Suzhou Dongshan Precision Manufacturing Co., Ltd. (002384.SZ)

1. What the company does

Dongshan Precision (the company trades under the English brand "DSBJ") makes the physical guts that go inside electronics. If you strip an iPhone, a Tesla dashboard, a 5G base station, or - most recently - an AI server, down to its constituent parts, Dongshan made several of the components you would find: the flexible ribbon cables that fold inside the phone, the rigid circuit boards that carry the chips, the touch-display glass stacks, the precision-machined metal housings, and now the optical transceivers that move data between AI accelerators as pulses of light.

It is, at its core, a contract precision-manufacturing group. It does not own famous end-products. It is a supplier - a very large one - to the people who do. The thread that runs through everything it makes is the same: high-volume, high-tolerance manufacturing of components where the engineering is hard, the yield is unforgiving, and the customer cannot easily switch.

The founding story explains why the business looks the way it does. In the 1980s, Yuan Fugen ran a sheet-metal and stamping workshop in Dongshan, a town near Suzhou, making electrical-cabinet casings and simple hardware. It was registered as "Dongshan Sheet Metal" (东山钣金) in 1998, and his two sons, Yuan Yonggang and Yuan Yongfeng, joined after university. When China's telecom build-out exploded in the 2000s, the brothers pushed the workshop up the value chain into precision sheet metal and die-casting for base-station equipment, winning their way into the supply chains of Ericsson and Nokia. In 2007 they dropped "sheet metal" from the name - they thought it sounded too low-end - and became "Dongshan Precision." The company listed in Shenzhen in April 2010 making base-station antennas, precision metal parts, and LED display devices.

The decision that made the modern company came in 2016. Dongshan paid roughly USD 610 million for Multi-Fineline Electronix (MFLEX), a US-listed maker of flexible printed circuits and a long-standing Apple supplier. In 2018 it bought Multek from Flex, adding rigid and high-layer-count boards. In two acquisitions a metal-bashing telecom-parts firm transformed into one of the largest circuit-board makers on earth and, through MFLEX, a deeply embedded member of Apple's supply chain. The most recent transformation, completed in late 2025, is the same playbook again: it bought Source Photonics, a vertically integrated optical-transceiver and optical-chip maker, to plug straight into the AI data-center build-out.

The company describes its current strategy as building "the dual core-hardware engine of AI compute: optical modules (including optical chips) plus AI PCB" - pairing its existing high-end circuit-board capability with Source Photonics' laser-chip capability.

The value proposition to a customer is consistency at scale. Apple does not award flexible-circuit volume to a vendor that cannot yield millions of defect-free, dimensionally identical parts on a schedule tied to a product launch. The thing Dongshan sells, more than any single product, is the ability to industrialize a hard component - to take something that is difficult to make once and make it ten million times reliably.

2. Business segments

Dongshan runs four reportable activities. Three are legacy (electronic circuits, optoelectronic display, precision manufacturing); the fourth, optical modules, arrived with the Source Photonics deal and was only consolidated from the fourth quarter of 2025, so its reported revenue share understates its true run-rate scale.

Electronic circuits (PCB and FPC) - roughly two-thirds of revenue

This is the core. The segment makes flexible printed circuits (FPC), rigid printed circuit boards (PCB), and the high-layer-count, high-density boards used in advanced electronics. It is built on the MFLEX (flexible) and Multek (rigid) acquisitions, which together gave Dongshan coverage from soft boards to hard boards in one group. By revenue, it ranks among the top handful of circuit-board makers globally, and is consistently cited as second or third worldwide in flexible boards.

The core capability is volume manufacturing of flexible circuits at consumer-electronics yields. FPC is genuinely hard: the substrate is a thin polyimide film, the copper traces are fine-pitch, the boards must survive being folded and flexed inside a device for years, and the qualification bar at a customer like Apple is brutal. That qualification - getting designed into a product, passing reliability testing, ramping yield - took MFLEX years to earn and is the moat. The newer, strategically loud part of this segment is "AI PCB": the high-speed, high-layer-count boards that carry AI accelerators and switches. Management has repositioned this capability as half of its AI compute story.

It exists as a distinct segment because the technology, customers, and capital intensity are different from metal work or displays - and because it was bought wholesale as two foreign businesses with their own factories and customer relationships. Competitively it wins on flexible-circuit scale and its Apple relationship; it loses, or at least bleeds margin, in commodity rigid boards where Chinese and Taiwanese capacity is abundant. Group-level commentary for 2025 noted that the traditional circuit base is under margin pressure even as it grows, which tells you this is the cash-and-scale engine rather than the margin engine.

