Nvidia’s most ambitious AI rack system just hit a wall. The Kyber NVL144, a next-generation rack-scale architecture that CEO Jensen Huang personally showcased at GTC just three months ago, has been pushed back by over 12 months to 2028 due to manufacturing problems with a critical circuit board component.
What went wrong
The culprit, according to semiconductor analyst firm SemiAnalysis, is the midplane printed circuit board. That’s the piece of hardware responsible for enabling the dense all-copper NVLink interconnect that makes the Kyber NVL144 tick.
Huang stood on stage at GTC and presented this architecture to the world roughly three months before the delay was announced on July 5, 2026. That timeline suggests Nvidia either didn’t know about the severity of the issue or believed it could be resolved quickly.
Making matters worse, the associated NVL72x2 back-to-back rack design has been outright canceled. That configuration was meant to serve as a bridge, offering expanded scalability by linking two NVL72 racks together. Its cancellation effectively caps the scaling potential of Nvidia’s current-generation systems until the company can deliver the Kyber architecture or find an alternative path forward.
