NewsNvidiaAI InfrastructureChip Manufacturing

Nvidia's Kyber Rack Delayed to 2028 – Rivals See Opening

Nvidia's next-generation AI infrastructure hits manufacturing snags. The delay opens a rare window for AMD and Google in the high-end market.

Kyber NVL144 delayed 12+ months to 2028

Nvidia's Kyber Rack Delayed to 2028 – Rivals See Opening

Nvidia's flagship infrastructure project is running into trouble: the Kyber NVL144 rack system, originally slated to launch in 2027 alongside Rubin Ultra chips, is now delayed by more than a year to 2028. Manufacturing problems with a critical circuit board are to blame, according to research firm SemiAnalysis, which reported the setback on Monday. Nvidia did not comment to CNBC.

Quick Facts

  • Kyber NVL144 packs 144 of Nvidia's most powerful chips into a single rack cabinet, arranged in vertical compute trays to boost density and reduce latency
  • The PCB midplane (a specialized multi-layer printed circuit board) is causing manufacturing challenges, SemiAnalysis says it "remains challenging from a manufacturability standpoint"
  • The larger NVL576 system (eight racks linked via optical connections) is also likely delayed or limited to small volumes
  • A backup plan – linking two current-generation Rubin racks together – was cancelled after cloud service providers rejected it as awkward and operationally burdensome

The Timing Crunch

The delay hits Nvidia at a vulnerable moment. The company maintains an aggressive annual release cadence, but manufacturing capacity is clearly hitting limits. SemiAnalysis warns that Nvidia now has "no proven solution to expand the scale-up world size for Rubin Ultra" – more than a technical hiccup, it's a strategic gap.

For context: Nvidia's current Rubin systems are in full production and begin shipping this fall to eight cloud partners, including Amazon Web Services, Microsoft Azure, and Google Cloud.

Who Wins?

The delay opens doors for competitors. AMD and Google (with its in-house chips) could gain traction in the high-end segment – a rare opportunity in Nvidia-dominated territory. For these vendors, it's a chance to reach top AI labs that would otherwise wait for Nvidia.

What This Means for German Enterprises

German AI infrastructure operators and cloud providers should watch this closely. Anyone counting on Kyber systems will need to adjust plans – or explore whether AMD or Google solutions offer viable alternatives for 2027/2028. The delay also signals that even Nvidia cannot scale indefinitely. That could ease price pressure and create negotiating room. For German data centers and KI service providers, it's worth looking beyond Nvidia.

Sources

Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.

Share
← All articles

All analyses are based on i6eal's own measurements or on clearly labelled sources. Figures are snapshots and may change; corrections are disclosed transparently.