Optical modules (Source Photonics) - the growth bet

This segment designs and makes optical transceivers - the devices that convert electrical signals into light and back, letting data move between servers, switches, and accelerators inside data centers over fiber. Source Photonics, founded in 2010, runs an IDM/vertically integrated model: it designs and fabricates its own optical chips (lasers, EML/electro-absorption-modulated lasers), builds the optical sub-assemblies, and assembles and tests the finished modules. Its product range spans 10G all the way to 800G and 1.6T.

The core capability that justified the roughly RMB 5.9 billion (USD ~687 million) purchase is in-house optical-chip design. Most module makers buy their laser chips from a small group of suppliers (notably the Western incumbents). Source Photonics is one of the few that designs its own, which means it controls cost, supply, and the technology roadmap rather than depending on a vendor that is also a competitor. It has shipped 800G modules in volume and has 1.6T modules in validation with the major AI customers.

This is unambiguously the growth bet and, despite contributing only a few percent of 2025 revenue (one quarter of consolidation), management has stated it already accounts for more than half of consolidated profit by Q1 2026 because its margins are far above the group average. Its competitors are the Chinese module champions (Innolight, Eoptolink) and the Western photonics incumbents (Coherent, Lumentum); see Section 5. Within the group it is the margin engine and the strategic story being sold to the market and to Hong Kong IPO investors.

Optoelectronic display - roughly a fifth of revenue

This segment covers LED display devices and touch-display modules. The LED business makes small-pitch and Mini/Micro-LED packaged devices (down to fine pitches like P0.5-P1.5) used in indoor and outdoor high-definition display walls, plus Mini-LED backlight units; it is described as a global top-three LED display-device maker. The touch-display business makes touch panels and display modules and has moved into in-vehicle screens.

The capability is fine-pitch LED packaging and display-module integration - the IMD, MiniCOB, and MiP packaging routes that let many tiny LEDs be placed and bonded at display resolutions. It exists separately because it is a different technology stack (optics and packaging, not circuit etching or laser fabrication) and serves different end markets (display OEMs, signage, automotive interiors). Competitively it sits among the leaders in fine-pitch LED but in a segment where Chinese LED capacity is plentiful and pricing competitive; LED display-device revenue has at times shrunk. Within the group it is a mature, scale-driven contributor rather than a growth driver.

Precision manufacturing (metal components and automotive) - roughly a tenth of revenue

This is the company's origin - precision sheet metal, die-cast and machined metal structural parts and assemblies - now redirected at two end markets: communications infrastructure (base-station and equipment housings) and, increasingly, new-energy vehicles (structural and functional metal parts, thermal components). The 2025 acquisition of France's GMD Group (an automotive parts maker, roughly EUR 100 million including debt) added European automotive manufacturing and customers and deepened the auto push.

The capability is high-tolerance metal forming and assembly at automotive and telecom quality standards - the discipline the Yuan family built the company on. It exists as its own segment because the inputs (metal, casting, machining) and the qualification regimes (automotive IATF-style quality) differ from electronics. Competitively it is one supplier among many in a fragmented metal-parts world; it wins on its existing relationships (it can sell metal parts to the same Apple and Tesla accounts it already serves with boards). Within the group it is the legacy cash cow being repurposed as an automotive growth option via Tesla and GMD.

3. Products and business detail

The product catalogue spans four families. In circuits: single- and multi-layer flexible printed circuits, rigid-flex boards, high-density interconnect (HDI) rigid boards, high-layer-count and high-speed boards for servers, and the substrate-like advanced boards grouped under the "AI PCB" banner. In optical: optical transceiver modules across 10G, 25G, 100G, 200G, 400G, 800G and 1.6T speeds, plus the underlying optical chips - EML and other lasers - that Source Photonics fabricates itself, and 100-GBaud EMLs that enable 200 Gbps-per-lane signaling for 1.6T. In display: small-pitch LED packaged devices using IMD, MiniCOB and MiP routes (1010, 0606, 0404 device families), Mini-LED backlight units, touch panels, display modules, and in-vehicle screens. In precision manufacturing: machined and die-cast metal structural parts, base-station housings, and automotive structural and functional components.

What makes these products hard is concentrated in two places. Flexible circuits demand process control over thin-film handling, fine-line etching, lamination and reliability under repeated flexing, all at consumer-launch volumes and yields - the kind of capability that requires years of qualification at a customer before a single board ships. Optical chips demand semiconductor-grade epitaxy and fabrication of compound-semiconductor lasers, then precision alignment of optics to fiber within microns; the in-house chip capability is rare and is the reason Source Photonics was worth buying rather than building.

