Nvidia's Kyber NVL144 reportedly pushed back more than a year, Asian suppliers drop

Editor's take

Nvidia's planned Kyber NVL144 AI server rack has been pushed back to 2028 due to difficulties in producing its complex circuit boards. This delay suggests that the manufacturing challenges for high-density, high-performance AI hardware are more significant and intractable than previously anticipated.

The impact extends beyond Nvidia, as it highlights the critical bottlenecks in the supply chain for advanced AI infrastructure. Companies reliant on cutting-edge AI compute, from cloud providers to enterprise AI developers, will need to recalibrate their roadmaps. This setback also underscores the immense engineering hurdles in scaling the physical components that underpin ever-more-powerful AI models, a stark contrast to the rapid software advancements seen with models like GPT-4.

Future attention should focus on whether Nvidia can overcome these specific circuit board manufacturing issues, or if alternative architectural approaches might emerge to circumvent such supply chain constraints. The ability of other hardware vendors, such as AMD with its upcoming Instinct MI300 series, to capitalize on any persistent Nvidia capacity limitations will be a key indicator of broader market shifts.