NVIDIA Corp (NVDA) has swiftly refuted a report claiming a delay in its next-generation "Kyber" artificial intelligence (AI) rack until 2028, with a company spokesperson stating, "Our roadmap is intact." The initial report, published on Sunday by SemiAnalysis, alleged that the Kyber rack-scale architecture, designed to house its 2027 Rubin Ultra chip, would face a delay of over 12 months, pushing its debut to 2028.
SemiAnalysis elaborated in a post on X, stating, "Just three months after Jensen Huang showcased Kyber NVL144 at GTC, it has hit a major setback and been delayed by over 12 months to 2028." The report further claimed that due to strong objections from hyperscale cloud customers regarding practical operations, NVIDIA has canceled the alternative NVL72x2 back-to-back rack design. Additionally, citing current challenges with CPO (co-packaged optics), a larger system called NVL576, which would connect eight Oberon racks, "may also be delayed or limited to small shipments."
The Kyber design is a server cabinet that packages 144 of NVIDIA's most powerful chips into a single unit, enabling them to work together like a giant computer to provide the computational power required for AI companies to train and run their most advanced models. The design features GPUs mounted in vertically-oriented compute trays to increase density and reduce latency, and was slated for launch in 2027 alongside NVIDIA's next-generation rack-scale system, Vera Rubin Ultra.
Industry analysts have downplayed the significance of the reported delay. An integrator within NVIDIA's Taiwan server supply chain reportedly indicated that any production delay for the Kyber rack would have minimal impact, affecting neither NVIDIA's chip upgrade cycle, its market position, nor the existing supply chain.
Jordan Klein, an analyst at Mizuho Securities' trading desk, commented on Monday, suggesting investors have seen this pattern before. He characterized the recurring news about NVIDIA product delays due to manufacturing issues as "attention-grabbing noise." Paul Triolo, a partner at consulting firm Albright Stonebridge Group, echoed this sentiment, stating that the delay rumors "should not be over-interpreted as affecting NVIDIA's long-term core importance in AI data center infrastructure," noting the company has previously faced and overcome similar challenges.
However, the report also notes that as power supply remains a key constraint for AI data center spending in the U.S., a delay in delivering advanced systems could provide more time to address certain core power bottlenecks.