Zhen Ding Technology Holding Limited (4958.TW): Deep Dive

Research as of 19 October 2026. Zhen Ding reports quarterly with a December year-end. Q3 2026 results are expected around 11-12 November 2026: the Q3 2025 results came out on 11 November 2025, and the 13 October 2026 Hong Kong investor forum disclosed no Q3 financials. The latest reported period is therefore 1H26 / Q2 2026, released at the investor conference on 11 August 2026, 69 days ago. Q3 2026 monthly revenue has already been published.

1. What the Company Does

Zhen Ding makes the circuit boards inside electronics. Every phone, laptop, AI server and optical transceiver has one or more boards: thin sheets of insulating material with copper wiring etched onto them, onto which chips and components are soldered. Zhen Ding does not design the devices and does not assemble them. It builds the wiring layer that everything else sits on, to the customer's design and in very large volumes.

Its specialty has been boards that are hard to make. The largest product family is the flexible printed circuit (FPC), a bendable board that folds around the inside of a phone, connects the camera to the main board, or runs through a smartwatch strap. The second is the high-density rigid board: substrate-like PCBs (SLP) and HDI boards fine enough to carry a smartphone's processor, and high-layer-count (HLC) boards up to 34 layers for AI servers. The third is the IC substrate, the tiny multi-layer board a chip die sits on before it reaches the main board. In 2025 the group had the most PCB revenue in the world for the ninth year running, by the company's own count, which matches Prismark's 2025 ranking (I-Connect007; SLL PCB citing Prismark).

How it became this. The operating core is older than the listed company. Avary Holding's predecessor, Fukui Precision Component (Shenzhen), was set up in Shenzhen on 29 April 1999 inside the Hon Hai (Foxconn) group and renamed Avary Holding (Shenzhen) in May 2017 (Avary corporate records via Sina / Baidu). Zhen Ding Technology Holding was formed in 2006 as the offshore holding company over these board plants. It listed on the Taiwan Stock Exchange in 2011, which was the group's first IPO (Zhen Ding group page via Pitchbook / company profile). Its "-KY" suffix marks a foreign-incorporated issuer.

Several later decisions shaped today's business:

• 2018: Avary listed on the Shenzhen exchange (002938.SZ) on 18 September 2018. This put the main operating company, and its mainland plants in Shenzhen, Qinhuangdao and Huai'an, under a separately listed Chinese entity that Zhen Ding controls.
• 2019: Leading Interconnect Semiconductor Technology (Liding) was founded in Shenzhen as the IC substrate arm. This was the move from boards toward chip packaging (163.com on Liding's HK filing).
• November 2020: Boardtek Electronics was acquired through a share swap. Boardtek was founded in 1987 as Yuan Feng Electronics and gave the group a Taiwanese rigid-board and high-layer-count maker. That is the capability AI servers now need.
• 2025-2026: the AI pivot. A Thailand plant opened, a Kaohsiung "AI Park" was built for ABF substrates and advanced boards, and from early 2026 the company began what management calls the largest capacity expansion in its history (Section 3).

Zhen Ding used to be a phone-board supplier with one dominant customer. It is trying to become a supplier of the interconnect inside AI data centres: server boards, optical-module boards and chip substrates. The phone business still funds that shift.

The problem it solves. Device and system makers need boards that are thinner, finer-pitched, more layered and lower-loss each generation, delivered in tens of millions of units at consistent yields and on the product's launch schedule. Few manufacturers can do all of this at once. The company's pitch is "one-stop shopping" across flexible, rigid, HDI and substrate technology (vocus summary of the March 2026 call).

Why it is hard to replicate (headline only). Fine-line boards are made with the modified semi-additive process (mSAP), which grows copper traces instead of etching them away. It took the group years to reach high yields with it, and many entrants never did (Section 5). Flexible boards at phone volumes also need large, tightly controlled plants that cannot be built quickly.

An example in practice. Take an 800G or 1.6T optical transceiver, the pluggable module that turns electrical signals into light to link GPUs across a data centre. A module maker such as InnoLight or Eoptolink designs the module. Zhen Ding builds the small, dense board inside it, with traces fine enough that the leading-edge versions need mSAP. Zhen Ding has to qualify the board with the module maker, ramp yields as the module goes to volume, and keep supplying as the cloud provider places orders. Management named these module makers as its primary optical customers (Fugle memo, June 2026 investor meeting). A phone works the same way at larger scale: the board layout is co-developed with the phone maker's engineers about a year before launch, then shipped in the tens of millions within a few months.

At the May 2026 AGM, management put the strategic claim this way:

"AI is redefining the role of PCBs, elevating them from traditional signal-interconnection components to critical system carriers for high-performance computing." (I-Connect007, Zhen Ding 2026 AGM, 29 May 2026)

2. Business Segments

Zhen Ding reports revenue by application, not by legal entity: Mobile Communications, Computing & Consumer, Server/Optical, and IC Substrates. Management usually groups the last two as the "AI infrastructure" bucket. Underneath, there is also a split by technology (flexible vs rigid vs substrate) and by listed entity (Avary in Shenzhen, Liding heading to Hong Kong, and Boardtek and the Kaohsiung operations in Taiwan). This section follows the application split, because that is how management talks about priorities.

2025: Mobile 61.3% and Server/Optical + IC Substrates 11.6% (Nikhs Substack analysis of 4Q25); IC substrates "over 6%" (Fugle memo, Dec 2025 investor meeting). 1H26: 53.6% / 24.8% / 21.6% (BigGo, 1H26 call).

2.1 Mobile Communications

What it does. Flexible boards and rigid-flex assemblies for smartphones (display, camera module, battery and antenna connections), plus SLP and HDI main boards that carry the application processor. The end market is premium smartphones, which in practice means one dominant customer (Section 4). Foldable phones and AI-enabled handsets raise the board content per device. Management says a foldable carries two to three times the board value of a standard phone (Fugle, Dec 2025).

