What happened

Google says Project Suncatcher, its plan to run AI data centers on satellites, is about to fly real hardware for the first time. In a post published on September 24, 2026, Travis Beals, Senior Director of Paradigms of Intelligence at Google, wrote that the project "is scheduled to embark on its first test in orbit, launching a prototype satellite to evaluate how Google Tensor Processing Units (TPUs) perform in space."

The satellite was built with Planet (Planet Labs PBC). It rides on SpaceX's Transporter-18 rideshare, one payload among many. Google's post says only "next week". SiliconANGLE reports a launch on October 1 aboard a Falcon 9 from Vandenberg Space Force Base in California. Rideshare dates often slip, so treat that as the scheduled date, not a promise.

According to that report, the satellite is called MVP. It is about the size of a refrigerator, carries four Trillium TPUs and gets roughly 1 kW from its solar panels. The chips run in bursts of about 15 minutes before they have to cool down. Several outlets put its expected service life at around one year. None of those figures appear in Google's own post.

Why it matters

AI's hard limit is increasingly power, not chips. Suncatcher's pitch is to go where the power is. "In low Earth orbit, satellites can access near-constant sunlight, generating up to eight times more solar power than on Earth," Google writes. A satellite in the right orbit has no night, no clouds and no grid connection to wait years for.

Until now that pitch rested on lab work and a research paper from November 2025. MVP produces the first real data from orbit. It also came sooner than planned. When Google announced the moonshot on November 4, 2025, the next step was "a learning mission in partnership with Planet to launch two prototype satellites by early 2027." MVP is an extra single-satellite flight added ahead of that pair. The pair has not been dropped.

For anyone buying compute today, nothing changes. MVP sells no capacity, and Google makes no claim that it will serve users or run Gemini. Cloud TPU customers will see no difference. What it changes is the evidence: whether a data-center chip survives launch and orbit stops being a paper question.

Google is not alone. SiliconANGLE lists SpaceX, Starcloud, which raised money in 2026, and Andreessen Horowitz-backed Orbital among the companies chasing space-based AI compute. Google does have one advantage those rivals lack: it designs its own accelerators, so what it learns in orbit can shape the next TPU.

What the satellite is testing

Beals is clear that this flight is an experiment, not a demo: "This first launch is about seeing what works, identifying points of failure, and applying those findings to future missions." Google's post names three stresses the hardware has to survive.

The ride up

Reaching low Earth orbit takes about 10 minutes, with sustained acceleration of up to 10 g. Individual parts, the TPU chips included, can see 50 to 100 g. Google shook the satellite on all three axes at launch frequencies, and the team says it was "pleasantly surprised" that the hardware held up.

Radiation

The TPUs went into a proton beam at UC Davis's Crocker Nuclear Laboratory while running AI workloads, with the team logging errors such as bitflips. Google says its Trillium TPUs "can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission." That is not the same as error-free. In the 2025 research, the chip's high-bandwidth memory first showed irregularities after a cumulative 2 krad(Si). A shielded five-year mission is expected to deliver about 750 rad(Si), and testing went up to 15 krad(Si). The margin is real, but the memory is the weak point.

Cooling

This is the least-proven part. Space has no air, so a chip cannot be cooled by fans. Heat can only leave by radiating away. Google is testing "a combination of heat pipes and radiators to cool the chips", and so far has tried it only in a thermal vacuum chamber on the ground. The reported 15-minute run bursts show how tight the heat budget is. MVP is the first time the cooling system will run in orbit.

What's next

MVP is a single satellite with no laser links. A real space data center would need many satellites working as one machine. Future Suncatcher satellites would each carry dozens of TPUs and fly in tight clusters, linked by lasers that carry very high bandwidth over very short distances. Google compares the pointing accuracy needed to "hitting a coin-size target from miles away while both points are in motion." Its plan: "We'll test our work on this in 2027 when we put two satellites in orbit." That is the Planet-built pair from the original announcement, which Planet says will fly in tandem with high-bandwidth cross-links.

The long-range design is much larger. The November 2025 paper sketches an 81-satellite cluster with a 1 km radius at about 650 km altitude, in a dawn-dusk sun-synchronous orbit, with neighbors 100 to 200 meters apart. One ground-based link pair reached 800 Gbps each way, 1.6 Tbps in total. The economics depend on launch costs falling below $200 per kilogram by the mid-2030s, which is Google's own projection and not a quoted price.

So the milestones to watch are these: whether Transporter-18 flies on schedule, what Google reports about TPU errors and temperatures once MVP is working, and whether the two-satellite laser test stays on track for 2027. As of September 25, 2026, the launch has not happened. Google has also released a four-part video series that covers each problem in turn: machine learning in space, hardware survival, cooling and keeping a constellation connected.