This is one area where SpaceX has a major role in the conversation.
SpaceX builds its own rockets, produces large numbers of satellites and operates one of the world's largest satellite constellations.
The company has also discussed extremely ambitious plans for AI computing in orbit, including a potential constellation of up to 1 million AI-computing satellites.
Google has not announced a comparable final number for Project Suncatcher.
A 2025 research paper suggested that Google's concept could involve satellite clusters containing around 81 satellites, although the company has not said how many clusters a full operational system would require.
So, for now, the numbers remain part of the bigger research question rather than a confirmed deployment plan.
This Isn't Going to Happen Overnight
It's easy to look at a project like this and imagine that giant AI data centers will soon be floating above our heads.
Reality is probably much slower.
There are still enormous technical and financial questions to answer.
How much will launches cost?
How long will the hardware survive?
How will satellites stay cool?
How will they communicate?
How much electricity can they generate?
How will they be repaired or upgraded?
And perhaps the biggest question of all:
Will computing in space actually make economic sense?
Google appears to understand that these answers won't come quickly.
The company describes Project Suncatcher as a long-term research effort, comparing it with technologies such as autonomous driving and quantum computing, where years of experimentation were necessary before practical applications became possible.
That mindset is important.
The first satellite doesn't need to prove that space-based AI data centers are ready for mass deployment.
It simply needs to teach engineers something they didn't know before.
The Beginning of a Much Bigger Experiment
Google's partnership with satellite company Planet Labs also shows that the idea is being taken seriously as a research project.
Planet Labs cofounder and CEO Will Marshall has described the project as potentially viable over the long term, while also acknowledging that it will require significant research and development investment.
And that's probably the best way to look at Project Suncatcher right now.
It's not a finished product.
It's an experiment.
But it's an experiment happening at a fascinating moment in the technology industry.
AI is consuming more computing power than ever before. Data centers are becoming larger and more energy-intensive. At the same time, companies are searching for new ways to generate electricity, manage heat and expand computing infrastructure.
Space may eventually become part of that equation.
For now, Google is taking small steps: launch a satellite, test the chips, study radiation, experiment with cooling, measure power requirements and learn from what goes wrong.
The journey from a small experimental satellite to a functioning orbital AI data center could take years.
But every major technology has to start somewhere.
And this one has just begun its journey above Earth.
