NVIDIA's Kyber-Rack AI server platform delayed to 2028 due to advanced packaging and manufacturing bottlenecks.

Nvidia reportedly faces a delay exceeding 12 months for its next-generation Kyber AI rack architecture, potentially pushing the system's launch into 2028.

The Kyber NVL144 system is designed to connect 144 high-performance chips in a single rack to support advanced artificial intelligence training and inference workloads. The architecture was expected to arrive alongside Nvidia's Vera Rubin Ultra platform in 2027.

According to SemiAnalysis, manufacturing challenges involving the PCB midplane—a specialized circuit board connecting electronic modules within the rack—are driving the delay. The research firm also noted that Nvidia's larger NVL576 system could face delays or limited initial volumes.

The setback raises questions about whether Nvidia's annual product release schedule is running into manufacturing limits as AI systems become increasingly complex.

A proposed alternative combining two current-generation racks was reportedly canceled after cloud providers objected to the added operating complexity and costs.

Nvidia's current Rubin systems remain in full production and are expected to begin shipping this fall to major cloud partners. SemiAnalysis projects Nvidia's data-center compute revenue could run 20% above Wall Street consensus in the second half of fiscal 2027.