Nvidia Kyber AI rack delayed to 2028 as circuit board manufacturing hits roadblock

Nvidia Kyber faces over 12-month manufacturing delay

Nvidia Kyber, the company's next flagship AI rack system, has been delayed by more than 12 months to 2028, according to research firm SemiAnalysis 1

1

. The setback centers on manufacturing challenges with a specialized multi-layer circuit board called the PCB midplane, which serves as the critical connector between electronic modules within the system 2

2

. This AI rack system was originally slated to debut in 2027 as part of Nvidia's next-generation rack-scale architecture, designed to house the company's Rubin Ultra AI chips.

The Kyber system represents a significant leap in AI infrastructure scaling, packing 144 of Nvidia's most powerful GPUs into a single server cabinet that functions as one unified computing unit. The design mounts graphics processing units in vertical compute trays rather than horizontally to boost density and reduce latency, providing the computational horsepower needed to train and run the most advanced AI models 1

1

.

Manufacturing complexity derails Nvidia's product roadmap

The manufacturing delay stems from the PCB midplane's complexity, which SemiAnalysis described as "challenging from a manufacturability standpoint" 1

1

. This component ties together all the vertically mounted chips, and Nvidia's suppliers have struggled to produce it at scale 2

2

. The setback extends beyond the base Kyber NVL144 rack architecture to the larger NVL576 system, which links eight racks via optical connections. That more ambitious configuration is also likely delayed or limited to small production volumes, the research firm noted 1

1

.

Nvidia attempted to develop a backup plan by bolting two current-generation racks together to achieve similar performance. However, cloud service providers and hyperscalers rejected this approach, pushing back heavily against what they viewed as an "odd design and heavy operational burden" 1

1

. Nvidia has since cancelled that alternative, leaving the company without a proven solution to expand scale-up capabilities for Rubin Ultra systems.

Opening in high-end AI chip market for competitors

The delay creates a rare technical opening for rivals AMD and Google, whose in-house chips are already winning business from top AI labs 1

1

. For a company that maintains a breakneck annual release cadence, missing a critical milestone on Nvidia's product roadmap is unusual and signals mounting strains across its product lines. The situation underscores concerns that Nvidia's aggressive development schedule is colliding with fundamental manufacturing limits 1

1

.

The real challenge facing Nvidia extends beyond chip design to the factories assembling these complex systems, which are already operating at capacity. Competitors building custom silicon are betting that manufacturing pace, rather than design superiority, represents Nvidia's most vulnerable point 2

2

. Asian technology and circuit board stocks declined following the report, reflecting broader nervousness in the AI chip sector.

Near-term strength despite long-term uncertainty

Despite the Kyber setback, Nvidia's current-generation Rubin systems remain in full production and will begin shipping this fall to eight cloud partners, including AWS, Microsoft Azure, and Google Cloud 1

1

. SemiAnalysis projects that Nvidia's data-center compute revenue will run 20% above Wall Street consensus in the second half of fiscal 2027, demonstrating continued near-term dominance 1

1

. Nvidia shares fluctuated minimally in premarket trading, last down less than 0.1% at $194.79 1

1

.

The question facing AI companies and cloud service providers is whether this gap in Nvidia's ultra-high-end offerings will meaningfully shift market dynamics. With the largest AI models requiring ever-greater computational resources, any delay in scaling capabilities could influence procurement decisions for 2027 and beyond. Nvidia did not respond to requests for comment on the reported delays 1

1

.