TL;DR — Key Takeaways
- SpaceX plans to launch its first AI-focused orbital data centers in the fourth quarter of 2027, accelerating its previously indicated 2028 timetable.
- The company has asked the FCC for authority to deploy as many as one million AI-capable satellites, signaling a potentially massive long-term orbital computing network.
- SpaceX plans to base its first-generation Starmind satellites on a space-optimized version of NVIDIA’s Vera Rubin NVL72 architecture.
Skeptics have dismissed orbital data centers as mere science fiction, but SpaceX is planning to launch its first AI data centers into orbit in the fourth quarter of 2027, with CEO Elon Musk forecasting that the system will attain “significant scale” in 2028.
In fact, the 2027 target moves SpaceX’s timetable forward. The company had previously indicated that orbital data centers could begin launching in 2028.
The project could shift a significant amount of AI computing from data centers on Earth to satellites. SpaceX has filed an application with the FCC seeking approval for as many as one million satellites capable of processing AI workloads.
SpaceX has not announced how many AI satellites it expects to launch in 2027 or 2028, and the application for one million appears to support a longer-term ambition rather than a near-term plan. Still, a network approaching one million satellites would turn orbital computing from a specialized experiment into a new class of AI infrastructure.
NVIDIA’s Vera Rubin
SpaceX plans to use a space-optimized version of NVIDIA’s Vera Rubin NVL72 system for its first generation of Starmind AI satellites. The companies are adapting the hardware to contend with the power and reliability requirements of computing in orbit. Musk said in an X post that the orbital system will be “simpler, lower cost, denser and lighter than a traditional rack.”
The Vera CPU is built specifically for AI agents, and NVIDIA claims Vera can complete AI agent tasks up to 1.8 times faster than x86 CPUs. The processor is designed to keep GPUs supplied with data, which helps avoid idle GPU capacity as it handles AI workloads. This could support the intense demands of space-based computing.
SpaceX’s AI1 satellite design highlights the scope of the company’s plan. The model is expected to stand 20 meters high with a 70-meter wingspan. It is designed for peak power of 150 kilowatts, with about 120 kilowatts available for average compute workloads. Expected latency is approximately three milliseconds.
SpaceX will need the launch capacity to deploy massive numbers of satellites, making the success of its reusable Starship rocket a critical piece of the project.
The company’s 2027 launch plans support its overall effort to make orbital computing a major part of its business. SpaceX has aggressively promoted space-based data centers as an alternative to Earth-based facilities, which require enormous amounts of electricity, cooling equipment and physical space.
SpaceX has plenty of ambitious competitors. Orbital has filed an application to launch up to 100,000 satellites. Blue Origin and NVIDIA-backed Starcloud are also pursuing space-based data center technology, and Google is developing its Suncatcher orbital computing initiative.