Jensen Huang, chief executive of Nvidia Corp., speaks next to the Vera Rubin Ultra Kyber Compute Tray and the Vera Rubin Ultra Kyber NVLink MidPlane during a keynote speech at the Nvidia GTC conference in San Jose, California, US, Monday, March 16, 2026.
Bloomberg | Bloomberg | Getty Images
NvidiaThe next marquee product — a Kyber rack-scale structure designed to house 2027 Rubin Ultra chips — has been delayed by more than 12 months to 2028, according to research firm SemiAnalysis, the latest in a series of reported setbacks that raise questions about the AI giant’s product roadmap.
Nvidia rejected the SemiAnalysis report saying, “Our road is right.”
Shares of the intelligence giant rose nearly 1% on Monday.
Kyber is a server cabinet that packs 144 powerful Nvidia chips into a single unit to work together as one supercomputer, providing the AI horses companies need to train and run their most advanced models.
The design puts graphics processing units in computer trays that sit vertically instead of horizontally to increase density and reduce latency, and was scheduled to debut with Vera Rubin Ultra, Nvidia’s next-generation rack-scale system, in 2027.
The delay is due to difficulties in manufacturing a key circuit board at the heart of the system, SemiAnalysis said in a post on Monday.
“The Kyber NVL144 rack architecture is delayed until 2028 as the mid-plane PCB remains a challenge from a manufacturing perspective,” the company said, referring to the special multi-layer printed circuit board that connects the electronic modules to the system.
The NVL576 – a larger system that connects eight racks via optical connections – is also likely to be delayed or limited to smaller volumes, the research firm said.
The reported delays add to growing problems in Nvidia’s product lines, underscoring concerns that Nvidia’s annual release cadence is bumping up against production limits.
The backup plan – to combine two current-generation Nvidia racks for the same power – was also scrapped after cloud customers rejected the design as unattractive and too expensive to use. “It has since been canceled due to intense pressure from CSPs [cloud service providers] and hyperscalers over their unconventional design and heavy workloads,” SemiAnalysis said.
That leaves Nvidia “without a proven solution to increase the size of the world’s Rubin Ultra,” SemiAnalysis said, predicting that would give it to competitors. Advanced Micro Devices again Googleits in-house chips are already winning business from top AI labs, a rare technology opening at the high end of the market.
The delay “should not be over-analyzed as it affects the long-term importance of Nvidia to the AI of data infrastructure construction,” said Paul Triolo, a partner in consulting DGA-Albright Stonebridge Group, noting that the company “has faced these challenges before, and has worked with vendors to overcome technical problems.”
Since power supply is likely to remain a major barrier to data center AI spending, Triolo said, “the delay in getting to the most advanced systems could just mean that new systems are ready at a time when the US can work to overcome some of the significant energy constraints that now grip the industry.”
Nvidia’s current generation Rubin systems are in full production and begin shipping this fall to eight cloud partners, including Amazon Web Services, Microsoft Azure and Google Cloud. SemiAnalysis also projects that Nvidia’s data center computing revenue will spend 20% more than Wall Street’s consensus in the second quarter of fiscal 2027.
While Huawei and other domestic manufacturers “may find time,” China’s ecosystem is increasingly diverging from — rather than chasing — the Nvidia-led model, Triolo said. “The issue is not yet clear, but how much better will China’s AI stack be in 2030,” he said.