Core capability. Making very thin, very fine multi-layer flexible circuits at volumes few rivals can match, and launching a new design every year on a fixed calendar. The company's mSAP experience also started here: SLP for phones needed fine-line rigid boards years before AI optics did.

Why it is a separate segment. It has a different demand rhythm (an annual launch cycle with a heavy second half), one buyer setting the specifications, and economics driven by yield and utilisation on very large plants rather than by pricing.

How it fits. This is the cash cow and the scale base. It pays for the AI build-out and absorbs fixed costs. Management still calls mobile a growth area through foldables and AI phones, but its share of the mix falls each quarter by design.

Main rivals: Nippon Mektron, Dongshan Precision (through its MFLEX unit), and Korean and Taiwanese FPC makers.

2.2 Computing & Consumer

What it does. Boards for notebooks, tablets, smartwatches, earbuds and, more recently, AI glasses and AR headsets. Management claims Zhen Ding is the largest global supplier of boards for AI glasses (Fugle, Dec 2025). It also mentions humanoid robots as an emerging application, with no disclosed revenue.

Core capability. The same flexible and HDI know-how as Mobile, applied to smaller and odder form factors where miniaturisation decides the design.

Why it is separate. It has a different product cycle and a broader, though still concentrated, customer base, and its volumes are more exposed to PC and wearables cycles than to the phone calendar.

How it fits. A second consumer pillar that diversifies the phone rhythm a little and acts as an option on new device categories. Management ranks it last of the group's product lines for profitability (Section 9 context).

Main rivals: the same FPC field as Mobile, plus Taiwanese HDI makers in notebooks.

2.3 Server/Optical

What it does. Two different products that management reports together:

• AI server boards: high-layer-count motherboards and accelerator boards ("HLC") and the denser "iHDI" boards for GPU and custom-ASIC platforms. The company showed boards of up to 34 layers at Semicon Taiwan in September 2026 (Digitimes, 2 Sep 2026).
• Optical-module boards: the small, very fine boards inside 800G and 1.6T transceivers, plus next-generation formats such as XPO and NPO.

Core capability. For optics, mSAP fine-line production at high yield (Section 3). For servers, a newer capability built partly on Boardtek's rigid-board heritage and partly on large new Chinese and Thai capacity.

Why it is separate. Different customers (module makers, GPU and ASIC vendors and their board assemblers rather than phone makers), different materials (ultra-low-loss laminates and low-expansion glass cloth), and 3-5 year long-term agreements with prepayments instead of annual phone allocations (Section 4).

How it fits. This is the growth engine and, according to management, the best-margin product line in the group. Optical boards rank first in profitability, and AI server boards rank above consumer computing (BigGo, 1H26 call). Most incremental capex goes here.

Main rivals: in optics, Unimicron and Compeq; in AI servers, Victory Giant, WUS, Gold Circuit and TTM.

2.4 IC Substrates

What it does. Two families of chip-package substrates. BT substrates (memory, RF and smaller packages) made up about 65% of the substrate mix in late 2025. ABF substrates (large flip-chip packages for CPUs, GPUs, ASICs and networking chips) made up about 35%, and roughly 60% of ABF revenue already came from AI-related products (Fugle, Dec 2025). Package types include FCBGA, FCCSP, WBCSP and module substrates (163.com on Liding).

Core capability. Large-body ABF. Management says it has reached 158×158mm at 28 layers, which it describes as comparable with tier-one rivals (vocus summary of the March 2026 call; vocus summary of the 13 Oct 2026 forum). This is a self-assessment, and the benchmark it is chasing is Ibiden.

Why it is separate. Substrates are a semiconductor-packaging product with different equipment, cleanroom standards and qualification chains (chip designers and OSATs). They are also housed in a separate legal entity, Liding, which filed for a Hong Kong main-board listing on 2 July 2026. Zhen Ding directly and indirectly controls 60.75% of Liding before the listing (163.com).

How it fits. A long-dated strategic option that has only just become profitable. Management says ABF turned profitable in Q1 2026 (Fugle, June 2026).

Main rivals: Ibiden, Unimicron, Shinko, AT&S, Kinsus and Nan Ya PCB.

2.5 Geographic mix and export history

Two different geographies matter here, and they point in opposite directions.

Where sales are billed. In 2024, about 77% of product sales were billed to the United States, 13% to mainland China, 5% to Taiwan, 1% to Singapore and 4% elsewhere (2024 annual report data as surfaced in search results; the 2025 split was not located). The "US" figure reflects the procurement entity of the dominant American customer, not where boards physically go.

Where goods physically go. Boards ship to assembly plants in China, Vietnam, India and Taiwan. Management said only about 0.8% of 2024 revenue was shipped directly into the US (CNA, 12 Aug 2025), which is why it describes direct US tariff exposure as limited (Section 8.4 covers the indirect exposure).

Where boards are made. Production has been overwhelmingly in mainland China since 1999: Shenzhen, Qinhuangdao and Huai'an. Taiwan adds Boardtek's rigid-board plants and the new Kaohsiung AI Park. Thailand (Prachinburi) is the newest base, and the company profile also lists an India presence (company profile). The chairman has said the Thailand and Kaohsiung expansion is driven by geopolitics, not labour cost (Anue, 12 Mar 2026). Customers selling to Western hyperscalers increasingly want boards made outside China.

3. Products and Operations

3.1 The product families that matter

Flexible printed circuits (FPC). Thin polyimide film with copper circuits, made in single, double and multi-layer forms and as rigid-flex hybrids. In late 2025 flexible boards made up over 70% of revenue and rigid boards close to 30% (Fugle, Dec 2025). They are hard to make well because tolerances on bend radius, impedance and layer registration are tight on very thin material, and a phone uses many different FPC designs that must all launch on the same day.

SLP and HDI. HDI boards use laser-drilled micro-vias to pack more wiring into each layer. SLP goes further, borrowing substrate-style mSAP to reach line widths a conventional subtractive (etch-away) process cannot. The company's mSAP experience comes from phone SLP. Management calls its AI-era versions "iHDI" (intelligent HDI).

