Nvidia might have seen massive success as the backbone of the AI revolution, but today there is some bad news for the company. The firm’s Kyber NVL144, a next-generation AI rack system, has been pushed back to 2028 due to difficulties the company is experiencing manufacturing its key circuit board.
The Kyber NVL144 is a server cabinet that will come with 144 Nvidia Rubin Ultra chips (due in 2027). The Kyber NVL144 turns the chips into a single unit so they can work together, giving AI companies the power they need to train and run the most advanced models. With the 12-month setback, however, it could give AMD and Google the opportunity to leap ahead.
The report of delays comes from SemiAnalysis, which said: “Kyber NVL144 rack architecture has been delayed to 2028 as the PCB midplane remains challenging from a manufacturability standpoint.” SemiAnalysis added that NVL576 is also likely delayed or at least limited to small volumes.
It has also been reported that Nvidia’s backup plan, to connect two of Nvidia’s current racks together for similar power, has also been scrapped due to the awkward and costly design. This leaves Nvidia in a very precarious position that its competitors could try to use to their advantage.
Apparently, Nvidia’s current Rubin systems are in full production and will begin shipping them in the fall to eight of its cloud partners, including Amazon Web Services, Microsoft Azure, and Google Cloud. Interestingly, SemiAnalysis also said that Nvidia’s data-center compute revenue will run 20% above the Wall Street consensus in the second half of its fiscal year 2027, so it’s not all bad news.
Source: SemiAnalysis (X)