Nvidia’s Kyber rack system delayed to 2028 over manufacturing snags
Jensen Huang, CEO of Nvidia Corp., speaks next to the Vera Rubin Ultra Kyber Compute Tray and Vera Rubin Ultra Kyber NVLink MidPlane during his keynote at the Nvidia GTC conference in San Jose, California, U.S., on Monday, March 16, 2026.
Bloomberg | Bloomberg | Getty Images
NVIDIA’s next major product, the Kyber rack-scale architecture designed to house 2027 Rubin Ultra chips, has been delayed by more than 12 months to 2028, according to research firm SemiAnalysis; This is the latest in a series of negative reports that raise questions about the AI giant’s product roadmap.
Kyber is a server cabinet that packs 144 of Nvidia’s most powerful chips into a single unit so they can work together like a giant computer, providing the horsepower AI companies need to train and run their most advanced models.
The design mounts graphics processing units on compute trays placed vertically rather than horizontally to increase density and reduce latency. Nvidia’s next-generation rack-scale system Vera Rubin Ultra launches in 2027.
SemiAnalytics said in a post on Monday that the glitch was caused by difficulties manufacturing the key circuit board at the heart of the system.
“The Kyber NVL144 rack architecture has been delayed to 2028 as the PCB midplane remains challenging for manufacturability,” the firm said, referring to the custom, multilayer printed circuit board that interconnects electronic modules within a system.
The NVL576, a larger system that connects eight racks via optical links, will also likely be delayed or limited to small volumes, the research firm said.
Nvidia did not respond to CNBC’s request for comment.
The reported delay adds to growing pressures on Nvidia’s product lines and underscores concerns that Nvidia’s excessive annual release pace is colliding with production limits.
A backup plan that bolted two of Nvidia’s current-gen racks together for similar power was also scrapped after cloud customers rejected the design as awkward and costly to run. “Canceled due to intense reaction from CSPs” [cloud service providers] and because of the peculiar design and heavy operational burden of hyperscalers,” SemiAnalytics said.
SemiAnalysis predicted that this could give Nvidia an advantage over competitors, saying that “Rubin Ultra has no proven solution left to scale up globally.” Advanced Micro Devices And GoogleIts in-house chips are already winning business from top AI labs, a rare technical opening at the top end of the market.
Nvidia’s current-generation Rubin systems are in full production and will begin shipping this fall to eight cloud partners, including Amazon Web Services, Microsoft Azure and Google Cloud. SemiAnalytics also predicts Nvidia’s data center computing revenue will be 20% above consensus on Wall Street in the second half of fiscal 2027.
Nvidia shares fluctuated in premarket trading, last falling less than 0.1% to $194.79.
