A PCB is becoming the most crowded business in the AI industry chain.

Over the past two years, the most prominent segment in the artificial intelligence industrial chain has belonged to chips.

NVIDIA's GPUs have become core assets that draw global capital market attention. Servers, optical modules, and data centers centered on computing power have also become the main line of repeated trading in the A-share market.

But in 2026, a new change is emerging.

The market has begun to shift its focus from the upstreammost chip segment to links closer to the manufacturing end.

PCB (Printed Circuit Board), a component that was long considered part of the traditional electronics manufacturing industry, is now becoming a key part in the upgrading of AI infrastructure.

Since September, the PCB industrial chain has remained continuously active.

From high-speed PCBs to copper clad laminates, and then to copper foil materials, many listed companies have attracted market attention.

As of the close of trading on September 23, Aohong Electronics maintained its strong momentum, notching 7 daily limit ups in 9 trading days; although Daxiang Shengya has issued a clarification stating that the ultra-thin PCB fiberboard in its wood-based panel products is not an electronic component, does not form part of a PCB, and is only used to pad under the copper clad laminate (raw material for PCB) to play an auxiliary and protective role, its stock price still recorded 4 consecutive daily limit ups; in addition, multiple stocks including Zhongyi Technology, Jitai Co., Ltd., Jiangnan New Materials, and Yiton Electronics have also hit their daily limit ups one after another.

What the market is trading is no longer just the "AI concept", but a more practical question: when AI computing power continues to expand, who can support the next-generation computing infrastructure?

The answer is pointing to a circuit board.

01

The Underestimated PCB That Is Bottlenecking AI Computing Power

The AI competition over the past two years is essentially a computing power race.

Whoever has more GPUs and larger data centers will have stronger large model training capabilities.

Therefore, the capital market first focused on chips.

AI chips, servers, and optical modules have become the most sought-after directions in the industrial chain.

However, as AI infrastructure continues to expand, a new problem has emerged: if the number of GPUs keeps increasing, where is the bottleneck that truly limits the development of AI?

The answer is shifting from chips to the entire system.

The reason is very simple.

AI servers are no longer the same type of equipment as traditional servers.

Traditional servers rely more on CPUs for computing, while AI servers require a large number of GPUs to work in coordination.

This means higher data transmission volume, higher signal frequency, greater power consumption, and more complex connection requirements.

The stronger the chip performance, the higher the requirements for the infrastructure that carries the chips, and the importance of PCBs is being rediscovered in this process.

PCB is known as the "mother of electronic products". In the past, it mainly undertook the connection function. Mobile phones need PCBs, computers need PCBs, and automotive electronics also need PCBs.

But AI servers are changing this logic. Because what AI servers need are not ordinary PCBs, but high-speed, high-frequency, low-loss PCBs. They need to carry data transmission between GPUs, switching chips, high-speed storage, and power systems.

A single PCB begins to determine whether the entire computing power system can operate efficiently. This is why the value of the AI industrial chain is shifting from chips to infrastructure.

02

Apart from Chips, There is Also a "Copper Foil War" in AI Hardware

If you further disassemble a high-end PCB, you will find that the real competition is not only in PCB manufacturing, but the upstream materials are becoming a new battlefield.

Behind a high-performance PCB, copper foil, fiberglass cloth, resin, and copper clad laminate are required. Among them, copper clad laminate determines the performance of PCB, and what AI servers need are high-frequency and high-speed materials.

In the process of high-speed data transmission, signal loss has become a new limiting factor, so the importance of low-loss materials keeps rising.

HVLP copper foil is a representative of such materials.

Compared with ordinary copper foil, HVLP copper foil can reduce high-speed signal transmission loss, and is one of the important materials for high-speed PCBs in AI servers.

According to the research report of Soochow Securities, the demand for high-end HVLP copper foil dedicated to AI servers in 2026 is about 24,000 tons, a year-on-year increase of about 260%; the demand is expected to further rise to 50,000 tons in 2027.

However, the problem of high-end materials cannot be solved simply by expanding production capacity. It requires equipment upgrading, process accumulation and customer certification.

This means that the AI industrial chain is forming a new contradiction between supply and demand: not all PCBs are scarce, what is really scarce are high-end PCBs and materials that can meet the requirements of AI servers.

This is why the capital market has begun to re-evaluate PCB enterprises, and Shanghai Dianxun Co., Ltd. is a typical case among them.

In the past, investors looked at Shanghai Dianxun Co., Ltd. more as a PCB manufacturing enterprise. But in the AI era, the market has begun to re-understand it.

The core value of the company is not just manufacturing PCBs, but being able to produce high-end PCBs that meet the requirements of high-speed servers.

This change means that PCB enterprises are transforming from traditional electronic manufacturing enterprises to AI infrastructure suppliers.

The same change is also taking place in material enterprises, with Shengyi Technology representing the copper clad laminate direction.

In the past, material enterprises were often in a relatively low position in the industrial chain. But in the AI era, a new logic is taking shape: materials determine the upper limit of performance.

Without better materials, chip performance cannot be fully released, and server speed cannot be further improved.

Therefore, AI competition is not only competition for chips, but also competition for materials.

At the same time, enterprises including Dongshan Precision and Shenghong Technology are also continuously making layouts in the high-end PCB field.

They are all facing the same industrial trend: the growing demand for AI servers is pushing the PCB industry to shift from scale competition to technology competition.

In the past, the one with lower cost had more advantages; in the future, the one that can produce higher-end products will have more advantages.

03

AI "Shovel Seller", It's PCB's Turn

Every industrial revolution will go through a similar process.

At the very beginning, the capital market focuses on the most eye-catching technologies.

In the Internet era, attention was paid to platforms; in the new energy era, attention was paid to complete vehicles and batteries; in the initial stage of the AI era, attention was paid to GPUs and large models.

However, as the industry enters the large-scale development stage, the market begins to look for infrastructure, and PCB is exactly at this stage.

It is not the most glamorous segment of the AI industry, but it is an unavoidable link for the large-scale implementation of AI.

Because artificial intelligence does not eventually run in concepts, it runs in servers, servers run in data centers, and data centers run on a large number of basic components.

Therefore, the AI industrial chain will definitely return to manufacturing in the end.

However, the capital market also needs to see the other side of the story.

The PCB industry has experienced multiple cycles in the past — rising demand, enterprises expanding production, production capacity release, and price decline.

The biggest difference this time is that the demand brought by AI is more concentrated in the high-end field.

Enterprises that will truly benefit in the future need to meet three conditions: first, have the capability for high-end products; second, be admitted to the supply chain of core customers; third, convert demand into profits.

The capital market can trade on industrial trends in advance, but what ultimately determines the value of an enterprise is still orders and financial data.

This is why the AI PCB market trend will not simply replicate the past electronics cycles.

The future competition is not about who has more production capacity, but about who masters the most scarce manufacturing capabilities in the AI era.

Over the past two years, people have been looking for those who make the AI brain, then those who connect the AI brain, and now those who support the operation of the AI brain.

A PCB seems to be just a circuit board in electronic equipment.

But in the AI era, what it carries is not only circuits, but the entire computing power infrastructure — from GPUs, to optical modules, to PCBs, and then to copper foils, copper clad laminates and testing equipment.

Artificial intelligence will eventually return to manufacturing.

And a PCB is becoming the new "shovel selling business" in the AI era.

This article is from the WeChat official account "Alpha Workshop Research Institute", the author is Beichen, and it is published with authorization from 36Kr.