High-layer-count (HLC) boards. Thick multi-layer rigid boards for servers and switches. AI platforms push layer counts into the 30s, use ultra- and extreme-low-loss laminates, and need tight back-drilling to keep signals clean at 224G SerDes speeds. Quartz-glass ("Megtron 9"-class) laminates are brittle and abrasive and wear out standard drill bits (Nikhs Substack), which is one reason the company is building a dedicated mechanical-drilling centre in Thailand.

Optical-module boards. Small, dense boards inside transceivers. As data rates rise, line and space shrink. At 1.6T the industry uses roughly 20-30 micron lines. The 3.2T generation, expected commercially around 2028, needs 10-15 microns, 18-20 layers and lower-loss materials, which pushes production firmly into mSAP (Cloud News citing Digitimes, Oct 2026). Management said 1.6T boards were about 5% of optical volume but about half of optical revenue (BigGo, 1H26 call). Each step up in data rate raises the value per board much faster than the unit count.

IC substrates. BT and ABF, as described in Section 2.4. The critical capability is large-body ABF. Bigger AI packages need bigger, flatter, higher-layer substrates, and yields fall quickly as body size grows.

3.2 How a board is made (for a flowchart)

1. Design-in. The customer's engineers and Zhen Ding's application engineers co-design the stack-up, choose materials and set impedance targets. For a phone this happens about 12 months before launch. For AI platforms it runs alongside the chip platform's development.
2. Material sourcing. Copper-clad laminate (CCL), glass cloth (including low-expansion "T-glass"), copper foil, polyimide film and ABF film. Glass cloth comes from suppliers such as Nittobo, Taiwan Glass and Grace Fabric, and laminate from eight to ten CCL vendors including Doosan (BigGo, 1H26 call).
3. Inner-layer imaging. Etched (subtractive) for conventional boards; seed-layer, pattern and electroplate (mSAP) for fine-line boards and substrates.
4. Lamination. Layers are stacked and pressed. Registration across 20-34 layers is the yield killer.
5. Drilling. Laser micro-vias for HDI and substrates; mechanical and back-drilling for HLC.
6. Plating, surface finish, solder mask.
7. Electrical test and reliability qualification, then shipment to the module maker, OSAT or EMS assembler.

Equipment lead times are 12-18 months, and the company books key raw materials 12-18 months ahead (vocus, 13 Oct 2026 forum). In this industry, capacity is decided more than a year before it can be used.

3.3 The manufacturing footprint and plant events

At the 1H26 call the company said it had 13 new plants under construction and 16 more in planning, with capacity coming online between 2026 and 2030 (I-Connect007, 1H26; UDN, 11 Aug 2026). Investor-forum summaries quote different counts (22 under construction and 9 planned, from a base of 28 plants rising toward 59 by 2030). This report uses the figures from the company's own 1H26 release. Either way, the plant count roughly doubles within four years.

The current capex mix is weighted toward rigid boards: about 65-70% for rigid (servers and optics), 15-20% for IC substrates and 10-15% for flexible boards (Fugle, June 2026). The size of the capex budget and how it was revised are covered in Section 9.

On sourcing, management says the scarcest inputs, including T-glass, high-end copper foil and drill bits, are booked through 2026-2027, and talks are under way for 2029-2030 supply. It also notes that "suppliers are also choosing their customers" (BigGo, 1H26 call).

4. Customers

4.1 Who buys

Four distinct buyer groups:

1. A US consumer-electronics company (Mobile and Computing & Consumer). Zhen Ding's filings do not name it. Avary's 2025 annual report shows its five largest customers at about 89% of sales, with the largest, "Customer A", making up the great majority (Avary 2025 annual report, via search extract). Industry coverage consistently identifies Zhen Ding as Apple's main flexible-board partner (Nikhs Substack), and the 77% US billing share (Section 2.5) fits that.
2. The Hon Hai group. Avary's annual report names the Hon Hai group and its subsidiaries as a major customer ("Customer B") and treats it as a related party, because Foxconn (Far East) is Zhen Ding's largest shareholder (Section 10). Hon Hai assembles devices and servers, so some of Zhen Ding's boards pass through it.
3. Optical-module makers. Management named InnoLight and Eoptolink as its main optical customers (Fugle, June 2026). They in turn sell to the large cloud providers.
4. GPU and ASIC platform owners and their board assemblers (servers), and chip designers and OSATs (substrates). Management refers to "GPU/ASIC customers" without naming them. It says new ABF capacity will bring in "large international customers" beyond the existing base (Q1 2026 call summaries).

4.2 How each buys

Phones and devices. Decisions sit with the device maker's product-design and procurement teams, not with the assembler. Suppliers are qualified per component and per model a year or more before launch. The buyer typically splits each part between two or three qualified suppliers and resets the allocation every product cycle. Zhen Ding wins share by solving the hardest new design (a new camera module flex, a foldable hinge flex, a thinner SLP), shipping at yield on day one, and having enough capacity to absorb launch surges.

Optical modules. The module maker's engineering team chooses the board supplier, but the cloud customer's qualification of the finished module makes the board supplier hard to swap mid-program. The deciding factors are fine-line capability at the next data rate, yield and available capacity. Management says customers had already begun pre-booking 2027 capacity by March 2026 (Yahoo/MoneyDJ, Mar 2026). In a shortage, the buyer's main question is whether the supplier can deliver the volume.

AI servers. Board suppliers are put on approved vendor lists by the GPU or ASIC platform owner and then allocated volume by the assemblers. Qualification runs with the platform's development, and once a board is qualified for a platform generation it tends to stay for that generation. Zhen Ding is a relatively new entrant in AI servers compared with Victory Giant and others.

Substrates. The longest cycle. Chip designers qualify substrates together with their packaging partner, and a large ABF qualification can take a year or more. That is why substrate revenue lags capacity.

