According to research firm SemiAnalysis, NVIDIA's Kyber rack-scale architecture, which is designed to integrate 144 of its Rubin Ultra chips into a single unit for enhanced AI processing, has been delayed by over a year to 2028. This delay is attributed to difficulties in manufacturing a critical multi-layer printed circuit board essential for the system's functionality.
Additionally, a proposed workaround involving the combination of two existing racks has been scrapped due to negative feedback from cloud service providers, who found the design impractical and costly.
This situation raises concerns about NVIDIA's ability to maintain its rapid product release schedule amidst manufacturing constraints, potentially allowing competitors like Advanced Micro Devices and Google to gain an advantage in the high-end AI market.
Despite these setbacks, NVIDIA's current-generation Rubin systems are set to begin shipping this fall to major cloud partners, and projections indicate that the company's data-center compute revenue could exceed Wall Street expectations by 20% in the latter half of fiscal 2027. Following the news, NVIDIA's shares experienced slight fluctuations, trading down less than 0.1% at $194.79