Home/Research/China's PCB Supply Chain: Real Chokepoints Inside the Rubin Narrative
中文
May 26, 2026 阅读中文版

China's PCB Supply Chain: Real Chokepoints Inside the Rubin Narrative

Executive Summary

In late May 2026, China’s A-share PCB supply chain became one of the hottest corners of the AI infrastructure trade. The trigger was simple: the market started to price a much higher PCB bill of materials for NVIDIA Rubin and GB300 servers. High-layer boards, high-speed materials, OAM boards, NVSwitch trays, ABF substrates, and M9 copper-clad laminates were all bundled into one “AI-server chokepoint” narrative.

This report is not a single-stock call. The more important question is structural: when a supply chain is repriced as a group, which companies are actually close to a technical or capacity chokepoint, and which are merely carried by the same diagram?

My conclusion is that the AI-server PCB opportunity is real, but it is not evenly distributed across the A-share PCB chain. The more concrete chokepoint candidates are concentrated in three areas: Wus Printed Circuit’s high-layer and high-speed boards, Victory Giant’s OAM and HDI exposure, and Shengyi Technology’s high-speed CCL materials. Pengding, Shennan Circuits, Kinwong, Dongshan Precision, and others each have real businesses, but the Rubin narrative requires a stricter split between capability, verified design wins, and financial conversion.

A Supply-Chain Map Is Not an Order Book

The biggest analytical mistake in the AI-server PCB trade is to treat a supply-chain map as if it were an order book. A map can show capability, potential participation, or ecosystem position. It does not prove that a company has secured volume orders from NVIDIA, ODMs, or server OEMs. It also does not prove that such orders will become sustained excess profits.

This distinction matters. Rubin and GB300 servers do raise the technical bar for PCBs: higher signal speeds, more layers, stricter material-loss requirements, and more complex internal interconnects. High-end PCB and CCL content per server should rise. But higher content at the system level does not give every supplier equal pricing power.

A cleaner framework has three layers.

The first is capability: does the company have high-layer, high-speed, HDI, OAM, ABF, or large server-board manufacturing capability?

The second is verification: do we have evidence on customer, platform, share, yield, or production timing?

The third is conversion: is the opportunity visible in revenue, margin, orders, or capital expenditure?

In the current A-share PCB narrative, there is plenty of evidence on the first layer. The second layer is partially obscured by NDAs and secondary sourcing. The third layer will require 2026 interim results and second-half ramp data.

Three Positions Closer to the Chokepoint

KSINQ’s local research points to three positions that deserve to be separated from the broader list.

The first is Wus Printed Circuit. The market repeatedly links the company to AI-server high-layer and high-speed boards, with keywords such as 78-layer M9, 800G, and NVSwitch tray PCBs. The attraction is that the technical parameters are concrete: layer count, material grade, and server position all map to AI-server architecture. The caveat is equally important: NVIDIA’s supplier disclosure is not transparent, and many share figures still come from brokers, industry media, or secondary references rather than primary-source confirmation.

The second is Victory Giant. The market discussion centers on GB200 / B300 OAM boards, five-step HDI, and AI-server PCB share. Victory Giant has one advantage over more indirect beneficiaries: its first-quarter 2026 financials already looked stronger than companies such as Pengding, suggesting that high-end data-center demand has begun to show up in reported numbers. The risk is that valuation has moved first, while customer mix and platform share still need to be verified in later filings.

The third is Shengyi Technology. Shengyi is not simply a PCB manufacturer. It is an upstream CCL materials company. The AI-server PCB bottleneck is not only board processing. It is also low-loss material, high-speed stability, yield, and reliable supply. KSINQ’s local notes repeatedly refer to M9 high-speed CCL, yield advantages, and Rubin-linked supply-chain claims. These point to an important fact: if AI-server PCB content rises, upstream material suppliers may have a cleaner margin lever than some downstream assemblers. But many specific share and yield numbers remain secondary-source claims and still need company disclosures, customer evidence, or interim results.

These three companies are not “certain winners.” A better description is that their stories contain more concrete and testable anchors. The market does not only need stories. It needs stories that can be verified.

