SpaceX Plans to Launch First Batch of Orbital AI Data Centers in 2027
Elon Musk has provided a rough timeline for his deployment plan of up to one million optimized AI workloads in orbital data centers. Musk posted on X on Monday that SpaceX has collaborated with Nvidia to design a space optimized version of the Vera Rubin NVL72 system, which is scheduled to be launched into orbit in the fourth quarter of 2027 and achieve “significant scale” in 2028. He said, “Our design is simpler, cheaper, denser, and lighter than traditional racks.”
Musk’s schedule announcement coincides with SpaceX’s commitment to deploying Nvidia Vera CPUs to accelerate its next-generation proxy AI workloads. In addition to expanding the AI infrastructure of the Grok platform, SpaceX will also leverage Nvidia’s support for the first generation “Starmind” AI satellite based on the optimized Nvidia Vera Rubin NVL72 system.
Targeting AI optimization
Nvidia explained that the overall goal is for SpaceX to utilize Vera CPUs to accelerate application workloads, enabling AI agents to “act faster while maintaining continuous GPU supply and full utilization. This comes at a time when the expected growth of proxy AI will change the form of network traffic – AI agents will have conversations with other AI agents and need to extend the session time. Cisco recently emphasized that AI agents generate approximately 450% more traffic on average than humans performing the same tasks.
Nvidia claims that Vera is the first CPU designed for AI agents, with a task completion speed improvement of up to 1.8 times compared to x86 CPUs. Ian Buck, Vice President of Very Large Scale and High Performance Computing at Nvidia, said: “Proxy AI requires a new type of computing system – built not only to generate answers, but also to take action.” Nvidia has stated that it is optimizing Vera’s power, thermal management, and reliability for its orbital data centers.
Relying on Starship
SpaceX’s orbital data center plan clearly requires far more than Nvidia chips. The plan for rapidly deploying AI satellites also depends on Starship, SpaceX’s reusable heavy rocket system. After the latest test flight of Starship in July, SpaceX plans to perform the first tower capture of the Starship’s upper deck in the next test mission.
Starmind Project
As for Starmind, the first generation AI1 model is 20 meters tall with a wingspan of 70 meters, supporting peak power of 150 kilowatts and an average calculated power of about 120 kilowatts, with a delay of about 3 milliseconds. In June, Musk stated that the reference design for AI1 can support both Nvidia GB300 and the new Vera Rubin platform, and SpaceX will also provide reference designs for TPU chips designed for Google.
Other companies are also advancing orbital data center projects, including Google, Blue Origin, and Nvidia supported Starcloud. The startup company Orbital, which is developing an orbital AI inference platform, recently submitted a plan to launch 100000 satellites.