Physics Deals Nvidia a Rare Handicap as 78-Layer Circuit Board Stalls Its AI Supercomputer Ambitions

The artificial intelligence revolution has long been a story of silicon and software, but Nvidia’s latest roadblock shows that the next chapter will be written by materials science. The chip titan’s ambition to dominate the highest echelons of AI computing has hit a tangible wall: a 78-layer printed circuit board that manufacturers simply cannot mass-produce at the required quality. The result is a delay of at least a year for the company’s flagship “Kyber” architecture, forcing a broader reconfiguration of its product roadmap.

At the center of the snag is the NVL144 server rack system, part of the Kyber platform originally slated for launch in 2027. SemiAnalysis reported on July 6, 2026, that the system will now arrive in 2028. The culprit is a highly complex midplane PCB that stacks 78 layers of circuitry — a density that current fabrication techniques cannot handle at commercial scale. The bottleneck has ripple effects across Nvidia’s lineup: the NVL576 model is now heavily delayed or destined for limited production, and an alternative double-rack design dubbed NVL72x2 was scrapped outright after pushback from key cloud customers. Even the processor itself has been scaled back: the top-tier “Rubin Ultra” chip has been downgraded from a quad-die to a dual-die configuration to keep manufacturing within achievable tolerances.

The pause in Kyber’s timeline opens a strategic window for rivals. While Nvidia wrestles with its engineering challenges, competitors can seize an opportunity to close the credibility gap in large-scale AI clusters. AMD’s Helios platform and the MI500X accelerator, as well as Google’s TPU 8t and TPUv8i chips, are now better positioned to pitch themselves as viable alternatives in the hyperscaler segment. For the moment, though, the existing product lines — Oberon and Rubin — continue shipping normally. Nvidia plans to deliver the current-generation Rubin racks to eight major cloud partners, including AWS and Azure, by fall 2026. The immediate revenue engine remains intact, even if the next-generation “super rack” has slipped.

Should investors sell immediately? Or is it worth buying Nvidia?

Markets have already begun pricing the delay, and the pain is spreading beyond Nvidia itself. Asian suppliers such as Ibiden and Samsung Electro-Mechanics saw double-digit percentage declines in their share prices after the 2028 timetable for Kyber became public. For Nvidia’s own stock, the reaction has been more contained. Shares were trading at €171.08 on the day of the report, a modest loss of 0.52%. That leaves the stock 15.52% below its 52-week high of €202.50 set in mid-May, but still 4.12% above its 200-day moving average of €164.31. The 14-day relative strength index sits at 42.8, indicating neither overbought nor oversold conditions — a consolidation phase in which investors are weighing the delay against still-robust fundamentals.

Year-to-date, the stock is up 6.19%, and over the past twelve months it has gained 26.71%. Analysts remain bullish on the long-term thesis, with an average price target of €263.59 — representing a 54.1% upside from current levels. The market valuation of roughly €4,124 billion reflects a wager that Jensen Huang’s team will eventually conquer the 78-layer barrier. But for now, Nvidia’s most daunting challenge is no longer proving demand for AI; it is mastering the mechanical engineering and supply chain logistics required to satisfy that demand at scale. The next quarterly earnings report, when Nvidia is expected to break its silence on the matter, will be the first real test of whether the market’s patience matches the company’s revised timetable.