Nvidia's Kyber NVL144 Delayed More Than a Year — What It Means for AI Hardware

Nvidia's Kyber NVL144 Delayed More Than a Year — What It Means for AI Hardware

Nvidia's next-generation AI server rack system, Kyber NVL144, has been delayed by more than 12 months, pushing its release to 2028. According to a report from SemiAnalysis, the delay stems from extreme manufacturing difficulties with the system's PCB midplane — the central circuit board that connects every node and GPU together.

The news comes just three months after Nvidia CEO Jensen Huang showcased Kyber NVL144 at GTC 2026.

The Core Problem: PCB Midplane Manufacturing

The Kyber NVL144's PCB midplane has proven to be extraordinarily difficult to manufacture without defects. This component serves as the backbone of the entire system, interconnecting all GPUs and nodes at unprecedented speeds.

Market Turmoil: Asian Suppliers Crash

The news sent shockwaves through the Asian supply chain, with major Nvidia suppliers seeing sharp stock declines: Ibiden -10%, Kingboard Laminates -18%, Elite Material -10%, Samsung Electro-Mechanics -11%.

More Than Just a Delay: Cancellations

The NVL72x2 alternative design has been entirely scrapped. The Rubin Ultra 4-die chip has been canceled. CPO-NVSwitch will not arrive until the Feynman generation.

Nvidia's Contingency Plan

Nvidia plans to sell more Oberon-Rubin racks in the existing form factor. Analysts say the Kyber delay does not mean AI spending is shrinking.

Competitors Poised to Gain Ground

AMD's MI500X, Google's TPUv8i Broadfly, and other AI accelerators now have more time to capture market share.

Analysis

The Kyber NVL144 delay marks a pivotal moment for the AI hardware industry. While Nvidia remains dominant, this signals that even the market leader faces significant engineering challenges.