4.3 Why customers choose Zhen Ding

• Breadth. It can supply flexible, rigid, HDI, SLP and substrates from one supplier, which matters to a phone maker buying many board types per device.
• mSAP yield. In the 1Q26 Q&A, management said only five of thirteen companies that tried SLP for consumer products stayed in production, and that its own mSAP yields are above 95% after more than nine years (Fugle, June 2026). These are company claims, but they explain why it is getting optical orders.
• Capacity in a shortage. It is building more capacity than its rivals while key materials are rationed.
• A non-China option through Thailand and Taiwan.

4.4 Switching costs

These are moderate per model and high per generation. A device maker can move share between qualified suppliers each year, but adding a new supplier means a full qualification cycle and yield risk at launch. In optics and servers, the module or platform qualification ties a board supplier to that generation. Switching mid-generation would mean requalifying a product already shipping to hyperscalers.

4.5 Concentration

The concentration is real and structural. The consumer customer still provides the large majority of revenue. In 2025 smartphones were roughly 61-63% of revenue (Nikhs Substack), and most of Computing & Consumer serves the same customer. That reflects quality, since this customer qualifies only the best suppliers. It also hands the customer pricing power and gives Zhen Ding a seasonal revenue pattern (Section 8.2). The AI push is partly an effort to dilute this concentration. Management has said explicitly that the AI mix should narrow the peak/off-peak gap (MarketScreener / BigGo summary of 1H26).

4.6 Contract structures

• Consumer: annual per-model allocations with no multi-year volume commitments. Revenue visibility runs one cycle ahead.
• AI (servers, optics, substrates): management says most AI orders are under 3-5 year long-term agreements, many with prepayments, with visibility to 2028 (BigGo, 1H26 call; vocus, Oct 2026). Substrate orders are described as committed to 2028 as well (Yahoo, Apr 2026). The terms of these agreements, including volume floors, take-or-pay clauses and price mechanisms, are not disclosed. An LTA in a shortage protects the buyer's supply more reliably than it protects the seller's volume.
• Price pass-through: management says material price increases are being passed on to customers (vocus, Oct 2026).

5. Competitive Landscape and Moat

5.1 Industry structure

The PCB industry is large and fragmented at the bottom, but concentrated at the high end. Prismark's 2025 ranking puts the five largest players at a little over 29% of a roughly US$86bn bare-board market. Zhen Ding is first, followed by Unimicron, Dongshan Precision, Nippon Mektron and Shennan Circuits. The rest of the top ten are TTM, WUS, Victory Giant, Compeq and Tripod (SLL PCB citing Prismark, May 2026). Zhen Ding's share is about 7% of the global market. Chinese-headquartered makers account for about 36% of production value and Taiwanese makers about 30% (same source).

Zhen Ding is not competing in one market but in four sub-markets with different leaders.

Market caps from Pitchbook, Yahoo Finance, StockAnalysis, StockAnalysis, Investing.com and Yahoo Finance; peer-size reference only.

5.2 Where it wins and loses, by arena

Flexible boards (Mobile, Computing & Consumer): wins on scale. Zhen Ding's FPC capacity and yield across many designs per phone make it the default supplier for the hardest parts. Dongshan's MFLEX is the main challenger at the same customer. Each year the buyer can move share between the two, so Zhen Ding's position depends on winning the new, harder designs (foldable hinges, camera flexes) more than defending the old ones. Nippon Mektron, once the reference FPC maker, is now a smaller rival.

Optical-module boards: wins on mSAP. The 3.2T transition (Section 3.1) is narrowing the field to suppliers with established fine-line capability. Industry reporting names Unimicron, Zhen Ding and Compeq, with Tripod and WUS developing similar processes (Cloud News citing Digitimes, Oct 2026). Zhen Ding's advantage over Compeq is scale and capacity that is already funded. Against Unimicron it is roughly even, with Zhen Ding claiming the higher yields from its phone-SLP history. Its target of becoming the largest high-end optical PCB supplier by 2027 is a plan, not a result (Section 7).

AI server boards: a challenger. Victory Giant got into accelerator boards early and is entrenched with the dominant GPU vendor (Wikipedia: Victory Giant). WUS and Gold Circuit are strong in switch and server HLC, and TTM wins where buyers want a non-China, US-domiciled supplier. Zhen Ding's pitch is the iHDI/mSAP fusion for next-generation boards that look more like substrates. It also argues that its capacity is coming on stream as the GPU and ASIC platforms' board requirements outgrow incumbents' capacity. Its server share is small today. Management described the goal as a "jump-style breakthrough" (NextApple, 11 Aug 2026).

ABF substrates: behind. Ibiden and Unimicron hold the qualifications for the largest AI packages, and Shinko and AT&S are established. Zhen Ding's claim of tier-one capability (Section 2.4) has not yet been tested by a large qualification win. Its edge is capacity timing: management describes ABF as in tight supply, and Shen has predicted a "super shortage" of ABF within three years (TechNews, 7 Jul 2026).

5.3 Barriers to entry

• Process know-how. mSAP and large-body ABF are learned by producing at volume and absorbing years of yield loss. The "five of thirteen survived SLP" claim (Section 4.3) is the best available evidence.
• Capital. A single new plant costs billions of Taiwan dollars, equipment lead times are 12-18 months, and a new Chinese plant takes 2-3 years from groundbreaking (3-4 years in Taiwan or Japan) per management (BigGo, 1H26).
• Materials access. In the current shortage, glass-cloth and laminate makers ration supply toward large, established buyers.
• Qualification history. Customers prefer suppliers that have already delivered a generation.

These barriers are high against new entrants. They are weaker against well-funded incumbents moving into adjacent product lines. Chinese PCB makers have abundant capital and government support, and Victory Giant and Dongshan are both raising money in Hong Kong (Bloomberg, 29 Sep 2026).

5.4 Structural shifts

1. Boards are becoming substrate-like. AI boards use mSAP, ultra-low-loss materials and very high layer counts, which blurs the line between PCBs and substrates. This favours companies that already do both, and Zhen Ding is one.
2. Capital is flooding in. Almost every top-ten maker has raised capex. Zhen Ding's plan is the largest in absolute size, but it is not alone (Section 8.1).
3. Geographic split. Western customers want non-China capacity, and Thailand has become the main alternative (Gold Circuit, Zhen Ding and many others), which may tighten Thai labour and supply chains.
4. Optical architecture risk. Co-packaged optics could eventually shift value from pluggable-module boards to substrates (Section 8.6).

