Google will launch a refrigerator-sized prototype satellite carrying four custom Tensor Processing Units into low-Earth orbit on March 1, opening the first in-space hardware test for its Project Suncatcher effort to build AI data centers above the Earth.

The prototype, named MVP, will ride to orbit on a SpaceX Transporter-18 rideshare mission aboard a Falcon 9 rocket. Its job is blunt: establish whether Google's TPUs and the systems around them can survive launch and then operate through the radiation and thermal extremes of orbit.

The stakes for the broader concept are considerable. A solar panel in the right orbit generates five to eight times more power than the same panel on Earth, according to Travis Beals, senior director and Project Suncatcher lead. That advantage comes largely from a "dawn-dusk" orbit with nearly constant access to sunlight, combined with the cooling properties of space itself.

"A solar panel, in the right orbit, generates five to eight times more power than that same panel down on Earth," Beals said. The project targets nearly 24/7 solar power as an alternative to power-hungry terrestrial data centers.

The test satellite is a deliberately modest package. According to the New York Times, MVP carries four specialized TPUs — roughly the computing power of a single server in a data center — and supplies just one kilowatt of power to those chips. Google expects future orbital data centers to run on dozens of TPUs each, flying in linked clusters.

Surviving the ride

Launch loads are the first hurdle. Eric Stevens, director of systems engineering at imaging company Planet, one of Google's partners on the mission, described the forces involved: "It's a ten-minute ride on a rocket to low-Earth orbit, and during that the spacecraft is going to be exposed to tremendous vibration forces as well as sustained acceleration loads up to ten times the force of gravity."

"Individual components such as the AI chips can actually experience up to 50 to 100gs of amplified forces," Stevens added.

Google's earlier ground-based vibration testing showed its TPUs can hold up to forces as high as 100gs. Planet designed custom packaging to help the payload survive the ascent.

Once in orbit, MVP will answer simple AI queries and operate for a year, the Times reported. Beyond raw compute survival, Google will test connectivity between the satellite and the ground, and how to cool TPUs in vacuum conditions using radiation heat transfer.

Radiative cooling is standard practice for spacecraft, but the challenge here is scale. TPUs concentrate heat in a small area, and Beals told the Times that Google's chips can currently operate in space for about 15 minutes before needing to shut down for cooling.

"We're working on a number of different approaches for this, including a combination of heat pipes and radiators to cool the chips," Beals wrote in a blog post about the trial. "We'll test our work on this in 2027 when we put two satellites in orbit."

A crowded field forming

Google is not alone in chasing orbital compute. SpaceX, under its "Starmind" initiative, plans to launch up to 1 million orbital AI data satellites, with deployments starting in Q4 2027 and scaling through 2028. Blue Origin has outlined an orbital data center platform of nearly 52,000 satellites. Startup Orbital Inc. recently filed a plan with the US Federal Communications Commission to launch up to 100,000 AI data center satellites that would eventually deliver 10 gigawatts of orbital computing capacity. Nvidia-backed Starcloud is also active in the field.

The regulatory dimension is already in motion: Orbital Inc.'s FCC filing signals that large-scale orbital constellations for computing will move through the same spectrum licensing machinery as communications satellites, a process that historically takes years.

For now, Google's one-kilowatt, four-chip prototype remains a physics and engineering experiment rather than a commercial platform. The March 1 launch, a year-long orbital trial and the planned two-satellite test in 2027 will determine whether the power and cooling math of orbital data centers holds up outside the lab.