Why Pengding Looks More Indirect

Pengding is the most useful counterexample in this cycle. It is a high-quality PCB company with strong industry status, but that does not automatically make it a Rubin chokepoint.

Its historic strength is in Apple-related FPC, communication boards, and rigid-flex boards. It can benefit from better PCB risk appetite, and it may have indirect exposure through Zhen Ding and the Foxconn ecosystem. But whether the A-share listed entity itself is carrying core NVIDIA Rubin or GB300 high-value orders requires clearer disclosure.

The first-quarter 2026 peer split is also notable. Pengding’s revenue fell 1.25% year over year, and net income fell 8.4%. Victory Giant, Dongshan Precision, and Shengyi showed much stronger growth signals. If AI-server PCB demand has already started to appear in the financials of some suppliers, Pengding’s underperformance cannot be explained away by saying “the entire supply chain benefits.”

There is also a structural signal. Controlling shareholders Meigang Industrial and Jihui International reduced their holdings by 24 million shares for roughly RMB 2.34 billion on May 21, 2026, around the same time the market was discussing Rubin PCB value uplift. Insider selling does not prove the company lacks opportunity. It does mean that calling Pengding a core Rubin chokepoint requires a high evidence bar. Behavior disclosed in filings is harder evidence than a viral supply-chain diagram.

This does not make Pengding a bad company. It makes it look more like an indirect narrative beneficiary than the strongest currently evidenced AI-server PCB chokepoint.

ABF Substrates: Domestic Substitution Is Not Yet the Same Table

Another area where the narrative is often compressed too quickly is ABF substrates. Advanced substrates are crucial for AI accelerators, CPUs, GPUs, ASICs, and high-end packaging. The long-term importance of ABF is not in doubt. The short-term issue is that Chinese suppliers still trail the Japanese and Taiwanese leaders in the highest-end segment.

KSINQ’s local research is blunt: high-end ABF substrates remain dominated by companies such as Ibiden, Unimicron, and Nan Ya. Chinese suppliers are investing and catching up, but “having a program” is not the same as “being qualified for the highest-end platform.” Shennan Circuits has substrate capability, but the gap to leading-edge line width, layer count, yield, and customer qualification still matters.

That means investors should not translate “AI servers need ABF” directly into “A-share ABF suppliers are already at the core table.” This is a real long-term domestic-substitution direction. It should not be pulled forward into the income statement before evidence appears.

Valuation Moves Before Evidence, but Evidence Eventually Arrives

This PCB rally has a classic structure: price moves before evidence. Rubin volume ramp is mainly a second-half 2026 event. Listed-company interim results and third-quarter results will gradually reveal who is converting the story into revenue and margins. It is understandable that the market priced the value-content uplift first in May. It also means the margin for error is now lower.

The next verification windows are straightforward.

First, 2026 interim-result previews. True chokepoint suppliers should show stronger signals in orders, revenue, gross margin, or capital expenditure.

Second, customer-mix disclosure. Even if companies cannot name NVIDIA, investors should see more evidence around AI servers, overseas hyperscale customers, high-speed boards, or data-center board mix.

Third, yield and capacity. The bottleneck in high-end PCB is often not whether a company can build samples, but whether it can mass-produce at acceptable yield.

Fourth, insider and major-shareholder behavior. If shareholders sell into the narrative while the company does not provide order verification, the combination deserves attention.

Conclusion

China’s PCB supply chain is genuinely exposed to AI-server content growth. But inside the Rubin narrative, the valuable asset is not being listed on a diagram. It is a verifiable technical chokepoint that converts into financial results.

Wus Printed Circuit, Victory Giant, and Shengyi Technology have more concrete verification anchors. Pengding, Shennan, Kinwong, Dongshan, and others need to be evaluated according to their actual business position. Supply-chain mapping is a research starting point. It is not a substitute for design wins or reported financial conversion.

The real split will come after 2026 interim results: which companies were merely illuminated by the Rubin narrative, and which companies actually became part of the AI-server income statement.

Data and Sources

This report is an independent KSINQ market observation for informational purposes only. It is not investment advice. Data snapshot: May 26, 2026.