Week 47 · 2026Friday, 20 November 2026
Technology

SpaceX and xAI Developing Rack with 72 Nvidia Rubin Accelerators

In the Kessel project, compute modules and switches are planned to be connected via an 84-layer board instead of cable assemblies.

12 Nov, 18:11Read in Russian
ixbt.com

According to SemiAnalysis, SpaceX and xAI have based their design on Nvidia's Vera Rubin NVL72 server platform. The rack will retain its configuration with 72 Rubin graphics accelerators, 36 Vera central processing units, and nine NVLink switch units.

Compute modules and switches will be placed on opposite sides of a shared board. This connection could simplify rack assembly, but the board itself is complex to manufacture. A defect in one of the internal layers could render the entire board unusable.

According to analysts, SpaceX plans to purchase accelerators, processors, switches, and network components directly from Nvidia, while ordering the remaining parts from third-party suppliers. This will give the company greater flexibility in selecting board manufacturers and assembly partners.

SemiAnalysis reports that the rack is being designed for ground-based data centers and potential deployment in orbit. The timeline for bringing Kessel into operation has not yet been announced.