On Thursday, October 1, a SpaceX Falcon 9 lifted off from Vandenberg Space Force Base in California on the Transporter-18 rideshare mission. Among its 130 payloads was something unusual: a prototype satellite called MVP, built by Google with the Earth-imaging company Planet Labs, carrying four of Google's Tensor Processing Units — the custom chips that power much of the company's AI. Shortly after reaching orbit, Google confirmed contact. The search giant had just put a computer in space and phoned it.
The mission is called Project Suncatcher, and it's Google's first real step toward a genuinely strange idea: data centers in orbit.
What "MVP" actually is
Let's be precise about scale, because the phrase "orbital data center" invites the wrong mental image. MVP's four TPUs have roughly the compute power of a single server rack. Its solar panels generate about one kilowatt — enough to run a hair dryer. It will spend the next year answering simple AI queries with Gemini models, but only in bursts of about 15 minutes at a time, after which thermal limits force it to shut down and cool off.
It will keep circling Earth for about six years before burning up on reentry.
So this is not a data center. It's a test article — and Google is admirably upfront about that. "This first launch is about seeing what works, identifying points of failure, and applying those findings to future missions," said Travis Beals, Google's senior director for paradigms of intelligence. Two more satellites are planned for 2027.
Why orbit at all
The pitch is about electricity, not computers. The thing constraining AI infrastructure right now isn't chip supply — it's power. Data center electricity demand tied to AI workloads grew roughly 50% in 2025 alone, and utilities are years behind on the transmission and generation capacity needed to keep up. In the right low-Earth orbit, solar panels get near-constant, unfiltered sunlight — Google says up to eight times the energy the same panels would produce on the ground. No clouds, no night, no atmosphere in the way.
There's a second motive, and it's a political one: nobody wants a data center in their backyard. Six in 10 Americans say they'd be uncomfortable with one opening in their neighborhood. Orbit has no zoning board, no water-use permits, no town hall meetings. Elon Musk, Jeff Bezos, and Jensen Huang have all mused about orbital compute; Google just became the first to actually fly hardware.
The part nobody's selling: heat
Here's the detail I keep coming back to: 15 minutes on, then shutdown. Space feels cold, but vacuum is a near-perfect insulator — there is no air to carry heat away from a hot chip. Every watt a TPU burns has to be radiated away as infrared, slowly, through panels pointed at the cold dark. On Earth we cool data centers with fans, chilled water, and cooling towers; in orbit, your only coolant is the void, and the void is patient but slow.
That 15-minute duty cycle is the most honest number in the whole story. It tells you exactly where the physics bites. Google's engineers will spend the coming weeks measuring how the chips handle launch vibration, radiation, and the savage temperature swings of orbital day and night. The chips were pre-blasted with radiation at Crocker Nuclear Laboratory, because high-energy particles can flip a chip's bits from 1 to 0 and silently corrupt a calculation.
My take
Strip away the "data center" branding and this is a materials-science experiment with a rocket ride — and that's what makes it interesting. The questions being asked are the right ones: do modern AI chips survive radiation without expensive hardening? Can you reject heat fast enough to be useful? How do you maintain something you can never touch again?
For the constellation vision to make sense, a few things have to be true that aren't yet: launch costs per kilogram have to keep falling, the laser communication links between satellites have to work at scale, and the economics have to beat just building on the ground, where technicians exist. None of that is proven. But the 8x solar number is real physics, and the power crunch on the ground is real economics. When both are true at once, "absurd" ideas get test flights.
Why this matters for Mars
This is a Mars story in disguise. Everything Suncatcher is testing — radiation tolerance, passive heat rejection, autonomous operation with no repair crew — is the exact checklist for computing on the surface of Mars, where the atmosphere is too thin for fan cooling, radiation is worse, and the nearest technician is 140 million miles away. The optical laser links Google wants between its satellites are the same technology deep-space missions need for high-bandwidth communication. If AI ever runs on Mars, its ancestors will have been debugged in low Earth orbit, fifteen minutes at a time.
Sources
- CNN: Google is firing the first salvo in a brewing race to deploy AI data centers in orbit
- Engadget: Google Is Sending A Teensy-Tiny AI Data Center To Space
- MarketWatch: Google is sending its first AI data center to space
- StartupFortune: Google will launch four TPUs into orbit on October 1 aboard a SpaceX rocket