SpaceX will build its planned Starmind orbital data centers around Nvidia's Vera Rubin NVL72 rack-scale systems, with Elon Musk indicating that a slimmed-down, space-optimized version of the hardware could reach orbit by the fourth quarter of 2027.

The project marks a significant step toward establishing persistent AI computing infrastructure beyond Earth and positions Nvidia's accelerated computing architecture as the unified foundation for both terrestrial and space-based operations.

Each Vera Rubin NVL72 rack combines 72 chips that operate as a single coherent computer. According to Nvidia, individual chips deliver up to 25 times the compute performance of the previous-generation H100. The orbital configuration is being adapted specifically for the space environment; Musk has described the design as significantly simpler, denser, and lighter than conventional data center hardware, with modifications to address radiation exposure and thermal management in orbit.

SpaceXAI, the company's artificial intelligence division, plans to run its full suite of workloads on the platform, ranging from the Grok conversational model to the control systems governing its orbital fleet. Musk has publicly characterized the Vera Rubin system as the best AI computer currently available, underscoring how the same hardware stack is being positioned for both large-scale terrestrial training and eventual orbital deployment.

From Terrestrial Gigawatts to Orbital Agentic AI

The partnership covers both the rack-scale NVL72 system and Nvidia's Vera CPU, which SpaceXAI will use to accelerate the orchestration layer of its agentic AI workloads. The Vera processor incorporates 88 custom Olympus cores, NVIDIA Spatial Multithreading technology, and high-bandwidth LPDDR5X memory capable of delivering up to 1.2 terabytes per second of bandwidth.

According to Nvidia, this architecture achieves approximately 1.8 times faster task completion than traditional x86 CPUs across agentic AI, reinforcement learning, and data processing tasks. As SpaceXAI expands its infrastructure toward gigawatt-scale capacity, the Vera Rubin platform—integrating NVLink interconnects, Spectrum-X Ethernet networking, BlueField data processing units, and supporting software—is designed to operate as a consistent architecture across both planetary AI factories and orbital satellites.

That consistency matters because orbital compute is not being introduced as a separate experiment so much as an extension of the same stack Nvidia is pushing across its data center roadmap. By tying the space project to Vera Rubin and Vera CPUs, SpaceXAI is aligning its orbital plans with a platform that already spans networking, accelerators, and orchestration components, which helps explain why the announcement is being read as more than a one-off hardware test.

The financial viability of orbital compute remains an open question. Industry analysts currently estimate that space-based processing costs more than four times as much as equivalent ground-based capacity. Musk has contested this outlook, predicting that the cost equation will invert within the next few years and favor orbital deployment as AI infrastructure requirements continue to grow.

The announcement comes at a moment of intensifying scrutiny over terrestrial data center construction and follows earlier skepticism from OpenAI chief executive Sam Altman, who publicly described space-based data centers as ridiculous earlier this year. With corporate timelines firming and Nvidia now formally adapting its architecture for orbital constraints, space-based AI infrastructure is shifting from a theoretical proposition to a near-term operational target. The late-2027 timeline also gives the sector a concrete checkpoint for gauging whether the engineering and operating assumptions behind orbital AI can move from concept to deployment.

Source: Metaverse Post