Moat: Narrow - scale / cost advantage, process know-how (intangible), and qualification barriers

The flexible-board franchise has kept Zhen Ding on top of global PCB rankings for nine years, through the 2023 consumer-electronics downturn and against a well-funded domestic challenger at its largest customer (Sections 4 and 5.2). Its mSAP yield record is the main reason it is a credible optical supplier now. In AI servers and ABF, though, it is still the challenger, so the "moat" there is only an entry ticket. What would most erode the moat is the single-customer structure (Section 4.5): one buyer that resets allocations every year can move share to a cheaper qualified rival faster than any technology lead can protect it.

6. Industry

6.1 What drives demand

Two engines run on very different clocks.

Consumer electronics (still most of the industry's volume): phone, PC and wearable unit volumes multiplied by board content per device. Units are mature, so growth comes from content: more cameras, thinner devices, foldables and on-device AI that needs more power delivery and denser interconnect. This engine follows replacement cycles and consumer confidence.

AI data-centre infrastructure (the fastest-growing part): hyperscaler and sovereign AI capex flows into GPU and ASIC accelerator boards, switch boards and optical transceivers. Each new GPU generation adds board layers, moves to lower-loss materials and raises link speeds (400G → 800G → 1.6T → 3.2T), which increases the value per board more than unit counts. Substrate demand rises with the size and number of AI packages.

6.2 Size and growth

Prismark estimates the 2025 bare-board market at about US$85.8bn, up roughly 17% from US$73.6bn in 2024, which it attributes mostly to AI servers, high-speed networking and advanced substrates (SLL PCB citing Prismark). Zhen Ding, citing Prismark at its AGM, projected a 7.7% CAGR to about US$123bn by 2030 (I-Connect007, AGM). Other material presented at the 1H26 call suggests the high-end segments could roughly double over the same period (BigGo). The overall figure hides a split: mid-single-digit growth in consumer boards and much faster growth in HLC, HDI for AI, optical boards and ABF.

6.3 Position in the supply chain

PCB makers sit between materials suppliers (glass cloth, copper foil, CCL, ABF film, and chemicals and equipment from Japan, Taiwan and China) and the assemblers (EMS, module makers, OSATs), with brand owners and hyperscalers setting specifications from above. Margins in the chain have traditionally been thin for board makers. In 2026 that has partly reversed: in a materials shortage, the scarcest input (T-glass, high-end CCL, ABF) and the most capable fabs gain pricing power. One Taiwan research note described the shift as PCBs being "promoted to a quasi-semiconductor industry" (TechNews, Jul 2026).

6.4 Import substitution and geography

China makes over a third of global PCB value, and its domestic champions (Victory Giant, Dongshan, Shennan, WUS) are moving up-market with local capital-market support. Western buyers and the US government want non-China supply chains, which pushes capacity to Thailand, Vietnam, Taiwan and Malaysia. That is industry-wide demand for "China+1" capacity. It favours makers who can afford to build twice, and penalises those whose only scale is in China.

6.5 Regulation

There is no product licensing in the PCB industry. The relevant regulatory forces are trade (US tariffs on Chinese-origin goods, and possible Chinese export controls on materials and equipment), environmental permits (plating and chemical discharge, which limit where plants can be built in China), and investment screening for Taiwanese companies investing in China. Customers' own supplier audits and sustainability scoring act as a quasi-regulatory gate.

6.6 Cyclicality

The PCB industry is classically cyclical. It tracks consumer-electronics inventory cycles, overbuilds capacity in upturns and cuts prices in downturns. The 2023 contraction was double-digit. The current AI cycle differs in three ways: demand is driven by capital expenditure rather than consumers, customers are signing multi-year agreements, and materials (not board capacity) are the binding constraint. Overbuilding in this cycle would therefore show up as an air pocket once the shortages ease, probably in 2028-2030 when the plants now under construction come online, rather than as an immediate price war.

6.7 Tailwinds and headwinds

Tailwinds: AI capex intensity; higher data rates in every new generation; ABF scarcity; device-content growth from foldables and AI glasses; customers paying prepayments to secure capacity.

Headwinds: materials shortages (T-glass, high-end CCL); every major board maker raising capex at the same time; possible architecture shifts (co-packaged optics, chip-on-wafer-on-PCB); trade fragmentation; and a consumer-phone market growing only modestly.

7. Growth Triggers

Ranked by potential to change the business. All are management statements of future intent.

• Optical-module boards: from niche to world's largest. Management targets becoming the largest supplier of high-end optical-module PCBs globally by 2027. It expects optical revenue to grow "several fold" in 2026 as 1.6T dominates second-half orders, and says customers are already booking 2027 capacity. (FY2025 call, 12 Mar 2026; 1Q26 call, 12 May 2026; 1H26 call, 11 Aug 2026, repeated on all three; also the 13 Oct 2026 Hong Kong forum.) Next-generation XPO/NPO formats and the 3.2T transition (Section 3.1) are the follow-on. Development is under way, with commercial 3.2T expected around 2028.

• The 2026-2030 plant wave comes online. The company has 13 plants under construction and 16 more in planning, scheduled to open between 2026 and 2030 as customers qualify them. Key milestones: Huai'an HD campus (iHDI/mSAP) first production targeted for March 2027; Thailand Fab 2 into mass production in 2027, with Fabs 3 and 5 and the drilling centre following; new Thai buildings in pilot runs from Q4 2026. Management says 2027-2028 capex will stay at "high levels". (1H26 call, 11 Aug 2026; AGM, 29 May 2026; 13 Oct 2026 forum.)

