Analyst Firm Reports Year-Plus Delay to Nvidia's Next-Generation 'Kyber' AI Server; Asian Supplier Stocks Fall

On July 5, 2026, the semiconductor research firm SemiAnalysis posted on the social platform X that Nvidia's next-generation Kyber NVL144 rack — a server cabinet designed to pack 144 of the company's top-end AI chips into a single unit — has slipped by more than 12 months, pushing its expected launch from roughly 2027 to 2028^[1][2][3]. The firm attributed the delay to manufacturing problems with a 78-layer printed circuit board, called a "midplane," that it described as one of the most complex boards ever attempted for a commercial product^[2][5]. The timing stung: Nvidia had publicly demonstrated Kyber only about three months earlier, at its GTC conference, and SemiAnalysis's post leaned into that gap, headlined "MASSIVE DELAY" in all capital letters^[8].

Markets reacted within a day. On July 6, shares of Asian companies that supply circuit boards and materials for Nvidia's systems fell sharply — Japan's Ibiden dropped about 10%, Hong Kong-listed Kingboard Laminates fell as much as 18%, Taiwan's Elite Material lost about 10%, and South Korea's Samsung Electro-Mechanics fell about 11%, all intraday^[2][5]. Nvidia's own stock moved far less, slipping about 1.4% to $194.83, leaving its market value near $4.7 trillion^[2]. Nvidia did not respond to requests for comment and has neither confirmed nor denied the report^[2].

What Isn't in Dispute

Strip away the framing, and a fairly narrow set of facts holds up. SemiAnalysis, a for-profit research firm, is the sole originating source of the delay claim, first surfacing on X before being expanded into a paid subscription note^[3][8]. The claimed slip is more than 12 months, targeting 2028, and the stated cause is the 78-layer PCB midplane^[1][2][3]. Nvidia showed Kyber at GTC roughly three months before the report^[8]. The specific stock declines — the 10% to 18% range across Ibiden, Kingboard, Elite Material, and Samsung Electro-Mechanics, and Nvidia's own 1.4% dip — are consistently reported across outlets^[2][5].

Just as important is what the report does not touch: Nvidia's current-generation Rubin systems remain on track for mass production, with shipments to major cloud customers such as Amazon Web Services and Microsoft Azure expected to begin in the second half of 2026^[7][11]. Whatever is happening with Kyber, it concerns a future, higher-end product rather than what Nvidia is shipping right now.

The Pressure Underneath

Each party in this story is operating under a structural incentive that shapes how it talks about the delay, regardless of the underlying facts. Nvidia's valuation rests partly on its promise of a new AI architecture every year, so the company has a strong interest in minimizing any perception that manufacturing physics is catching up with that pace — while also not confirming a specific roadmap it may still be revising^[7]. SemiAnalysis, meanwhile, earns its reputation and its subscription revenue by being first and specific about supply-chain problems before a company admits them, which rewards confident, attention-grabbing calls made well ahead of any official confirmation^[3]. The Asian board and materials makers sit downstream of both: their businesses are financially tethered to Nvidia's flagship demand, so their stocks are primed to move violently on any credible roadmap change, confirmed or not^[5].

Beneath all of that sits a physical reality nobody seriously disputes. A 78-layer orthogonal PCB midplane is genuinely at the edge of what mass manufacturing can reliably produce, raising real, non-trivial problems in signal integrity, power delivery, and thermal design^[3][7]. Whatever the final launch date turns out to be, that underlying difficulty — and the industry's dependence on a small number of advanced-board suppliers — isn't going away.

How Each Side Sees It

Nvidia's strongest implicit position, even in its silence, is that no product its customers are actually waiting on is late: Rubin is shipping on schedule to the biggest cloud buyers, and Kyber is a further-out, extreme-density design where taking extra time on a first-of-its-kind board reflects engineering caution rather than failure^[7]. In this framing, a private analyst's timeline is not the company's official one, and the modest 1.4% stock move suggests investors mostly agree, at least for now^[2][7].

SemiAnalysis, by contrast, argues its track record of specificity is exactly the point — it points not only to the midplane but also to a rejected NVL72x2 stopgap design that cloud providers reportedly pushed back on, and to complications with co-packaged optics on an even larger configuration, as evidence that Nvidia lacks a validated way to scale its top tier on schedule^[3][8]. The Asian suppliers, for their part, are simply repricing to a plausible new reality: firms like Ibiden, Kingboard, Elite Material, and Samsung Electro-Mechanics are heavily geared to Nvidia-linked demand, so a report of a year-long delay to a flagship board is, in their view, reason enough for a sharp market response regardless of official confirmation^[2][5].

Competitors and cost-conscious buyers see opportunity rather than crisis. For AMD and Google, a delay at Nvidia's highest tier is read as a rare window to win data-center business with their own chips — AMD's processors and Google's custom TPU silicon — validating buyers who worried Nvidia's once-a-year cadence might be outrunning what factories can actually build^[7]. Separate investor analysis has flagged smaller suppliers, including Marvell, Tower Semiconductor, and Lattice Semiconductor, as likely beneficiaries if hyperscale cloud companies diversify away from reliance on a single vendor's roadmap^[10]. None of these rivals, however, has been shown to have a validated, drop-in replacement for the specific high-end tier Kyber was meant to fill^[7][10].

How the Coverage Split

Mainstream business outlets such as Bloomberg and CNBC led with the market reaction and were careful with their hedging, repeatedly framing the delay as a "report" from a single research firm and giving prominent space to Nvidia's non-response^[1][2]. Investor- and trader-facing sites like Seeking Alpha and GuruFocus kept similar qualifiers but showed more interpretive reach — GuruFocus, for instance, generalized a single product's slip into concerns for the entire "AI market"^[2].

Right-leaning and pro-market financial commentary emphasized the competitive angle, framing the delay as a rare technical window for rivals rather than any sign that underlying AI demand is weakening^[7][10]. Asian business media, including the Seoul Economic Daily, treated the story primarily as a regional supply-chain event, foregrounding the specific named suppliers and their exact percentage losses ahead of the broader American tech narrative^[5]. And the original source, SemiAnalysis's own X post, used unmistakably promotional language — "MASSIVE DELAY" in capital letters — a reminder that the account driving the entire episode also comes from a firm with a direct commercial interest in making its scoop sound as consequential as possible^[8]. Underneath all of it, the plain facts remain unchanged: a genuinely hard circuit board to manufacture, a cluster of Asian suppliers absorbing a real market hit, and a company that, as of this writing, has said nothing at all^[2][3][5].