Nvidia's Kyber NVL144 AI server has been delayed to 2028 due to circuit board manufacturing problems, according to analyst SemiAnalysis, triggering sharp sell-offs in Asian PCB suppliers. The company has also canceled two alternative system designs and a higher-capacity chip version, creating potential openings for competitors like AMD and Google. The delay highlights engineering challenges in scaling AI infrastructure while leaving near-term AI spending plans uncertain.
Summaries like this, in your inbox every morning.
Sign up free →What happened
Analyst firm SemiAnalysis reported that Nvidia's next-generation AI server system Kyber NVL144 has been pushed back more than twelve months to 2028 due to manufacturing defects in the PCB midplane, a central circuit board. The report also revealed two other cancellations: the NVL72x2 alternative design and the four-compute-die version of the Rubin Ultra chip. Shares in Japanese, Taiwanese, South Korean, and Hong Kong PCB makers fell sharply—Ibiden dropped as much as ten percent, Kingboard Laminates fell 18 percent, Elite Material lost ten percent, and Samsung Electro-Mechanics slid eleven percent.
Why it matters
Nvidia dominates AI chip supply; delays to its roadmap ripple through supplier earnings and investor sentiment. The cancellation of high-capacity designs means Nvidia currently lacks a proven way to scale the Rubin Ultra chip to very large systems, creating an opening for competitors like AMD's MI500X or Google's TPUv8i Broadfly. The stock weakness reflects profit-taking and growing uncertainty about Nvidia's expansion plans, which may give alternative AI platforms room to gain ground.
What to watch
A key interconnect technology called CPO-NVSwitch, which links multiple chips into larger systems, will not arrive until the Feynman generation (after Rubin). Nvidia plans to compensate by selling more Oberon-Rubin racks in the existing form factor. Investors are monitoring whether delays to Nvidia's product roadmap signal broader slowdowns in AI infrastructure spending or simply a shift in timing.
The Kyber delay marks a rare setback in Nvidia's multi-year dominance of AI chip supply. Jensen Huang had showcased the system at GTC just three months before SemiAnalysis's report, suggesting the manufacturing problems emerged or were disclosed relatively recently. The cancellation of both the alternative NVL72x2 form factor and the four-die Rubin Ultra version narrows Nvidia's product options: cloud providers and data center operators had rejected the unusual form factor and high operational overhead of the NVL72x2, pushing Nvidia back toward its existing Oberon-Rubin rack design.
The absence of a proven interconnect (CPO-NVSwitch arrives only in the Feynman generation after Rubin) leaves Nvidia temporarily unable to scale Rubin Ultra to very large systems, a gap that could allow AMD and Google to move upmarket. However, as Gary Tan of Allspring Global Investments noted, the delay does not necessarily reduce overall AI spending—it simply extends the timeline. The stock weakness has been attributed largely to profit-taking and growing uncertainty around expansion plans rather than a fundamental collapse in AI infrastructure demand. Still, the uncertainty creates competitive openings that alternative platforms like AMD's MI500X may exploit during the gap period.
No comments yet. Be the first to share your thoughts!
Log in to join the discussion
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
