Nvidia faces delay for next-generation Kyber AI platform

Traders Union
All News
Financial News
Nvidia AI rack system delayed

Highlights

Nvidia’s Kyber NVL144 system has reportedly been delayed to 2028.
Manufacturing challenges are centered on the PCB midplane.
Nvidia shares were shown at $194.83, down 1.39%.

Kyber delay exposes manufacturing hurdles

The Kyber NVL144 rack architecture has been delayed to 2028 because of manufacturing challenges tied to its PCB midplane, according to research firm SemiAnalysis. The system had been presented only months earlier as a major step in Nvidia’s roadmap for larger AI computing clusters.

Loading...

The tweet was deleted by the author.

But we saved everything 🙂.

The delay matters because rack-scale systems are becoming central to AI infrastructure. Rather than selling only chips, Nvidia is increasingly selling full systems that connect GPUs, memory, networking, and cooling into large computing units for cloud providers and AI labs.

The reported setback follows another change to the Rubin Ultra roadmap. Nvidia has reportedly canceled the original four-compute-die Rubin Ultra design and shifted to a smaller two-die version, a move that could cut real-world performance compared with the earlier plan. Tom’s Hardware reported that the four-die version was scrapped because of manufacturing execution concerns.

Alternatives also face pressure

SemiAnalysis also said Nvidia’s NVL72x2 back-to-back rack design, developed as an alternative to Kyber, has been canceled after pushback from cloud service providers and hyperscalers. The design was intended to expand the copper-based NVLink scale-up domain by placing two Oberon racks back to back, but customers reportedly viewed it as operationally difficult.

The result is a narrower path for Rubin Ultra. If Kyber and related systems are delayed, Nvidia may have fewer proven options to expand scale-up performance before later architectures arrive. The report said CPO NVSwitch, which could help solve some of the interconnect constraints, is not expected until the Feynman generation.

That creates an opening for competitors. AMD, Google, and other AI-chip developers are trying to win more data-center spending as cloud companies look for alternatives to Nvidia’s dominant systems. Nvidia remains the market leader, but delays in rack-scale infrastructure could make the next phase of competition more technical and less predictable.

Shares reflect the pressure

Nvidia shares fell 1.39% to $194.83 as chip stocks remained sensitive to any signs that demand for AI hardware could face manufacturing or design constraints.

For Nvidia, the issue is not only whether AI demand remains strong. It is whether the company can keep turning that demand into manufacturable systems at the scale required by hyperscalers. The reported Kyber delay does not erase Nvidia’s lead, but it shows that the AI buildout now depends on difficult engineering across chips, memory, boards, networking, and full racks.

We also reported Nvidia aims to "reinvent the PC" with new processors.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.