Nvidia's Kyber AI rack has been postponed to 2028.

      Nvidia's upcoming flagship AI machine has encountered a setback due to a single circuit board. The Kyber rack, which is designed to accommodate its 2027 Rubin Ultra chips, has been postponed to 2028. Research firm SemiAnalysis brought attention to the delay, and CNBC reported on it. Kyber is not a chip; it is a server cabinet that consolidates 144 of Nvidia’s most powerful GPUs into a single unit, allowing them to function like a colossal computer. This density is essential for training and running the largest AI models.

      One board, one major issue

      The delay arises from a less glamorous section of the setup. Kyber positions its chips in vertical trays to maximize capacity and reduce latency. Central to this arrangement is a multi-layer circuit board referred to as the PCB midplane. SemiAnalysis noted that this board “remains challenging from a manufacturability standpoint,” meaning Nvidia is currently unable to produce it at scale.

      The ripple effects extend from this point. A larger system called the NVL576 connects eight racks via optical cables, and it is likely facing delays or constrained to very low production volumes. Nvidia has not commented on this situation.

      No alternative plan

      Nvidia had considered a backup option. It intended to combine two current-generation racks to achieve comparable power. However, cloud customers rejected this idea due to the “odd design and heavy operational burden,” prompting Nvidia to abandon the plan.

      This leaves a gap, as SemiAnalysis states that Nvidia currently has “no proven solution” to scale its most powerful Rubin Ultra systems. For a company that typically introduces a new architecture every year, missing a step in its development is unusual.

      An opportunity for competitors

      This timing benefits competitors. Google and AMD are already securing contracts with leading AI labs using their own chips. A stumble at the high end gives them a rare technical advantage, notes SemiAnalysis. Additionally, it contributes to the existing unease within the AI chip market. Stocks related to Asian technology and circuit boards decreased following the report, according to Bloomberg.

      Despite this, Nvidia's immediate dominance remains intact. Its current Rubin systems are in full production and are set to ship this autumn to eight cloud partners, including AWS, Microsoft Azure, and Google Cloud. SemiAnalysis even predicts that Nvidia’s data center compute revenue will surpass Wall Street expectations by 20 percent, covering the second half of its 2027 financial year. Nvidia's stock barely fluctuated.

      The real challenge

      The underlying issue is manufacturing. Nvidia's annual release schedule is colliding with the production limits of its suppliers. This includes everything from advanced circuit boards to the memory that powers each GPU.

      The factories responsible for assembling these racks are already operating at full capacity. Competitors developing their own custom silicon are betting that the pace of production, rather than design, is where Nvidia is most vulnerable.

      This time, the toughest aspect of the AI boom is not the chip itself; it is the enclosure that houses it.