Orbital, a startup aiming to build a 10,000-satellite 'neocloud' in space, admits that the current economics of orbital data centers are not yet viable, according to go.theregister.com.
CEO Euwyn Poon told the outlet that while a constellation approach for AI inference makes sense, the cost of getting hardware into orbit remains a massive barrier. Currently, launching one kilogram into space costs approximately $7,000 via SpaceX’s Falcon 9 missions.
"Elon's stated goal is around $10 per kilogram, and we need to get there," Poon said. He noted that the company needs to see a reduction of several orders of magnitude in launch costs to make the solution economically feasible.
Engineering the space-based rack
While waiting for cheaper access to orbit, Orbital is developing a satellite roughly the size of four refrigerators. The unit is designed to house 100kW worth of computing gear.
Managing heat in a vacuum presents a significant challenge. To handle the thermal load, the satellites will utilize a solar array roughly the size of two halves of a tennis court and a radiator about half that size.
Orbital is focusing on the 100kW scale because larger units would require unwieldy solar panels and radiators. Poon described the process of redesigning chip cooling for space as a "fun and interesting challenge."
For its proof-of-concept flight planned for next year, the startup intends to use Nvidia’s Space-1 Vera Rubin module. The test flight will also evaluate radiation hardening technology to protect against cosmic particles that can corrupt data or damage hardware.
Reliability is critical since hardware repairs in orbit are currently impossible. The company's strategy also relies on competitors like SpaceX and Blue Origin providing spare launch capacity to Orbital, despite those companies planning their own orbital data centers.
