Nvidia's Kyber NVL144 AI Rack Delayed to 2028 Over PCB Midplane Manufacturing Issues
Supply‑chain hurdles push Nvidia’s flagship AI rack more than a year behind schedule, rattling Asian component makers.
- Nvidia's Kyber NVL144 rack now slated for 2028, over a year delay.
- Problem traced to difficulty fabricating the midplane PCB.
- Asian component supplier stocks fell sharply after the news.
- Delay could reshape AI hardware market and spur new interconnect research.
Nvidia announced that its next‑generation Kyber NVL144 AI rack will not ship until 2028, a slip of more than twelve months caused by problems fabricating the system’s midplane printed‑circuit board (PCB). The delay, first reported by research firm SemiAnalysis ahead of market open, reverberates through a supply chain that includes several Asian component manufacturers whose stocks have already taken a hit.
Core developments
SemiAnalysis, cited by multiple news outlets, said the Kyber NVL144 rack – the centerpiece of Nvidia’s strategy to dominate large‑scale generative‑AI workloads – is now expected to ship in 2028, pushing the original timeline back by over a year. The firm pinpointed “difficulties in manufacturing the midplane PCB” as the decisive bottleneck.
Both Moomoo feeds and a CNBC briefing echoed the same cause, noting that the midplane – the high‑density interconnect that links GPUs, CPUs and networking ASICs – requires precision that current production lines are struggling to achieve.difficulties in manufacturing the midplane PCBMoomoo
The delay is not merely a calendar adjustment. The Kyber platform is designed to house up to 144 Nvidia H100 GPUs, delivering petaflops of AI performance for hyperscale data centers. A postponement therefore stalls the rollout of new services that cloud providers have pledged to launch on the hardware, and it also postpones revenue that Nvidia had projected from the rack’s sales.
Asian suppliers that were slated to provide the PCB and related components saw their share prices tumble after the news broke. The‑decoder reported a sharp decline in Asian component stocks, while finance.biggo.com highlighted a broader sell‑off across the regional supply chain. The market reaction underscores how tightly linked Nvidia’s product cadence is to the health of its upstream ecosystem.
Why it matters
Nvidia’s AI racks are more than just high‑end servers; they are the de‑facto standard for training the largest language models. The Kyber NVL144, in particular, was marketed as the “next‑gen” system that would enable a new wave of generative‑AI applications, from real‑time translation to autonomous‑vehicle simulation. A delay therefore has a cascade effect: cloud operators must defer capacity expansions, AI startups may see longer wait times for compute, and competitors could gain a foothold if they can deliver comparable performance on schedule.
From a supply‑chain perspective, the midplane PCB issue highlights a growing vulnerability in the semiconductor ecosystem. Manufacturing a high‑speed, high‑density interconnect that can handle terabytes per second of data traffic pushes the limits of current PCB fabrication technology. The problem is not unique to Nvidia; other firms that rely on similarly dense interconnects have reported yield challenges, suggesting that the industry may need to invest in new materials or processes to keep pace with AI‑driven demand.
The delay also feeds into broader concerns about the concentration of AI hardware production in a handful of vendors. If Nvidia’s flagship product slips, the pressure on alternative providers – such as AMD’s Instinct line or emerging Chinese AI accelerators – could increase, potentially reshaping market dynamics in the next few years.
Differing viewpoints and reactions
While most outlets presented the delay as a straightforward supply‑chain setback, some analysts offered a more nuanced take. CNBC quoted industry observers who suggested that Nvidia may be using the postponement to fine‑tune the Kyber architecture, ensuring that thermal and power delivery constraints are fully resolved before mass production.
On the other hand, qz.com emphasized the immediate financial impact on suppliers, noting that “stock prices of Asian component makers have sunk” following the announcement. The article implied that investors are pricing in not only the lost revenue from the Kyber rack but also the broader uncertainty around high‑density PCB manufacturing.
A few commentators, such as those at finance.biggo.com, warned that the delay could be a leading indicator of systemic strain in the AI hardware supply chain. They argued that if Nvidia’s most advanced rack cannot meet its production targets, other firms that depend on similar supply‑chain nodes might face comparable setbacks.
What’s next
Nvidia has not released a detailed remediation plan, but the company’s historical approach suggests it will work closely with its PCB partners to resolve the yield issues, possibly by qualifying new fabrication facilities or revising the midplane design to improve manufacturability. The firm is also likely to keep customers informed about revised delivery windows, as many cloud providers have already allocated budget for the rack’s deployment.
For investors, the immediate signal is heightened volatility in the semiconductor supply chain. Watching the stock performance of Asian PCB and substrate manufacturers will provide a barometer for how quickly the problem is being addressed. In parallel, analysts will monitor Nvidia’s quarterly earnings for any guidance adjustments related to the Kyber timeline.
In the longer term, the episode may accelerate research into alternative interconnect technologies, such as silicon‑photonic or advanced organic substrates, that could sidestep the current PCB bottleneck. If those efforts bear fruit, the next generation of AI racks could emerge with a different architectural foundation, potentially reshaping the competitive landscape.