Google's First Suncatcher Orbital Data Center Test Launches October 1

Google will launch its first experimental satellite for Project Suncatcher, the company's orbital AI data center "moonshot," on October 1.

The satellite, called MVP, is roughly the size of a refrigerator. It carries four of Google's custom TPU AI accelerators — the same class of chips used on Earth to train AI models and run inference for token generation. Terrestrial data centers run thousands of these chips at a time, consuming enormous amounts of power. That consumption is one of the core reasons Google wants to move solar-powered AI hardware into space.

The stakes are considerable. AI boosters including Elon Musk and Jeff Bezos have pitched orbital data centers as an alternative to terrestrial facilities, which have drawn resistance in places like rural America. Google announced Suncatcher last year as its own entry in that race, framing it as a plan to build a constellation of AI satellites that could sate AI's hunger for power without the siting conflicts that increasingly slow ground-based construction.

MVP starts small. According to the New York Times, its solar panels supply only about one kilowatt of power — enough to run a microwave or a hair dryer. That figure alone signals how far the concept sits from a production data center today. But the goal of this first flight is validation, not capacity: Google wants to test whether its vision for orbiting AI compute holds up outside the lab.

Google did not build the spacecraft from scratch. The AI hardware inside is Google's own, but the satellite platform comes from Planet Labs, a satellite imagery firm. The original plan called for two custom satellites launching in 2027. Google wanted to move faster, so instead of waiting, the company integrated its chips into a satellite Planet Labs had already built for an early test.

That shortcut tells its own story. By riding on existing hardware rather than commissioning purpose-built spacecraft, Google trades engineering elegance for schedule — a sign the company treats orbital compute as a race worth entering now rather than a research question to defer.

The test also lands amid broader skepticism. Ars Technica's own reporting on orbital data centers has asked whether the concept is "economically viable" at all, given launch costs, thermal management, and the kilowatt-scale power budgets of even ambitious early designs. MVP's one-kilowatt array is a reminder of the gap between a demonstration and a constellation capable of meaningful AI training or inference throughput.

Still, October 1 gives Google a first real data point. If the TPU payload operates as intended in orbit, the company will have evidence that its chips can run in the space environment — the foundational step toward the two-satellite mission originally planned for 2027 and, eventually, toward a full AI satellite constellation. If it fails, Google will learn that lesson for the price of one refrigerator-sized spacecraft rather than a custom-built fleet.