英伟达罕见"掉链子"：主力产品延期一年，AMD和谷歌迎来窗口期

Original Title: "SemiAnalysis Reveals Pre-Market Intel: Nvidia's Kyber NVL144 Rack Delayed Over 12 Months Due to 'PCB Midplane Manufacturing Difficulties'"

Original Author: Long Yue

Original Source: Wall Street News

On the morning of July 6, semiconductor industry research firm SemiAnalysis published a series of six posts on X (formerly Twitter), revealing significant delays and multiple cancellations concerning Nvidia's Kyber NVL144 rack architecture. The news captured market attention in pre-market trading.

SemiAnalysis stated bluntly: "Major Delay: Just three months after Jensen Huang showcased the Kyber NVL144 at GTC, the product has encountered a significant setback, with a delay of over 12 months, pushing it to 2028."

The PCB Midplane: The Board Holding Back Kyber

SemiAnalysis attributes the direct cause of the Kyber NVL144 delay to a critical piece of hardware—the PCB midplane, also referred to by Nvidia as the "Orthogonal Backplane."

The firm stated: "The Kyber NVL144 rack architecture has been delayed to 2028 because the PCB midplane still faces major challenges in its manufacturing process. The NVL576, which uses CPO to connect 8x Oberon racks via NVSwitches, is also highly likely to be delayed or limited to small-volume production due to current CPO challenges."

The gray board Jensen Huang showcased at the GTC conference in March is precisely the orthogonal backplane for the Rubin Ultra (Kyber architecture) cabinet. Its function is to enable a 90° vertical interconnection between the compute trays and the switch trays—compute trays are inserted vertically, connecting board-to-board through this midplane with the rear switch trays, completely eliminating the traditional cable jungle.

Manufacturing this board is extremely challenging. According to the aforementioned technical analysis, the backplane uses M9-grade copper-clad laminate + quartz fabric (Q-fabric) + PTFE hybrid materials, with a layer count of 78 (laminated from three 26-layer boards), a line width/spacing of ≤25μm, meeting the ultra-high-speed signal integrity requirements at 448G+ SerDes rates.

Why is this board absolutely necessary? According to the technical analysis, the Rubin Ultra NVL144 rack needs to connect 144 GPUs within a single domain. Using a copper cable solution would require over 20,000 cables, increasing weight by more than 30% and causing severe signal attenuation. The orthogonal backplane is one of the few feasible solutions under current technical conditions.

Alternative NVL72x2 Solution Also Cancelled

Facing the manufacturing difficulties of Kyber, Nvidia attempted to develop a transitional solution—the NVL72x2 back-to-back rack architecture.

According to SemiAnalysis, the design concept was to place two Oberon racks back-to-back, using pure copper NVLink to expand the scale domain, thereby bypassing the manufacturing challenges of the Kyber midplane.

However, this plan ultimately did not materialize. SemiAnalysis states that the NVL72x2 "was cancelled due to strong opposition from cloud service providers and hyperscale data center operators concerning its unconventional design and heavy maintenance burden."

With both paths blocked, Nvidia now faces a temporary gap in scaling the Rubin Ultra's domain.

NVL576 Also Under Pressure; CPO Challenges Cannot Be Ignored

The delays are not limited to Kyber NVL144. SemiAnalysis also pointed out that the NVL576—a larger system connecting 8 Oberon racks via CPO (Co-Packaged Optics)—"given the current challenges with CPO, may also be delayed or limited to small-volume shipments."

CPO is an optical interconnect technology Nvidia introduced for the first time in the scale-out network during the Rubin Ultra phase. According to a research report previously published by SemiAnalysis in March 2026, the design for NVL576 is to maintain copper cable expansion within the rack while connecting NVSwitches between racks via CPO, forming a two-tier fully interconnected network.

However, the production maturity of CPO itself remains a variable. In its research report, SemiAnalysis explicitly stated that the CPO NVSwitch would not be officially ready until the Feynman generation.

Rubin Ultra Itself Downsized: 4-Chip Version Cancelled

Concurrent with the delay news, another significant product-level change was disclosed.

SemiAnalysis states that the 4-compute-chip version of Rubin Ultra has been cancelled, "leaving only the smaller 2-compute-chip version of Rubin Ultra, whose actual performance is roughly half of the 4-chip version."

This means that even if the Kyber rack eventually delivers on schedule, the compute ceiling per rack has been substantially lowered.

Regarding this, SemiAnalysis suggests that Nvidia will compensate for this gap by "significantly increasing sales of Oberon Rubin racks and Oberon Rubin Ultra racks."

Competitive Window: AMD and Google Could Benefit

The gap in the scale-out domain directly impacts Nvidia's competitive position in large-scale training scenarios.

SemiAnalysis notes: "Nvidia currently has no proven solution to scale the Rubin Ultra's scale-out domain, leaving room for competitors like the AMD MI500X or TPUv8i Broadfly to surpass Rubin Ultra in scale-out capability."

According to Nvidia's existing roadmap, the CPO NVSwitch will not appear until the next-generation Feynman platform. Until then, the scale-out ceiling for Rubin Ultra remains constrained.

At the end of its thread, SemiAnalysis indicated that these delays and cancellations have repercussions for the memory, PCB, and ODM supply chains.

The manufacturing difficulties of the Kyber midplane directly point to the technological bottlenecks of high-end PCB suppliers. The 78-layer ultra-high-density PCB, M9-grade copper-clad laminate, and PTFE hybrid material required for this midplane represent the current limits of PCB manufacturing technology.