For all the speculation about data centers in space, there aren’t many GPUs out there. As that begins to change, the business closest to orbital computing is starting to take shape.

The largest compute cluster currently in orbit was launched by Canada’s Kepler Communications in January, and has about 40 Nvidia Orin edge processors on 10 operational satellites, all interconnected by laser communication links.

The company now has 18 customers, and it announced its latest on Monday — SophiaSpace, a startup that will test software for its unique orbital computer aboard Kepler’s constellation.

Experts expect that we won’t see large-scale data centers like SpaceX or Blue Origin until the 2030s. The first step will be to process data collected in orbit to improve the capabilities of space sensors used by private companies and government agencies.

Kepler doesn’t see itself as a data center company, but as an infrastructure for applications in the space, CEO Meena Maitri told TechCrunch. It wants to be a layer that provides network services to other satellites in space, or to drones and aircraft in the sky below.

SOFIA, on the other hand, is developing passively cooled space computers that could solve a key challenge for large-scale data centers in orbit: keeping powerful processors from overheating without building and launching bulky, expensive active cooling systems.

In the new partnership, SOFIA will upload its proprietary operating system to one of Kepler’s satellites and attempt to launch and configure it into six GPUs on two spacecraft. Such activity is table-stacked in a ground data center, and this is the first time it will be tested in orbit. Ensuring the software works in orbit will be a key risk mitigation exercise for Sofia ahead of its first planned satellite launch in late 2027.

For Kepler, the partnership helps prove the utility of its network. Right now, it’s taking and processing data uploaded from Earth, or collected via host payloads on its spacecraft. But as the sector matures, the company expects third parties to begin connecting with satellites to provide networking and processing services.

Satellite companies are now planning future assets around this model, Mitry says, pointing to the benefits of offloading processing for more power-hungry sensors like synthetic aperture radar. The U.S. military is a key customer for this type of work as it develops a new missile defense system to predict and track threats from satellites. Kepler has already demonstrated a space-to-air laser link in a demo for the US government.

This type of edge processing—dealing with data where it is aggregated for faster response—is where orbital data centers will initially prove their value. That vision separates Sophia and Kepler from established space companies like SpaceX and Blue Origin, or startups like StarCloud and Aetherflux that are raising significant capital to focus on large-scale data centers with data center-style processors.

“Since we believe it’s more inference than training, we want more distributed GPUs that do the inference, rather than a superpower GPU that can accommodate the training workload,” Mitry told TechCrunch. “If this thing uses kilowatts of power and you’re only running 10% of the time, that’s not very helpful. In our case, our GPUs are running 100% of the time.”

And once these technologies are proven in orbit, well, anything can happen. Sophia CEO Rob DeMello said Wisconsin adopted a ban on data center construction last week, which some lawmakers in Congress are pushing for. Anything that limits data centers on Earth, in their view, makes space-based alternatives more attractive.

“There are no more data centers in this country,” DeMello mused. “It’s going to get weird from here.”