A small experimental satellite named MVP is set to carry Google's AI hardware beyond Earth's atmosphere for the first time.

Aboard SpaceX's Falcon 9 rocket, the spacecraft is scheduled to launch on Thursday, October 1, 2026, as part of Google's Project Suncatcher, an initiative the company has described as a step toward one day running artificial intelligence data centers in orbit.

The mission's passenger is not a full-scale orbital data center. MVP is a potential precursor to one: a satellite packed with four tensor processing units, the specialized chips Google designs for AI workloads. Together, those chips carry roughly the computing power of a single server sitting in a terrestrial data center.

One kilowatt from sunlight

Power for the experiment will come entirely from the satellite's solar panels, which will supply one kilowatt, sufficient, according to Google, for the scale of this test.

The satellite is expected to remain operational for one year, during which Google will study how its AI chips hold up in the environment of outer space.

That one-year window is designed around a narrow but meaningful capability.

While in orbit, the tensor processing units will be able to process short queries for up to 15 minutes at a stretch while Google's Gemini AI generates responses.

Once each 15-minute period ends, the chips must power down and cool off through a dedicated cooling system, a necessity because the fans that keep Earth-bound computers from overheating simply do not work in space.

Why cooling dominates the challenge

In a blog post laying out Project Suncatcher, Google identified thermal management as one of the defining problems for computing in orbit.

Tensor processing units generate a large amount of heat concentrated in a small area, and that heat must be diffused safely or the chips risk burning out.

On Earth, air moving across components carries heat away. In the vacuum of space, there is no airflow at all; heat can only be shed through radiators, which Google says demand a fundamentally different approach to cooling electronics than anything used in terrestrial data centers.

The satellite's story does not end when its year of operation concludes. After shutting down, MVP will continue circling the planet for a total of six years before Earth's gravitational pull drags it back down.

As it falls, the satellite will burn up passing through the atmosphere.

Space-Based AI Infrastructure

Google’s Project Suncatcher is exploring the use of space-based infrastructure to support large-scale artificial intelligence computing.

The concept involves solar-powered satellites equipped with AI processing chips and connected through high-speed optical links.

The approach is designed to use the abundant solar energy available in orbit while reducing reliance on terrestrial infrastructure.

Future satellite constellations could work together to process larger AI workloads, although significant challenges remain in areas such as thermal management, radiation protection, communications, and reliable operation in space.

Simple prompts, distant goal

For all the momentum behind the concept, both companies acknowledge the idea is in its infancy.

The MVP satellite's capabilities amount to answering simple AI prompts from space, a capability that stands far short of a full-scale data center operating above the atmosphere.

The gap between this experiment and the stated end goal is substantial. If the current round of experiments demonstrates that the project is feasible, it would likely be years, if not decades, before the orbital data center visions held by Google and Musk become reality.

What Thursday's launch offers, in the meantime, is a first empirical test: whether specialized AI chips can endure and function in the punishing conditions of orbit and whether the cooling constraints of a vacuum environment can be managed well enough to let those chips do meaningful work.

MVP's year in space, powered by a single kilowatt of solar energy and limited to quarter-hour bursts of computation, will provide Google's engineers with their earliest answers to those questions.

The results will help determine whether the path from a four-chip testbed to an orbital AI data center is one that can realistically be walked and how soon.