On the morning of July 6, semiconductor industry research firm SemiAnalysis posted six consecutive tweets on X (formerly Twitter), revealing significant delays and multiple cancellations related to NVIDIA’s Kyber NVL144 rack architecture. The news drew 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 suffered a major setback, with delays exceeding 12 months, pushing its launch to 2028.'
The Midplane PCB: The board that’s holding up Kyber
According to SemiAnalysis, the immediate cause of the Kyber NVL144 delay lies in a critical hardware component—the midplane PCB, which NVIDIA officially refers to as the 'orthogonal backplane.'
The firm added: 'The Kyber NVL144 rack architecture has been delayed until 2028 due to significant manufacturing challenges still faced by the midplane PCB. The NVL576, which connects 8x Oberon racks via CPO between NVSwitches, is also likely to be delayed or limited to small-batch production due to current CPO challenges.'
The gray board Jensen Huang displayed at this year’s GTC conference in March is precisely the orthogonal backplane for the Rubin Ultra (Kyber architecture) cabinet. Its function is to enable 90-degree vertical interconnects between compute trays and switch trays—compute trays are inserted vertically and connect directly, board-to-board, through this midplane to 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 a hybrid material composed of M9-grade copper-clad laminate, quartz fabric (Q-fabric), and PTFE. It features 78 layers—formed by laminating three 26-layer boards together—with trace widths and spacings of ≤25μm to meet ultra-high-speed signal integrity requirements at SerDes rates exceeding 448G.
Why is this board indispensable? Technical analysis indicates that the Rubin Ultra NVL144 rack must interconnect 144 GPUs within a single domain. Using conventional copper cabling would require over 20,000 cables, increasing weight by more than 30% and causing severe signal attenuation. The orthogonal backplane represents one of the few viable solutions under current technological constraints.
Alternative NVL72x2 design also canceled
Faced with Kyber’s manufacturing difficulties, NVIDIA had previously attempted to develop an interim solution—the NVL72x2 back-to-back rack architecture.
According to SemiAnalysis, the design concept of this solution involved placing two Oberon racks back-to-back and scaling the domain using pure copper NVLink interconnects to circumvent manufacturing challenges associated with the Kyber midplane.
However, this approach ultimately failed to materialize. SemiAnalysis stated that the NVL72x2 was canceled 'due to strong objections from cloud service providers and hyperscale data center operators over its unconventional design and heavy operational burden.'
With both paths blocked, NVIDIA faces a temporary gap in scaling options for Rubin Ultra.
NVL576 is also under pressure, as CPO-related challenges cannot be overlooked.
It is not only the Kyber NVL144 that has been delayed. SemiAnalysis also noted that the NVL576—a larger-scale system connecting eight Oberon racks via CPO (Co-Packaged Optics)—'may also face delays or be limited to low-volume shipments given the current challenges facing CPO.'
CPO is an optical interconnect technology NVIDIA is introducing for the first time at scale in the Rubin Ultra generation. According to a SemiAnalysis research report published in March 2026, the NVL576 design calls for copper-based intra-rack scaling while using CPO to connect NVSwitches between racks, forming a two-tier fully connected network.
However, the production readiness of CPO itself remains uncertain. SemiAnalysis explicitly stated in its report that CPO-enabled NVSwitches will not be fully ready until the Feynman generation.
Rubin Ultra itself has also been scaled back: the four-chip version has been canceled.
Alongside these delay announcements, a significant product-level change was also disclosed.
SemiAnalysis reported that the four-compute-die version of Rubin Ultra has been canceled, 'leaving only the smaller two-compute-die version of Rubin Ultra, which delivers roughly half the performance of the original four-die configuration.'
This means that even if Kyber racks are ultimately delivered on schedule, the per-rack compute ceiling has been significantly reduced.
In response, SemiAnalysis stated that NVIDIA will offset this shortfall by 'significantly increasing sales of Oberon Rubin racks and Oberon Rubin Ultra racks.'
Competitive Window: AMD and Google May Benefit
The gap in scalability directly affects NVIDIA's competitive position in large-scale training scenarios.
SemiAnalysis noted: 'NVIDIA currently lacks a proven solution to scale the Oberon Rubin Ultra’s scalability domain, creating an opening for competitors such as the AMD MI500X or Google’s TPUv8i Broadfly to surpass Rubin Ultra in scalability capabilities.'
According to NVIDIA’s current roadmap, CPO NVSwitch will not be introduced until the next-generation Feynman platform. Until then, the scalability ceiling of Rubin Ultra remains constrained.
SemiAnalysis concluded its tweet by noting that the aforementioned delays and cancellations will impact the memory, PCB, and ODM supply chains.
The manufacturing challenges associated with the Kyber midplane directly reflect technical bottlenecks faced by high-end PCB suppliers. The midplane requires a 78-layer ultra-high-density PCB, M9-grade copper-clad laminate, and PTFE hybrid materials—representing the current limits of PCB manufacturing technology.
Want to select stocks or get a stock diagnosis? Curious about the opportunities and risks in your portfolio? For all your investment questions,Just ask Futubull AI!
Edited by Joryn