
The space-based artificial intelligence data center project long envisioned by SpaceX Chief Executive Elon Musk is set to take its first step next year. A satellite carrying Nvidia's next-generation AI chip, Vera Rubin, will head to space in the fourth quarter of 2027, with plans to scale up the effort in earnest from 2028.
Musk said on X, formerly Twitter, on the 24th local time that SpaceX had developed the Vera Rubin NVL72 system in partnership with Nvidia, targeting a launch in the fourth quarter of next year, and that the business would reach a substantial scale by 2028. Each SpaceX AI satellite serves as a single server making up a space data center, while the data center itself is infrastructure formed by gathering thousands or more of these AI satellites.
The Vera Rubin NVL72, the rack for AI satellites jointly developed by SpaceX and Nvidia, is a redesign of a ground-based version adapted for the space environment. Musk explained that the rack has a simpler structure, lower cost and higher density relative to weight than existing ground-based racks.
Nvidia also announced that it would load computing equipment based on the Vera Rubin NVL72 onto SpaceX's first AI satellite, Starmind AI1. This marks the first attempt to realize the concept of an orbital data center, which processes data computations directly in satellite orbit rather than on the ground.
Musk first unveiled the design concept for a solar-powered space AI data center in a video released last June ahead of SpaceX's planned listing. The computing capacity of a single satellite presented at the time was up to 150 kilowatts, comparable to one of Nvidia's ground-based GB300 racks, with a wingspan reaching 70 meters.
According to his roadmap, the goal is to build 1 gigawatt of annual AI computing infrastructure in space by the end of next year, then expand it tenfold each year to ultimately reach the terawatt scale. To this end, SpaceX has applied to the U.S. Federal Communications Commission for approval to launch up to 1 million satellites for space data centers.

The challenge is cost. According to energy market research firm Wood Mackenzie, building a 1-gigawatt space data center is estimated to cost about $170 billion (approximately 235 trillion won). That is more than three times the cost of a ground-based data center of the same scale. Since about 60% of the total cost goes into building and launching the satellites, analysts say current costs must be cut by about 70% for the project to be economically viable.
In response, Musk projected that the reusable heavy-lift launch vehicle Starship would sharply drive down launch costs, and that the point at which operating AI in space becomes cheaper than on the ground would arrive far sooner than the market expects.
As ground-based data centers struggle to secure power, cooling water and land, a competitive landscape is also forming around space data centers. Still, technical hurdles remain to be overcome, including radiation exposure, thermal management, building communications infrastructure and maintenance. With Musk laying out a specific timeline for the first launch next year, the space data center concept, once regarded as science fiction, has moved into a phase where its actual business viability will be tested.






