Nvidia has hit an unusual snag. Its next-generation Kyber NVL144 rack system — the room-sized hardware that bundles many AI chips into a single unit — has been delayed by more than 12 months, pushing its rollout to 2028, according to the research firm SemiAnalysis.
The holdup is a manufacturing problem, not a design flaw. SemiAnalysis says the trouble lies in producing the rack's "midplane PCB," the large circuit board that connects the system's components. CNBC, citing SemiAnalysis, reported the delay ahead of the market open.
Kyber was meant to house Nvidia's Rubin Ultra chips, which had been slated for 2027. As reported by outlets including CNBC and Crypto Briefing, the slip to 2028 raises a broader worry: Nvidia's aggressive plan to ship a new AI architecture every year may be colliding with the hard limits of what factories can actually build.
For a company that dominates AI hardware and rarely misses, the misstep stands out. According to odaily.news, rivals AMD and Google now see a potential window of opportunity to close the gap while Nvidia's flagship is stalled.
Why it matters: Nvidia's chips power much of the current AI boom, so a year-plus delay to its most advanced system could reshape who supplies the industry's computing muscle — and hand competitors a rare opening.