• AI server boards ramping with next-generation GPU and ASIC platforms. Management guided that AI server revenue would "gradually scale up in 2026 and double in 2027" as customers' next-generation platforms reach mass production. It reiterated that iHDI and HLC products for GPU and ASIC customers are moving into volume, with server revenue rising sequentially through 2026. (3Q25 call, 11 Nov 2025; FY2025 call, 12 Mar 2026; 1H26 call, 11 Aug 2026, repeated.)

• ABF substrate capacity and new international customers. New ABF capacity (the Kaohsiung AI Park, with production from Q4 2026, and the Shenzhen second park, which broke ground in September 2026) is meant to bring in "large international customers" beyond the existing base. Management says ABF capacity will double in 2027, and targeted substrate growth of more than 70% for 2026. (1Q26 call, 12 May 2026; AGM, 29 May 2026; 1H26 call, 11 Aug 2026; 13 Oct 2026 forum.)

• Mix target: AI infrastructure at 45-50% of revenue by 2030. Management's long-range plan is for servers, optics and substrates to make up 45-50% of revenue by 2030, with intermediate steps of about 25% in 2026 and 30% in 2027. (1Q26 call, 12 May 2026; 1H26 call, 11 Aug 2026.) The target was raised from 25-30% in December 2025; that revision is discussed in Section 9.

• Liding's Hong Kong listing. The IC substrate subsidiary filed for a Hong Kong main-board listing on 2 July 2026, with CITIC Securities as sole sponsor. Completion is reportedly targeted for around Q1 2027. Proceeds are intended to fund substrate expansion. (Board approval announced 17 Apr 2026; 1H26 call, 11 Aug 2026.)

• New device categories in the consumer business. Foldable phones, which carry two to three times the board content, AI glasses (where management claims supplier leadership) and AI handsets were named as consumer growth drivers. (1H25 call, 12 Aug 2025; Dec 2025 investor meeting.) Analysts later flagged foldable-launch timing as a pressure (Section 8.2).

• Thailand as the qualified non-China base. Thailand's first fab reached breakeven in July 2026. Management says it will serve AI server, optical and communications customers who want production outside China. (3Q25 call, 11 Nov 2025; FY2025 call, 12 Mar 2026; 1H26 call, 11 Aug 2026.)

8. Key Risks

8.1 The capex bet arrives as the cycle turns

Mechanism. The company is roughly doubling its plant count into 2030 (Section 3.3) on the assumption that AI board demand keeps outrunning supply. Plants started in 2026 open in 2027-2029, when every rival's plants also open (Section 5.4). If AI capex pauses, a platform slips or materials suddenly free up, utilisation falls just as depreciation peaks, and fixed costs in a capital-heavy business hit profit hard.

Calibration. Medium probability, high impact. After the August 2026 call, domestic analysts cut their earnings forecasts for 2026 and 2027, citing higher depreciation from 2027 capacity, higher R&D and delayed foldable launches (Yahoo, Aug 2026). Even in the boom, depreciation is already putting pressure on earnings. Management frames the risk the other way round:

"Our biggest concern is undercapacity, not overcapacity." (Charles Shen, 1H26 call, 11 Aug 2026, as reported by BigGo Finance)

The protection is the multi-year agreements with prepayments (Section 4.6). How much that protection is worth depends on contract terms that are not public.

8.2 The consumer anchor weakens

Mechanism. Most revenue still comes from one consumer customer's devices (Section 4.5). Three paths hurt: a weak phone cycle; foldable and AI-device launches slipping (analysts already flagged a foldable delay in August 2026, Section 8.1); or the customer moving FPC share to Dongshan's MFLEX or another qualified supplier at the next allocation. Because the AI build-out is partly funded by the cash from this business, a weaker consumer year would squeeze funding at the peak of capex.

Calibration. High probability of some year-to-year drag, low probability of a structural loss. The relationship has lasted through many product cycles, but bargaining power sits with the buyer.

8.3 New plants and new products ramp slower than promised

Mechanism. Every growth trigger in Section 7 depends on bringing new plants to yield in new places (Thailand, Kaohsiung, Huai'an HD) on products where the company is new or second (AI server HLC, large-body ABF). The record is already mixed: Kaohsiung's timeline has slipped twice (Section 9). The group can make 1.6T optical boards, but the same yields on new lines in new countries are not guaranteed, and in ABF the yield benchmark is held by Ibiden (Nikhs Substack). Late qualification means depreciation without revenue (8.1), and in a shortage customers give the volume to whoever can deliver.

Calibration. High probability of at least one material delay, moderate impact for each delay.

8.4 Concentrated in China, selling to the West

Mechanism. Most capacity is in mainland China, and the largest new site (the CNY 40bn Huai'an HD campus) is in China too. Direct US exposure is under 1% of sales (Section 2.5), but the end customers are American and their boards end up in products sold in the US. Possible triggers: US tariffs or sourcing rules that reach components by origin rather than shipping destination; hyperscaler procurement mandates for non-China boards; Chinese export controls on materials and equipment; or Taiwan-China tensions. Any of these would leave Chinese capacity under-used while non-China capacity is still ramping.

Calibration. Low to medium probability of an acute event, high impact. Management's own explanation for Thailand and Kaohsiung is geopolitics (Section 2.5), which shows it takes the risk seriously. But China still holds most of the production base.

8.5 Materials, not demand, cap growth

Mechanism. T-glass, high-end low-loss laminates, specialised copper foil and drill bits are rationed. Supply depends on a few producers. Nittobo's capacity additions are a key swing factor (Nikhs Substack). If a plant is finished but the glass cloth is not available, it stands idle. Management says supply is secured through 2027 (Section 3.3). Beyond 2027 it is still negotiating, while competitors are bidding for the same suppliers.

Calibration. Medium probability, moderate impact. It delays revenue rather than destroying it, but it compounds 8.1.

