Nvidia's next AI server system, Kyber NVL144, has reportedly moved further out on the calendar, and the shift is already being felt across parts of the Asian technology supply chain.
According to analyst firm SemiAnalysis, the system has been pushed back by more than twelve months to 2028 because of manufacturing problems tied to circuit boards. The delay affects one of Nvidia's most ambitious AI infrastructure designs and arrives after a long run of investor enthusiasm around AI hardware.
Why Kyber NVL144 is delayed
The reported problem sits inside the hardware that holds the system together. SemiAnalysis points to the PCB midplane, a central circuit board that connects the individual components inside the rack.
That part has reportedly been extremely difficult to manufacture without defects. In a large AI server rack, the midplane is not a minor detail. It is the physical connection layer that helps many parts work as one system, so production quality matters directly to whether the design can scale.
The timing also matters. Nvidia CEO Jensen Huang had shown off Kyber NVL144 just three months earlier at the company's GTC conference. Now, according to the report, the system is not expected until 2028.
For customers planning data center expansion, a delay of more than a year can change procurement decisions. For suppliers, it can shift revenue expectations. For competitors, it can create space to make their own platforms look more practical or better timed.
Asian suppliers react to the report
The SemiAnalysis report triggered sharp stock moves among Nvidia-linked suppliers across Japan, Taiwan, South Korea, and Hong Kong. The reaction was especially strong because many AI hardware suppliers had already seen major gains.
Japanese PCB maker Ibiden, which counts Nvidia as its largest customer, dropped as much as ten percent. Kingboard Laminates fell 18 percent in Hong Kong. Elite Material lost ten percent in Taiwan, while Samsung Electro-Mechanics slid eleven percent in South Korea.
The selling followed large earlier run-ups. Samsung Electro-Mechanics had gained more than 600 percent this year, and Kingboard Laminates had gained more than 470 percent.
That context is important. The reported Kyber NVL144 delay did not land in a quiet market. It hit investors who had already priced in strong expectations for AI infrastructure demand and Nvidia's future system roadmap.
Other Nvidia designs were reportedly canceled
SemiAnalysis also listed additional setbacks beyond Kyber NVL144. One planned alternative design, NVL72x2, has reportedly been scrapped entirely.
That design would have placed two Oberon racks back to back. According to the report, cloud providers and large data center operators pushed back against the unusual form factor and the high operational overhead.
The report also says the more powerful version of Nvidia's upcoming Rubin Ultra chip has been canceled. That version would have used four compute dies. Only the smaller two-die version remains, delivering roughly half the real-world compute performance.
These changes point to a broader issue: building very large AI systems is not just about faster chips. The rack design, the physical layout, the interconnects, and the operating burden all matter. If a system is difficult to build or difficult to run, large customers may resist it even if the performance target is attractive.
The interconnect gap could help rivals
A key technology called CPO-NVSwitch is also not expected until the generation after next, called Feynman. CPO-NVSwitch is meant to link many chips into a single large system.
Without that technology in the Rubin Ultra generation, Nvidia reportedly lacks a proven way to scale Rubin Ultra into very large systems for now. SemiAnalysis says this gap could give competitors room to move in, including AMD's MI500X and Google's TPUv8i Broadfly.
Nvidia's reported answer is to sell more Oberon-Rubin racks in the existing form factor. That may help fill part of the gap, but it does not erase the larger timing issue around Kyber NVL144 and the delayed interconnect path.
The competitive takeaway is straightforward. If Nvidia's most advanced rack-scale systems take longer to arrive, alternative AI platforms have more time to compete for attention from cloud providers and large data center operators.
What the delay does and does not mean
The stock reaction was severe, but analysts cited in the source article do not frame it as proof that AI spending is weakening. Gary Tan of Allspring Global Investments said, according to Bloomberg, that a Kyber delay does not necessarily mean overall AI spending will shrink.
Instead, the delay shows that Nvidia's most ambitious system is taking longer than expected. The current stock weakness is mostly being driven by profit-taking, according to that view.
Shawn Oh of NH Investment & Securities pointed to growing uncertainty around Nvidia's expansion plans. That uncertainty gives alternative AI platforms more room to compete.
For now, the key issue is execution. Nvidia remains central to the AI hardware market described in the source, but the Kyber NVL144 delay, canceled designs, and later arrival of CPO-NVSwitch show how hard it is to turn ambitious rack-scale AI systems into manufacturable, customer-ready products.