- SemiAnalysis reported via X that Nvidia’s Kyber NVL144 — a next-generation AI server rack system featuring 144 GPUs that CEO Jensen Huang demoed at GTC just three months ago — has been delayed by more than 12 months and pushed to 2028 due to setbacks in printed circuit board construction; Nvidia did not respond to a request for comment, and the absence of a denial sent PCB supplier stocks sharply lower across Asia.
- The selloff in AI supply-chain stocks was severe and concentrated: Japan’s Ibiden Co. (Nvidia’s largest PCB client) fell as much as 10%; Kingboard Laminates fell 18% in Hong Kong; Elite Material dropped 10% in Taiwan; Samsung Electro-Mechanics slid 11% in South Korea — steep moves for stocks that had already gained 470-600% this year, reflecting how tightly stretched valuations were relative to the single customer driving the bull thesis.
- Analysts caution that a Kyber delay doesn’t necessarily imply weaker AI capex broadly — the more likely interpretation is that Nvidia’s most ambitious next-gen architecture will take longer to deploy, which isn’t entirely surprising given its complexity; the NVL72x2 back-to-back rack architecture that Kyber was designed to supersede remains in production, meaning near-term GPU demand and revenue are not directly impaired by the setback.
- The market reaction illustrates growing investor fragility in AI supply-chain stocks: the MSCI Asia tech sector gauge had already fallen 8.5% in the prior two weeks before Monday’s additional decline; NH Investment & Securities analyst Shawn Oh noted the delay creates “a wider competitive window for alternative AI platforms” and raises uncertainty about Nvidia’s next-generation scale-out roadmap — a concern that resonates given the growing competitive pressure from AMD, custom silicon from Google and Amazon, and Chinese alternatives.
What Happened?
Research firm SemiAnalysis published a report on X stating that Nvidia’s Kyber NVL144 — a next-generation AI server rack system housing 144 GPUs that was demonstrated by CEO Jensen Huang at GTC just three months ago — has hit major manufacturing setbacks and been delayed by more than 12 months, pushing the expected deployment to 2028. The reported issue centers on printed circuit board construction for the platform. Nvidia did not respond to a request for comment, and the absence of a denial was treated as tacit confirmation by markets. Asian PCB suppliers — which had run up dramatically on AI infrastructure demand — sold off sharply, with Ibiden, Samsung Electro-Mechanics, Kingboard Laminates, and Elite Material all falling between 10% and 18% on the news.
Why It Matters?
The Kyber NVL144 delay is significant for several reasons beyond the immediate supply-chain selloff. First, it reveals the technical difficulty of scaling AI server architectures to 144-GPU rack configurations — a complexity that Nvidia’s competitors may be able to exploit. Second, the severity of the PCB stock reactions (10-18% single-day drops on stocks up 470-600% YTD) illustrates how concentrated and fragile the AI infrastructure investment thesis has become: when a single unreported research note can move billion-dollar suppliers by double digits, it suggests position sizing and leverage in the supply-chain trade may be excessive. Third, the delay creates strategic uncertainty: hyperscalers planning their next infrastructure cycle around Kyber availability now need to recalibrate procurement timelines, which could affect near-term capex cadence even if long-term demand is unchanged.
What’s Next?
The immediate catalyst to watch is Samsung Electronics’ earnings report Tuesday — the first major AI supply-chain financial data point since the Kyber news broke. If Samsung’s AI-related revenue guidance is strong, it could partially offset Monday’s sentiment damage. Nvidia’s own earnings (typically reported in late August) will be the definitive test: if management confirms or denies the Kyber delay there and provides a revised timeline, markets will reprice the entire supply chain accordingly. In the interim, the Kyber setback modestly strengthens the competitive position of AMD (whose MI400 roadmap is unaffected) and custom silicon providers like Google’s TPUs and Amazon’s Trainium, which could benefit from any deployment gaps created by the delay. For investors, the question is whether this is a 12-month speed bump or evidence of a more fundamental constraint on Nvidia’s scaling roadmap.
Source: Bloomberg
