Tom's Hardware reports SpaceX is developing a 'Kessel' rack-scale system with 72 Nvidia Rubin GPUs and an 84-layer PCB midplane for data centers in space.

SpaceX and xAI are developing a custom version of Nvidia's Vera Rubin NVL72 rack-scale AI system, according to SemiAnalysis. The system, codenamed Kessel, replaces conventional NVLink cable assemblies with an 84-layer printed circuit board (PCB), partly for use in orbital data centers. Kessel keeps the NVL72's core configuration of 36 Vera CPUs, 72 Rubin GPUs, and nine NVLink switch trays, but uses a different solid-state interconnect architecture. SpaceX and xAI will continue to buy Nvidia hardware directly from the company, but they will source other components independently, which could give Elon Musk's company greater control over its hardware supply chain.

The Kessel system retains the basic layout of Nvidia's Vera Rubin NVL72 platform, which places 72 GPU packages across 18 compute trays and nine NVLink switch trays. Instead of Nvidia's conventional internal NVLink cable cartridges, SpaceX plans to position compute and switch trays back-to-back around a passive PCB midplane. The architecture resembles a smaller version of Nvidia's Kyber design, which has 576 AI GPUs (in 144 packages) and also uses a PCB-based interconnect rather than cable assemblies. According to the analysts, the design is intended for both terrestrial and orbital deployments.

The most distinctive feature of Kessel is its 84-layer PCB midplane, which uses ultra-low-loss M10 copper-clad laminates (CCLs). By comparison, Nvidia's Kyber relied on a considerably more complex 156-layer backplane built with M9 CCLs. Cutting the layer count by nearly half could substantially improve manufacturability and production yields, but the more advanced laminate will almost certainly raise material costs. Either way, an 84-layer PCB is exceptionally difficult to build. For context, modern high-end server motherboards typically use 15 to 20 layers, while server development boards use more than 20 layers to virtually eliminate electrical issues on the board itself.

Moving to a PCB midplane could simplify assembly by eliminating numerous high-speed cable connections, but it would not necessarily lower manufacturing costs. An 84-layer board built on ultra-low-loss M10 laminate is an expensive component that requires sophisticated manufacturing processes and precise quality control. Yields could also become a major obstacle, since a defect in any internal layer may render the entire board unusable. The economic viability of Kessel will therefore depend on whether savings from eliminating cables and procuring components independently can offset the cost of the complex midplane.

Perhaps more significant, Kessel departs from Nvidia's conventional system-level procurement model. Rather than buying Nvidia's assembled L6 compute boards or the L10 trays Nvidia plans to sell with its Vera Rubin generation, SpaceX reportedly intends to buy AI accelerators, CPUs, switches, and networking components directly from Nvidia and source other parts from third-party suppliers. It remains unclear how easily SpaceX and xAI can procure memory amid current shortages. This approach would let SpaceX choose its own PCB manufacturers, component vendors, and manufacturing partners, optimize hardware for its infrastructure (which will matter for orbital deployments), and potentially reduce intermediary costs. In exchange, the company would take on greater responsibility for system design, validation, manufacturing quality, and supply-chain management, a role most hyperscalers have been assuming in recent years.

For Nvidia, the implications are less clear-cut. Direct sales of CPU and AI GPU packages could reduce the company's revenue from complete compute-board assemblies. However, the arrangement could also encourage other large customers to build their own Nvidia-based systems and ultimately deploy more Nvidia hardware, which may matter as custom silicon from hyperscalers gains traction. For that to happen, though, Kessel must succeed decisively and offer clear advantages over systems Nvidia designs itself.