Nvidia, a leader in artificial intelligence (AI) semiconductors, may face delays in the launch of its next-generation AI server due to manufacturing challenges. Analysts suggest that if the development of servers that utilize high-performance AI chips is postponed, it could present opportunities for competitors like AMD and Google.
According to reports from CNBC and other outlets on July 6, semiconductor analysis firm SemiAnalysis predicts that the release of Nvidia's next server system, the 'Kyber NVL144,' could be delayed by more than 12 months, potentially pushing its launch to 2028.
The Kyber NVL144 is designed to connect 144 next-generation AI chips, known as 'Rubin Ultra,' to function as a massive computer.
The primary reason for the delay is identified as issues with printed circuit board (PCB) manufacturing. PCBs are essential boards that connect chips and components within the server.
SemiAnalysis noted that Nvidia's alternative structure, the 'NVL72x2,' was canceled due to significant operational burdens for cloud customers.
Additionally, the larger 'NVL576' system may also face delays or be produced in limited quantities. This system connects multiple server systems via optical communication, but it requires a high level of technical expertise to ensure stability when linking various devices.
There are also concerns that the performance of the Rubin Ultra chips may fall short of initial targets. SemiAnalysis reported that the design featuring four dies, which are critical components for processing, has been scrapped, leaving only a design with two dies.
However, Nvidia has not officially acknowledged these delays. According to CNBC, the company stated that there are no changes to its product roadmap. The key question remains whether Nvidia can mass-produce its next-generation AI server as planned.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.