NVIDIA's Kyber NVL144 Delayed 12 Months Due to Manufacturing Challenges

NVIDIA's Kyber NVL144 rack has been delayed for over a year due to issues with the printed circuit board (PCB) midplane. This delay may impact the company's AI market position and provide opportunities for competitors. The exact reasons and implications are still unclear.

SemiAnalysis reports that NVDA's next-generation Kyber NVL144 rack system has been delayed by more than 12 months, pushing its launch out to 2028, due to manufacturing difficulties with the rack's PCB midplane. Several other outlets, including reports citing Nvidia directly, have corroborated the roughly 12-month slip.

The midplane is reportedly one of the most complex printed circuit boards ever designed for a commercial computing product, using a dense, all-copper NVLink interconnect across dozens of layers so the rack's chips can operate as a single, massive compute unit. That complexity is described as the direct source of the production bottleneck, rather than a chip-design or yield problem.

The delay carries competitive weight: it opens a window for AMD, whose Instinct GPU and rack roadmap has been closing the gap on Nvidia's flagship systems, and for hyperscalers' own custom-silicon programs, including Broadcom-designed accelerators used by Google and other cloud providers, to gain relative ground while Nvidia's top-tier rack slips. GuruFocus and CNBC report the Kyber NVL144 is now expected to launch in 2028.

For hyperscalers, the slip also raises capex-timing risk: cloud providers that had budgeted for Kyber-based deployments may need to rephase spending plans or lean more heavily on current-generation Nvidia racks and third-party alternatives in the interim, a scheduling risk that could ripple through AI infrastructure capital spending over the next two years.

Related Stocks

Powered by SentiSense - Intelligent Market Analysis