8.6 Technology architecture moves value elsewhere

Mechanism. Value in this industry follows architecture decisions made by chip and system designers. Co-packaged optics could replace part of the pluggable-module market and shift value from module boards to substrates. Chip-on-wafer-on-PCB or glass-core substrates could reshape the board-substrate boundary. Rack-level board designs could change or slip, as shown on 6 July 2026, when a SemiAnalysis report of PCB midplane trouble in the dominant GPU vendor's next rack (Kyber NVL144) knocked more than 5% off Zhen Ding and other PCB stocks. The vendor said its roadmap was unchanged, and Taiwanese board makers said the bottleneck was not on the PCB side (Zero One Research, Jul 2026).

Calibration. Low near-term probability of a broad shift, medium long-term probability. The company is exploring glass-core and CoWoP (Nikhs Substack), but it cannot control the timing.

8.7 Profit and value leak through the group structure

Mechanism. The operating businesses sit in partly-owned listed or listing subsidiaries (Avary in Shenzhen, Liding in Hong Kong). One analysis estimated that over a third of FY2025 group net income went to minority holders (Nikhs Substack), and a Liding IPO would increase that share for the fastest-growing unit. The build-out is being funded partly through equity-linked bonds: a US$400m convertible in September 2025 and a US$800m convertible in September 2026 (BigGo, Sep 2026). Inventories and receivables have grown much faster than sales during the ramp (vocus, Oct 2026). Growth at the operating level can therefore reach the parent's shareholders diluted.

Calibration. High probability, moderate and slow-acting impact.

9. Walk the Talk

Six results events used (from the company's IR and exchange filings, with press and broker write-ups of each call):

1. 1Q25: results and investor meetings, mid-May 2025 (press release 13 May 2025)
2. 1H25 / 2Q25: investor conference, 12 August 2025
3. 3Q25: results and investor conference, 11 November 2025
4. FY2025 / 4Q25: investor conference, 12 March 2026
5. 1Q26: investor conference, 12 May 2026
6. 1H26 / 2Q26: investor conference, 11 August 2026 (69 days before this report)

Two non-results events also supply guidance: the December 2025 investor meeting and the 13 October 2026 Hong Kong forum. Detailed coverage of the 1Q25 event is thin and mostly about tariff exposure. The other five are well covered.

The capex revisions

The amount of money to be spent is the most important number this management gives. Over eight months it kept going up. In late 2025 the guidance was more than NT$30bn a year for 2025 and 2026 (Fugle, Dec 2025). Then, on 12 March 2026:

"The company is in the largest capacity expansion period since its founding; 2026 capital expenditure is expected to exceed NT$50 billion." (Charles Shen, FY2025 call, 12 Mar 2026, translated, Yahoo/MoneyDJ)

Within four weeks, on 7 April 2026, the figure became "over NT$80 billion". Management tied the increase to long-term agreements reached with several key customers on next-generation AI products (Anue, 7 Apr 2026). The allocation also changed. In March the guidance was 35-40% for substrates and 25-30% for servers and optics. By the June meeting it was 65-70% for rigid boards (servers and optics) and only 15-20% for substrates (Section 3.3). Raising capex when customers commit is defensible, and management says demand drove it. But a plan that grows by 60% in a month and changes its mix between product lines was not a well-anchored plan to begin with. Investors should treat capex guidance as a minimum the company expects to exceed, not a ceiling.

Kaohsiung keeps slipping

Kaohsiung was presented in August 2025 as heading for trial production by year-end. In November it became sampling in Q1 2026. By June 2026 it was trial runs in Q3 and production in Q4. No reason for the delays has been disclosed. This matters because Kaohsiung is where the Taiwan-made ABF and advanced boards for non-China customers are supposed to come from. The pattern so far is that dates for plants already built slip, while capex for plants not yet built goes up.

The 2030 target was raised

In December 2025, management's 2030 ambition was for servers, optics and substrates to reach 25-30% of revenue. By May 2026 it was 45-50%. The first half of 2026 supports the direction, since the mix nearly doubled in a year. But the target nearly doubled within five months without any new product launch, and the "AI-related ~70% of revenue" headline used since March 2026 counts AI-enabled phones too. Both are signs of optimistic framing.

What has held up is short-term operating guidance: quarterly seasonality, margin direction and the ABF break-even.

"Some say the PCB industry has a 'golden decade'; I would call it a 'golden age', not limited by ten years." (Charles Shen, 1H26 call, 11 Aug 2026, translated, Yahoo)

Assessment

This management is reliable on the next quarter and optimistic about the next five years. Near-term operating guidance (seasonal peaks, ABF break-even, mix shift) has been met or beaten. Long-range targets and capex have repeatedly been raised, while dates for facilities already under construction have slipped. Investors can rely on short-term operating guidance, should treat capex as a floor, and should discount long-range targets.

10. Ownership, Governance and Shareholder Friendliness

Part 1: Ownership and control

The largest shareholder is Foxconn (Far East) Limited, a wholly owned subsidiary of Hon Hai Precision Industry. It held about 26.5% (roughly 300 million shares) after completing a sale in October 2026, down from 28.5% (305.5 million shares) earlier (Anue, Oct 2026; MarketScreener holdings data). No other single holder of comparable size was identified. The rest is widely held by institutions and the public. Zhen Ding is foreign-incorporated (the "-KY" designation), so shareholder rights depend on its articles and Taiwan's listing rules for foreign issuers, not on Taiwan's Company Act directly.

Below the parent there are two layers of listed or listing subsidiaries: Avary Holding (Shenzhen, 002938.SZ), about 70% held according to a holding disclosure on MarketScreener, and Liding (60.75% controlled before its Hong Kong IPO, Section 2.4). For a minority shareholder in 4958, this structure means two things. First, a large part of operating profit and cash sits in subsidiaries with their own outside shareholders and boards (Section 8.7). Second, the largest shareholder is also a major customer and assembler (Section 4.1), so related-party trade is built into the structure.

