During a recent keynote at the Nvidia GTC conference in San Jose, California, Jensen Huang, the CEO of Nvidia Corporation, showcased a Vera Rubin Ultra Kyber Compute Tray alongside the Vera Rubin Ultra Kyber NVLink MidPlane. However, disappointing news emerged regarding Nvidia's future product lineup. The Kyber rack-scale architecture, which is designed to accommodate the highly anticipated 2027 Rubin Ultra chips, has experienced a significant delay of over a year, now set to launch in 2028 as reported by research firm SemiAnalysis. This delay raises concerns about the AI powerhouse's overall product strategy.
The Kyber system is essentially a server cabinet that integrates 144 of Nvidia's most advanced chips into one cohesive unit, enabling the computational power required for AI firms to train and deploy cutting-edge models efficiently. Its innovative design positions graphics processing units in vertical compute trays, which enhances density and minimizes latency. Originally, it was expected to roll out alongside Nvidia's next-generation rack-scale system, the Vera Rubin Ultra, in 2027.
According to SemiAnalysis, the delay is linked to challenges in manufacturing a crucial circuit board that is integral to the system's function. They noted, "The delay for the Kyber NVL144 rack architecture has been postponed to 2028 due to ongoing manufacturability issues with the PCB midplane." This multi-layer printed circuit board is essential for connecting various electronic modules within the setup.
In addition, there's a possibility that the larger NVL576 system, which connects eight racks via optical links, may also face delays or be limited in production capacity.
Nvidia has yet to respond to requests for comments regarding these developments. The delay further intensifies existing concerns about Nvidia's ambitious annual release strategy, which seems to be straining against current manufacturing capabilities.
Furthermore, a backup plan to combine two of Nvidia’s existing racks for comparable power has been abandoned following unfavorable feedback from cloud service providers. SemiAnalysis revealed that this approach was ultimately canceled due to significant pushback from cloud service providers and hyperscalers, citing its cumbersome design and operational complexities.
This situation leaves Nvidia without a viable strategy to enhance the scalability of the Rubin Ultra, which could present an unexpected opportunity for competitors like Advanced Micro Devices and Google, who have developed in-house chips and have already begun to capitalize on contracts from leading AI labs.
Currently, Nvidia's existing Rubin systems are in full production and are scheduled to begin shipping this fall to eight cloud partners, including Amazon Web Services, Microsoft Azure, and Google Cloud. SemiAnalysis also forecasts that Nvidia's data-center computing revenue will surpass Wall Street expectations by 20% during the latter half of fiscal 2027.
In premarket trading, Nvidia's shares showed some volatility, dipping slightly by less than 0.1% to $194.79.
