For years, Nvidia’s biggest competitive advantage has been designing the world’s most powerful AI chips. But as artificial intelligence infrastructure becomes larger and more sophisticated, manufacturing the systems around those chips is emerging as an equally formidable challenge.
According to CNBC, citing semiconductor research firm SemiAnalysis, Nvidia’s next-generation Kyber NVL144 rack system has reportedly been delayed from its planned 2027 launch to 2028 because of manufacturing difficulties involving the printed circuit board (PCB) midplane. Nvidia has not publicly confirmed the reported delay.
If the report proves accurate, it would illustrate a broader shift across the semiconductor industry. The bottleneck is increasingly moving beyond chip design to the engineering required to assemble hundreds of processors into a single high-performance computing system.
Rack Computing Gets Harder
Kyber NVL144 represents Nvidia’s strategy of selling complete AI infrastructure rather than standalone graphics processors. The rack-scale system is designed to integrate 144 processors into a tightly connected platform capable of training and running increasingly large AI models.
While advanced chips remain central to performance, systems of this scale depend on sophisticated networking, power delivery, cooling and complex PCB designs that connect every component with minimal latency.
According to CNBC’s report, SemiAnalysis identified the PCB midplane as the primary engineering hurdle delaying production. Manufacturing these large, high-density boards while maintaining reliability at commercial scale is considerably more complex than producing individual semiconductor packages.
The reported setback highlights how AI computing is entering an era where success depends not only on transistor technology but also on the ability to manufacture increasingly intricate systems.
Annual Roadmap Faces Pressure
The reported delay also comes against the backdrop of Nvidia’s accelerated product strategy.
In 2024, Chief Executive Jensen Huang announced that the company would transition from launching major AI computing platforms every two years to introducing new architectures annually. Reuters reported that the move was intended to match the rapid pace of AI development and customer demand while keeping Nvidia ahead of competitors.
A faster release cycle, however, leaves less room for unexpected manufacturing challenges. Even if chip development proceeds on schedule, delivering complete rack-scale platforms requires coordination across semiconductor fabrication, advanced packaging, networking components, thermal management and large-scale system assembly.
The reported Kyber delay, if confirmed, suggests that these engineering challenges could become increasingly important as AI infrastructure grows more ambitious.
Customers Demand Complete Platforms
Cloud providers are no longer buying individual accelerators alone. They are investing in integrated computing platforms that can support thousands of AI processors working together.
That evolution raises the importance of every component inside a rack. A delay in one critical subsystem can affect deployment timelines for entire AI clusters.
CNBC also reported that, according to SemiAnalysis, Nvidia abandoned an alternative architecture known as NVL72x2 after customer feedback indicated the design was less attractive operationally. Nvidia has not commented publicly on that report.
Although such roadmap adjustments are common in semiconductor development, they demonstrate the increasing influence of hyperscale cloud providers, whose technical requirements now shape the industry’s product strategies.
Industry Enters New Phase
The reported Kyber delay is significant not because it signals weakness in Nvidia’s chip design capabilities, but because it reflects the growing complexity of AI infrastructure itself.
As AI models become larger and data centres expand, competitive advantage will increasingly depend on integrating processors, networking, cooling and power systems into reliable, manufacturable platforms.
Nvidia remains the industry’s dominant AI hardware supplier, and the company has not confirmed the reported roadmap changes. However, the CNBC report serves as a reminder that leadership in artificial intelligence is no longer determined solely by designing faster chips. It also depends on successfully manufacturing some of the most complex computing systems ever built.
If AI’s next chapter is defined by rack-scale computing rather than individual processors, manufacturing execution may become just as important as semiconductor innovation.
