Nvidia denies report that Kyber NVL144 AI rack has been delayed to 2028

Nvidia pushed back this week on a report that its Kyber NVL144 rack-scale AI system had slipped to 2028, saying the company’s roadmap had not changed. The denial came after SemiAnalysis said Kyber, a next-generation rack design tied to Nvidia’s Rubin Ultra platform, had been delayed by more than a year because of manufacturing problems involving a key circuit board, according to CNBC and Yahoo Finance. (finance.yahoo.com) Nvidia responded with a short statement — “Our roadmap is intact” — and the stock rose about 1% on July 6 after the company’s rebuttal. ### What exactly was being disputed? SemiAnalysis said Kyber NVL144 had been pushed from 2027 to 2028, according to reports cited by CNBC and other financial outlets. The report said the issue centered on production of a specialized midplane board used in the rack-scale system. (nvidia.com) Nvidia did not address the manufacturing details in public. Yahoo Finance and other outlets reported only the company’s denial and its statement that the roadmap remained intact. ### What is Kyber NVL144 in Nvidia’s product stack? Nvidia presented Kyber as part of its Vera Rubin generation of AI infrastructure at GTC in March 2026. (cnbc.com) Nvidia’s own product materials describe the Rubin platform as a rack-scale and pod-scale system that combines GPUs, CPUs, networking, memory movement and power infrastructure into a single AI supercomputer design. The NVL144 label refers to a large multi-GPU rack configuration that sits above individual chips and servers in Nvidia’s lineup, according to the reports on the dispute. (cnbc.com) That matters because customers buying AI capacity at the hyperscale level are increasingly evaluating complete systems rather than standalone accelerators. (finance.yahoo.com) ### Why did investors react so quickly? Nvidia shares moved higher after the company denied the report, with multiple market reports describing a gain of roughly 1% on July 6. The market reaction suggested investors were focused less on a single rumor than on whether Nvidia’s annual cadence for new AI infrastructure was still credible. (nvidia.com) CNBC said the reported delay had raised questions about whether Nvidia’s rapid release schedule was running into manufacturing limits. Barchart’s syndicated report said Kyber is a central piece of Nvidia’s next infrastructure transition, which is why any suggestion of slippage drew immediate attention. (cnbc.com) ### Why does a rack matter as much as the chip? Nvidia’s own Rubin materials describe the platform as a system built from multiple purpose-built racks, not just a faster GPU. The company says the design is meant to reduce communication and memory bottlenecks across very large AI workloads, a point that has become increasingly important as cloud providers build larger training and inference clusters. (finance.yahoo.com) That is why the reported bottleneck, if it had been confirmed, would have mattered beyond a component issue. A delay at the rack level would affect how quickly Nvidia can ship complete AI infrastructure systems, not just silicon. CNBC attributed that concern to the SemiAnalysis report about manufacturability. (cnbc.com) ### What should readers watch next? March 16, 2026 was the date Nvidia used GTC to outline the Rubin platform and its rack-scale architecture. The next concrete signals are likely to come from Nvidia product updates, partner deployment announcements and any future disclosures tied to Rubin and Kyber system availability. (nvidia.com) Nvidia has not announced a change to the Kyber timetable. Until the company gives a more detailed schedule, the public record is a disputed delay report on one side and Nvidia’s statement that its roadmap remains intact on the other. (finance.yahoo.com) (developer.nvidia.com) (cnbc.com)