#gStocksTokenizedStocksLive

Kyber NVL144 Ambition Delayed, Nvidia Faces Real Test in AI Rack Production Line

Nvidia was once again in the global market spotlight early this week, after the latest report from semiconductor research firm SemiAnalysis revealed that the next-generation AI rack architecture, Kyber NVL144, is experiencing significant delays. This news was first shared through a series of posts on platform X on the morning of July sixth, and within hours it directly pressured technology stocks in the Asian region.

Kyber NVL144 Pushed Back to 2028

According to the report, the launch of Kyber NVL144 is now estimated to shift from the original plan of 2027 to 2028. This delay amounts to more than twelve months from the schedule previously showcased directly by Jensen Huang on the GTC stage just three months ago. The main cause lies in manufacturing difficulties of the main circuit board called the orthogonal midplane, a component with an ultra-dense design of up to seventy-eight layers using high-grade copper and quartz fabric materials. The required precision level in production yield, impedance, and thermal management of this component is said to be a major challenge for Nvidia's current supply chain.

Kyber NVL144 itself is designed as a high-density AI cabinet capable of integrating up to one hundred forty-four top-tier GPUs in a single unit. GPUs in this architecture are placed vertically, not horizontally like previous generations, to increase computational density while reducing inter-chip communication latency. This system becomes a crucial foundation for the Vera Rubin Ultra platform, which is touted to support training and inference of giant-scale AI models.

Backup Plan Also Canceled

In addition to the main delay, SemiAnalysis also revealed that the NVL576 system, a larger configuration connecting eight racks via optical interconnects between NVSwitches, will likely also be delayed or only produced in limited volumes due to challenges in optical interconnect or CPO technology. Not stopping there, Nvidia's backup plan, the NVL72x2 design, which essentially connects two current-generation cabinets to approach Kyber's power, has also been completely canceled. This cancellation occurred after strong protests from cloud service providers and hyperscalers, who deemed the design awkward and too expensive in terms of long-term operations and maintenance.

This condition directly means that Nvidia currently does not have a truly proven solution to scale up the Vera Rubin Ultra computing world, a rather heavy admission for a company known for its very aggressive annual product launch cadence.

Impact on Ecosystem and Competitor Opportunities

This delay immediately opens a gap for Nvidia's competitors. Analysts say AMD, with its product line like the MI500X, and Google's internal chip solutions, have potential to gain more room to close the gap in the high-performance AI computing segment. Market reaction was swift, with technology stocks in Asia related to Nvidia's supply chain weakening once the report spread, while several analysts began highlighting other companies like Marvell Technology, Lattice Semiconductor, and Tower Semiconductor as potentially benefiting from this additional time window.

Nevertheless, it is important to note that the current-generation Rubin system continues with full production and is scheduled to start shipping this fall to eight major cloud partners, including Amazon Web Services, Microsoft Azure, and Google Cloud. Several analysts remain optimistic, estimating that Nvidia's data center segment revenue in the second half of fiscal year 2027 could still potentially surpass Wall Street consensus by up to twenty percent, although the shadow of Kyber's delay continues to weigh on short-term sentiment.

Manufacturing Capability Limits Amid Unabated Demand

The Kyber NVL144 case serves as a real mirror that rapid innovation in the AI industry is not solely about chip design, but also about the ability to realize it on a massive industrial production scale. Demand for AI infrastructure remains very high to this day, but the supply chain is encountering real technical hurdles at the advanced circuit board level, an area that has often been overlooked compared to chip performance itself.

For technology companies relying on large-scale AI infrastructure, such delays could alter project timelines and compute capacity acquisition strategies over the next few years. Investors and industry players will now continue to monitor how Nvidia navigates these manufacturing obstacles while striving to maintain its dominant position on the global AI computing stage.