Rumors Rattle AI Computing Chain! NVIDIA Denies Kyber Project Delay
N.R. Finch
Nvidia on July 6 denied that its next-gen rack-scale AI server Kyber NVL144 faces delays, but Asia-Pacific supply-chain stocks had already dropped 10–18% in a single session, exposing how fragile high-flying AI valuations are to any delivery-risk signal.
How did this sell-off start?
Research firm SemiAnalysis reported that Kyber NVL144 hit a manufacturing bottleneck in advanced PCB (printed circuit board — the physical backbone that connects chips) production, potentially delaying the project by over 12 months to around 2028.
This means → Nvidia's key milestone for AI scale-out — linking more GPUs into larger clusters — could slip, directly undermining the market's timeline for "when next-gen compute actually ships."
Nvidia responded via an official email: "Our road map is intact." It disclosed no specific R&D milestones or mass-production schedules.
How hard did Asia-Pacific suppliers get hit?
Japan's Ibiden, a core Nvidia PCB supplier, fell as much as nearly 10% intraday. Hong Kong-listed Kinwong dropped as much as 18%.
Taiwan's EMC (聯茂電子) closed down 10%. South Korea's Samsung Electro-Mechanics fell over 10% intraday.
In plain terms = the closer a supplier sits to Nvidia and the more "high-end" its component, the harder it was sold — the market priced delivery risk straight into share prices.
Why did one unverified rumor cause this much damage?
The MSCI Semiconductors & Tech index had already dropped nearly 10% over two weeks, with crowded AI trades unwinding fast. The sector was already fragile at elevated levels.
This means → the rumor itself was not uniquely destructive. After two years of a relentless AI rally, valuations and expectations sat at historic highs — any unverified negative signal was going to be amplified.
Multiple analysts characterized the sell-off as a "high-valuation sentiment amplifier" rather than a fundamental shift.
What do institutions say — has the AI capex cycle peaked?
Gary Tan, portfolio manager at Allspring Global Investments, argued the delay would only slow deployment of Nvidia's next-gen rack architecture. It would not alter global cloud providers' long-term demand for AI infrastructure. Core industry fundamentals remain intact.
He cautioned, however, that market sentiment is fragile. Uncertainty around supply-chain delivery timelines can easily trigger profit-taking and sector rotation.
South Korea's NH Investment & Securities added that the sell-off had clear regional and sector-wide characteristics. Markets are now re-pricing the delivery risk of next-gen compute platforms.
What has this episode actually changed?
The industry's core logic is shifting: AI has moved from a forward-expectations-driven phase into a delivery-verification phase.
Demand is not the problem — top-tier cloud providers' AI capex remains at peak levels. But supply-side bottlenecks are surfacing: advanced rack integration, cutting-edge PCB fabrication, and high-end semiconductor packaging are now the binding constraints on how fast compute hardware actually ships.
In plain terms = the market used to ask only "how much will AI spend?" Now it asks "once the money is committed, can the hardware actually be built on time?" — the pricing anchor has shifted from growth potential to delivery capability.
Content is for reference only, not financial advice.