Google is sending a refrigerator-sized satellite into space to test whether AI data centers can live up there. The project is called Suncatcher, and the first test satellite, dubbed MVP, will launch on October 1 aboard a SpaceX Falcon 9 rocket.
The launch is a proof-of-concept for a vision that has been floated by tech leaders like Elon Musk and Jeff Bezos. The idea is simple: put AI training hardware in orbit, powered by the sun, and connect it back to Earth at high speed. Google’s version is a moonshot. The question is whether the first shot even works.
The Satellite That Carries Four Chips
MVP is about the size of a refrigerator. Inside it carries four of Google’s custom TPU AI accelerators. These chips are used to train AI models and run inference to generate tokens for AI workloads. On Earth, a data center runs thousands of these chips, which consume massive amounts of power. That is one of the reasons putting solar-powered AI hardware in orbit looks attractive.
The satellite does not come from Google’s own engineering shop. The AI hardware is Google’s, but the spacecraft itself comes from Planet Labs, a satellite imagery firm. The original plan was to launch two custom satellites in 2027. Google decided to move faster by integrating its chips into a satellite Planet Labs had already built.
Power Is the Limiting Factor
The satellite’s solar panels supply only about one kilowatt of power, according to The New York Times. That is enough to power a microwave or a hair dryer. It is not much, but it is the whole point of the test. The satellite will operate for just a few months.
The low power output means the satellite cannot run the kind of heavy computation that Earth-bound data centers handle. What it can do is validate whether the hardware holds together in orbit and whether the concept is sound. If MVP survives its mission, Google has a working prototype. If it fails, the company knows the problem is somewhere in the design.
The Launch and What Comes Next
MVP will launch as part of Transporter-18, a rideshare mission. That means it is hitching a ride on a SpaceX Falcon 9 rocket alongside other payloads.
Eventually, Google wants networks of orbiting satellites that communicate over high-speed laser connections. That will likely require expensive dedicated launches. This lone satellite is a test of several technologies at once, not a full deployment.
Why Space Matters for AI
The appeal of Suncatcher is straightforward. Data centers on Earth consume vast amounts of power. Putting AI hardware in orbit, where it can draw on the sun directly, cuts some of that cost. The trade-off is that the hardware becomes harder to reach and repair.
The comparison is stark:
| Feature | Terrestrial Data Center | Orbital Data Center |
|---|---|---|
| Power Source | Grid electricity | Solar |
| Distance From Users | Direct | Through satellites |
| Maintenance | Easy | Hard |
| Environmental Impact | High | Lower |
The table shows why the trade-off exists. The satellite’s low power output is a feature of the test, not a failure. It is meant to show that the hardware can survive in orbit at all.
The Risks of the Test
A few months is a short window for a satellite. Anything that fails will fail fast. The test is a bet on the future of computing. It is a bet that the cost of launching hardware outweighs the cost of running it on Earth. Whether that bet pays off depends on whether MVP survives its mission and whether Google can scale the concept beyond a single satellite.
What We Know So Far
- Satellite: MVP, about the size of a refrigerator
- Chips: Four custom TPU AI accelerators
- Power: About one kilowatt from solar panels
- Launch: October 1, on SpaceX Falcon 9 via Transporter-18
- Mission length: A few months
- Next milestone: Potential full deployment of a constellation
Our Take on the Launch
This is a bold move by Google, but the scale of the ambition is mismatched with the scale of the test. A refrigerator-sized satellite with the power of a hair dryer is not a warning shot. It is a gentle probe.
The test is valuable. It answers the question of whether the hardware survives the trip to orbit and whether it works at all. But it does not answer the question of whether the concept scales. For that, Google needs to launch bigger satellites with more power and more chips.
The satellite is a prototype, and prototypes are supposed to fail. The question is whether MVP fails in a way that tells Google something useful, or fails in a way that sends the project back to the drawing board. Either outcome is information. Neither is embarrassment.
The test is scheduled for October 1. The world will know whether the prototype survived within a few months.
See the video the story is built around at Ars Technica.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.