What's Happening?
Nvidia has announced a delay in the release of its Kyber NVL144 rack system, now expected in 2028. The delay is attributed to manufacturing challenges with the PCB midplane, a critical component that connects vertical compute trays. This postponement
affects infrastructure teams relying on the system for large-scale model training, as it narrows validated scale-up options. The interim NVL72x2 rack architecture was also abandoned due to operational complexity concerns from cloud customers. This development has led to a decline in Asian PCB and supplier stocks, reflecting market concerns over the delay.
Why It's Important?
The delay in Nvidia's Kyber NVL144 rack system is significant as it impacts the planning and procurement strategies of hyperscalers and training clusters that depend on this technology for AI model training. The postponement could lead to a bottleneck in capacity expansion for AI infrastructure, affecting the rollout of advanced AI applications. The decision to abandon the NVL72x2 interim solution further complicates the situation, leaving infrastructure teams with fewer options to meet their scaling needs. This delay underscores the challenges in manufacturing complex AI hardware and its ripple effects on the tech industry.
What's Next?
With the delay in the Kyber NVL144 rack system, infrastructure teams will need to reassess their procurement and capacity planning strategies. Nvidia's future announcements regarding alternative solutions or adjustments to the NVL576 system will be closely monitored. The market will also watch for any shifts in supplier strategies or new partnerships that Nvidia might pursue to address the manufacturing challenges. The delay may prompt other tech companies to accelerate their own developments to fill the gap left by Nvidia's postponed release.