Part 2: Governance

Notable flags only:

• Related-party trade with the largest shareholder. Avary's 2025 annual report treats the Hon Hai group as a related party and lists it among its major customers (Avary 2025 annual report summary). The amounts were not located for this report.
• Shared leadership across the group. Chairman Shen Ching-fang (Charles Shen) chairs both Zhen Ding and Avary. The company's board disclosure lists his earlier roles, without dates, as general manager of Yao Wen Electronics, general manager of Hualien Ideal Land, executive vice president of Pacific Securities, vice president of Asia Securities and a section chief at the Export-Import Bank of the Republic of China (board disclosure, via search extract). He also held the president role until it passed on 1 March 2025.
• An independent director became the chief executive. Chien Chen-fu was appointed president (general manager) in early 2025 after serving as an independent director of Zhen Ding for the previous five years and sitting on its management strategy committee. Per the announcement, he was deputy director of industrial engineering at TSMC from 2005 to 2008 and has served as executive vice president of National Tsing Hua University (Anue, 27 Feb 2025). Moving a long-serving independent director into management is unusual and reduces the board's independent bench.
• The largest shareholder sits on the board. Foxconn (Far East) is a corporate director represented by a named individual. The board also includes COO D.J. Lee and the independent directors Yeh Hsin-cheng, Hu Ching-ying and Chen Chun-chung (board listing, via search extract).

Auditor changes, opinion qualifications, insider share pledges and the link between executive pay and measurable targets could not be verified from the filings reviewed.

Part 3: Capital returns

Dividends. Cash dividends per share were NT$3.275 for FY2023 (ex-dividend June 2024), NT$4.80 for FY2024 (ex-dividend 12 June 2025) and NT$3.45 declared for FY2025, adjusted to about NT$3.42 after convertible-bond conversions increased the share count (ex-dividend 8 June 2026, paid mid-July 2026) (Anue; MoneyDJ; Fugle, Mar 2026). The dividend rose and then fell, following earnings, with a policy of paying out about half of profit (the company said the FY2025 payout held at around 50%).

Buybacks and dilution. Recent ~90 days (since 21 July 2026): MoatMap records no buybacks by Zhen Ding. Older, from the three-year search: no share repurchase programme by the listed parent was found for 2023-2026. The buybacks found were by the subsidiary Avary: a programme announced in October 2023 and a further RMB 300-500m programme announced in a later Zhen Ding filing on Avary's behalf, whose date and execution could not be verified (MoneyDJ). These shrink Avary's share count, not 4958's. The parent's share count is rising. Conversions of the September 2025 bond made the dividend-record count about 1.079 billion shares in mid-2026, and Foxconn's post-sale holding implies about 1.13 billion by October 2026. The US$800m bond issued in September 2026 adds potential further dilution. The share count three years ago could not be verified, so a precise three-year net change is not given.

Verdict: Neutral. The company pays out about half of earnings in cash, but it is funding the build-out partly with convertible bonds that keep increasing the share count, and it has bought back no parent shares.

11. Insider Activities

All recent insider activity in the last 12 months comes from one holder: Foxconn (Far East) Limited, the corporate director and roughly 26.5% shareholder (Section 10). MoatMap records three insider sale filings: 5,544,000 shares (0.50% of shares outstanding) on 2026-07-17, 5,544,000 shares (0.50%) on 2026-08-19 and 5,036,000 shares (0.45%) on 2026-10-01 (TWSE insider transfer filings, 2026-07-17, 2026-08-19, 2026-10-01). Hon Hai's own disclosure says the block actually sold was 5,544,000 shares in total, between 1 September and 12 October 2026, at an average of NT$584.96, for about NT$3.24bn and a realised gain of about NT$2.68bn. Hon Hai's board approved the disposal on 17 July 2026 with the stated purpose of "realising investment gains" (Anue, Oct 2026; LTN). The three MoatMap filings therefore look like successive pre-transfer declarations for the same block, renewed as earlier ones lapsed partly unexecuted, rather than three separate sales. The Taiwan press described the completion as the "alarm lifted" on an overhang (LTN). No director or officer bought or sold shares in the window. Read: neutral. This is a disclosed partial profit-taking by a strategic holder, about 2% of its own stake, after a large share-price rise. It does not reflect a view from operating management, and Hon Hai keeps a controlling-size block.

12. Scenarios

Bull case. The 1.6T ramp becomes a multi-year shortage and Zhen Ding comes out of 2027 as the largest optical-board supplier, with first 3.2T qualifications on its mSAP lines. One or two next-generation GPU or ASIC platforms qualify its iHDI boards at volume, so servers become a second large engine instead of a hope. Kaohsiung and the new Shenzhen ABF park bring in a large international substrate customer and the Liding IPO funds the rest without further parent dilution. Thailand fills up with customers who need non-China supply. The consumer business holds steady as foldables and AI glasses arrive late but arrive. By 2028-2029 AI infrastructure is close to half the business, the seasonal swings management has always lived with are much smaller, and the plant wave meets contracted demand rather than empty order books.

Base case. Management delivers most of the AI mix shift, but later and less smoothly than the 2026 slides suggest. Optical boards are the clear success and substrates grow from a small base. Servers arrive in steps as platforms launch, with Victory Giant and others keeping the incumbent positions. As Section 9 shows, plant dates keep slipping by a quarter or two while capex keeps rising, so depreciation runs ahead of revenue through 2027 before utilisation catches up. The consumer business is flat to slightly up, with annual share contests at its main customer. The group ends the period clearly more AI-weighted and less seasonal, with more shares outstanding, more minority leakage and a bigger fixed-cost base than in 2025.

Bear case. Risks 8.1 and 8.3 compound. New plants in Huai'an, Thailand and Kaohsiung open late into a 2028 market where every rival's plants also open, while materials (8.5) ease and remove the shortage that justified the contracts. An architecture change (8.6), such as co-packaged optics gaining share or a rack design slipping, takes value out of the optical-module boards the capex was weighted toward. Meanwhile 8.2 hits the funding base: a weak phone cycle or an allocation loss to MFLEX reduces the cash the build-out relied on. With most capacity still in China, a tariff or sourcing-rule shock (8.4) leaves Chinese plants under-used before Thailand can absorb the work. Shareholders of the parent absorb the result through the leakage and dilution in 8.7.