Nvidia delays next-gen AI rack system launch to 2028 due to manufacturing issues
Nvidia Corp., a prominent player in the artificial intelligence chip sector, recently experienced a significant setback regarding its anticipated Kyber rack-scale architecture. Initially scheduled for release in 2027, the debut has been postponed to 2028 due to manufacturing complications, according to insights from a media source. This development raises concerns about Nvidia’s product trajectory and its ability to maintain its competitive edge in a rapidly evolving market.
The Kyber architecture is designed to accommodate 144 of Nvidia’s advanced Rubin Ultra chips within a single server cabinet, facilitating a collective computing power necessary for AI firms to effectively train and implement sophisticated models. The architectural innovation features vertically stacked graphics processing units, enhancing density while minimizing latency—a vital factor for high-performing AI systems.
The delay is attributed to difficulties associated with producing a critical multi-layer printed circuit board, which is essential for the function of the system. Research firm SemiAnalysis elaborated on this issue, indicating that the Kyber NVL144 architecture’s postponement stems from challenges in manufacturability, particularly with the midplane circuitry that interlinks electronic modules.
In addition, another larger system known as NVL576, which aims to connect eight racks through optical links, faces similar uncertainties regarding its release timeline and volume capacities. The situation highlights mounting pressures across Nvidia’s diverse portfolio, suggesting that its ambitious annual release schedule may be colliding with practical manufacturing constraints.
Furthermore, an alternative plan to merge two existing Nvidia racks to replicate the desired computational power was reportedly rejected by cloud service providers, who viewed the design as cumbersome and too costly. This outcome has left Nvidia without a viable solution to enhance the capabilities of the Rubin Ultra architecture.
Despite these challenges, experts maintain that such delays are not necessarily catastrophic for Nvidia’s long-term positioning in AI infrastructure development. Observers point out that Nvidia has navigated comparable hurdles previously and often found solutions through collaboration with suppliers. Industry projections also indicate Nvidia’s data center revenue may exceed Wall Street expectations in the near term, suggesting resilience amidst present difficulties.
Meanwhile, global competitors, including Advanced Micro Devices and Google, are positioning themselves to capitalize on Nvidia’s setbacks, particularly in acquiring contracts with leading AI research entities.
In navigating this complex landscape, experts assert that Nvidia’s ongoing developments aim to remain relevant in the fierce competition for AI dominance, even as potential limitations in power supplies prompt the industry to reconsider its foundational growth strategies.
#business #technology
