Nvidia’s Next-Generation AI Rack Delays Open Door for AMD and Google

Nvidia uses high-density AI racks to link large numbers of GPUs as it targets the scale-up computing market in data centers. Its next-generation Kyber NVL144 had been viewed as a key product for the Rubin Ultra platform. If its advanced packaging and PCB designs cannot be completed on schedule, the setback could hinder Nvidia’s ability to maintain its performance lead and disrupt customers’ deployment timelines.

Semiconductor research firm SemiAnalysis said PCB and advanced-packaging bottlenecks could delay Kyber NVL144 mass production by more than a year, until 2028, while Rubin Ultra’s specifications could also be scaled back. The resulting gap in Nvidia’s product lineup could give AMD’s MI500X and Google’s TPU an opportunity to compete for major AI infrastructure orders. The reports did not disclose any specific financial amounts.