Nvidia's Rumored 10% PCB Price Cut Denied by Multiple Parties; Kyber Backplane Solution May Be Delayed to 2028
0xBroomberg
A rumor that Nvidia demanded a 10% price cut from PCB makers was widely denied by industry insiders, who say pricing power still sits upstream; the real risk flagged by Jefferies is that the Kyber backplane PCB may slip to at least 2028, trimming the 2027 AI PCB market forecast by roughly 5%.
Does the "10% price cut" rumor hold up?
Shanghai Securities News checked with multiple PCB makers and market participants — respondents broadly denied the claim and called it a clear exaggeration.
Huachuang Securities analyst Xiong Yiyu noted that pricing is negotiated between makers and clients, and PCB makers still hold the upper hand.
A PCB-listed-company chairman added: persistent upstream material price hikes have shifted supply-chain pricing from semi-annual/annual lock-ins to real-time quotes every 1–3 months — actual pricing power sits with upstream material suppliers.
This means → PCB makers are not passive order-takers; with cost pressure flowing down from upstream, they have leverage to reject one-sided cuts.
Is AI-server PCB pricing the same game as consumer electronics?
A private-equity investment manager pointed out that AI-server PCBs carry higher value-add, and downstream clients show markedly greater price acceptance — a fundamentally different pricing logic from consumer-electronics PCB.
He characterized the current dynamic as "guided price management," not "forced price cuts."
In plain terms = what cloud builders fear most right now is not high prices but supply-chain delivery failure; stable delivery matters far more than squeezing unit cost.
Is Rubin's delay really caused by one PCB supplier?
Another rumor blamed Shenghong Technology's slow capacity expansion for Nvidia's Rubin shipment delays — also denied by multiple sources.
The Rubin platform spans GPUs, HBM — high-bandwidth memory — advanced packaging, PCBs, switches, power supplies, liquid cooling, and rack systems: many critical links, of which PCB is just one.
TrendForce research manager Gong Mingde listed the real causes of delay: longer-than-expected HBM4 certification, CX8-to-CX9 network upgrade adaptation, whole-rack power management after a sharp rise in single-chip power draw, and tuning of higher-spec liquid-cooling solutions.
He expects Rubin AI servers to ship in small batches by late Q3 2026, ramping in Q4; full-year shipments will still center on GB300, with Rubin at only 10–20%.
What does the Kyber backplane delay mean?
A Jefferies research note flagged that the Kyber orthogonal backplane PCB — a design that replaces copper cables with a printed-circuit-board backbone inside the rack — planned for Rubin Ultra in 2027, may slip to at least 2028 due to high technical complexity.
Rubin Ultra in 2027 will most likely stick with the Oberon architecture (i.e., NVL72).
This means → if Kyber slips to 2028, Jefferies estimates the 2027 global AI PCB market will be ~5% below its prior forecast, and the CCL — copper-clad laminate, the core raw material for PCBs — market ~8% below.
If the delay extends further or the project is canceled, 2028 PCB and CCL markets could drop by 11% and 16%, respectively.
Who actually benefits from the delay?
Jefferies also stressed that the delay does not change PCB's long-term growth thesis: switch boards and mid-boards are still upgrading to higher-spec materials such as M9, M10, and PTFE.
If the Oberon architecture's life cycle extends, copper-cable demand that was set to be replaced by backplane PCBs will be preserved — copper-cable makers stand to gain.
Upstream glass-fabric and CCL suppliers, facing persistently tight supply, are expected to maintain stronger margins than the PCB fabrication segment.
In plain terms = backplane delayed → copper cables stay in use one more year → cable makers get an unexpected bonus window; upstream material suppliers, meanwhile, profit regardless of architecture because demand outstrips supply.
What is the key variable after 2027?
Whether the Kyber solution lands on schedule is the pivotal checkpoint for whether AI PCB growth expectations hold beyond 2027.
This reflects a broader reality: the AI hardware supply chain does not accelerate in a straight line — each new architecture faces simultaneous-readiness hurdles across HBM certification, power management, and thermal solutions before volume production.
For investors: short-term rumors (a 10% price cut, one supplier causing delays) are loud but poorly supported; the Kyber timeline is the variable that actually moves the 2027–2028 industry-size forecast.
Content is for reference only, not financial advice.