Nvidia has reportedly pushed back its next-generation AI server rack, the Kyber NVL144, by more than a year to 2028. According to analyst firm SemiAnalysis, the delay stems from manufacturing problems with the rack's circuit boards — specifically what Tom's Hardware describes as PCB midplane issues.
The Kyber rack is the physical enclosure that packs Nvidia's chips together into a single high-density AI computer. When the boards that tie those chips together don't come together cleanly in production, the whole system slips. SemiAnalysis says the holdup amounts to a delay of more than 12 months.
The fallout wasn't limited to Nvidia. According to The Decoder, Asian suppliers tied to the product saw their market value fall by up to double-digit percentages, and QZ reported supplier stocks sinking on the news.
There's a second casualty. Reports say a stopgap solution — associated with the more powerful Rubin Ultra variant — has also been axed. Tom's Hardware attributes that cancellation to customer pushback, while The Decoder frames the Rubin Ultra variant itself as canceled.
Several outlets, including TIKR, note that the stumble hands Nvidia's rivals a rare opening. Nvidia has dominated the market for AI computing hardware, and manufacturing hiccups that push a flagship product out by a year give competitors room to court customers who can't wait.
All of these accounts trace back to the same source, SemiAnalysis, so the details remain reports rather than confirmed by Nvidia.
Why it matters: AI data centers are being built on a tight timeline, and a year-long slip in Nvidia's next big rack ripples through its suppliers, its customers, and its competitors all at once.