Manufacturing is concentrated in China (Suzhou, Yancheng and other mainland sites) for circuits, displays and metal parts, with the MFLEX/Multek heritage giving it advanced board capacity. Source Photonics adds optical-chip fabrication and module assembly, and the company is building additional module capacity in Thailand (targeting tens of millions of modules a year) to serve customers wanting non-China supply. GMD adds European automotive plants. The company sells globally: its circuits go into US, European and Asian device brands; its optical modules go to the global AI hyperscaler and networking customers; its display and metal parts serve a mix of domestic and export accounts.

The milestones that reshaped the business are acquisitions rather than organic launches: the 2016 MFLEX purchase (entry into flexible circuits and Apple), the 2018 Multek purchase (rigid boards), the 2023 acquisition of Suzhou Jingduan to push into EV components, the 2025 GMD acquisition (European automotive), and the 2025 Source Photonics acquisition (optical, the pivot to AI compute). The November 2025 Hong Kong H-share filing - an "A+H" dual listing to fund capacity expansion and repay acquisition debt - is the financing milestone that underwrites the next phase.

4. Customers

The customer base is concentrated at the top and broad below it. The single largest customer is Apple, reached primarily through the flexible-circuit business inherited from MFLEX. Disclosure around the Hong Kong listing and prior filings shows the largest customer accounting for more than half of revenue - the top customer was around 50% of revenue in the first nine months of 2024, and Apple's share had been reported above 55% in 2023. The second material relationship is Tesla, served across multiple product lines (circuits including FPC, in-car displays, and functional metal structures) as the company's new-energy-vehicle anchor. Source Photonics brings a different roster - AI hyperscalers and networking customers, with Meta historically a major Source Photonics account and Nvidia involved in validating 1.6T modules.

Inside these customers the buying decision sits with sourcing and supplier-quality engineering teams, not a single procurement manager. For a part like a flexible circuit in a phone, the decision is made years ahead during the product design cycle: the component is designed in, the supplier is audited and qualified, sample lots are reliability-tested, and only then does volume flow. The selection criteria are yield, capacity to meet a launch ramp, quality history, price, and increasingly geographic flexibility. The sales cycle is long for design-ins and then highly repetitive once qualified.

Customers choose Dongshan for proven high-volume yield, an existing qualified position, and the ability to supply several component types to the same account. The switching costs are real: requalifying a new flexible-circuit or optical-module supplier means repeating audits and reliability testing and risking a launch, so an incumbent in good standing tends to keep the socket. This is why concentration is best read as two-sided - it reflects deep, sticky qualification at the world's most demanding electronics buyer, and it is simultaneously the company's single largest risk (Section 8).

Contract structure is the consumer-electronics norm: framework supply agreements with volumes driven by the customer's product forecasts and purchase orders, repricing pressure each cycle, and revenue that is seasonally tied to flagship product launches (heavy in the second half for the Apple-linked circuit business). The optical business is earlier-stage and order-driven, ramping with hyperscaler capex; management has framed optical demand as visible from customer order indications rather than smooth contracts.

5. Competitive landscape

Dongshan competes in three distinct arenas, and its position differs sharply across them.

In circuit boards, it is one of the larger players globally in flexible circuits but sits in a crowded field. Direct comparables include Zhen Ding Technology (the world's largest flexible/PCB maker, Apple's lead FPC supplier), Nippon Mektron/Mektec, Taiwan's Unimicron and Compeq, and Chinese rigid-board specialists such as Shennan Circuits and WUS Printed Circuit. Dongshan wins on flexible-circuit scale and its Apple position; it is exposed in commodity rigid boards where capacity is abundant and pricing thin. The newer "AI PCB" sub-arena pits it against high-speed-board specialists who have moved fastest into AI server boards.

In optical modules, Source Photonics is a mid-tier global vendor (commonly ranked around 8th-10th worldwide) competing against the two Chinese champions - Innolight and Eoptolink - who dominate the high-speed 800G/1.6T sockets at the hyperscalers, and against the Western photonics incumbents Coherent and Lumentum. Source Photonics' edge is its in-house optical-chip (EML) design, which few peers have; its weakness is scale and customer share relative to Innolight, which holds the majority of Nvidia-linked volume. A structural dynamic worth noting: the Western incumbents have at times refused to sell their best laser chips to the Chinese module leaders, which makes in-house chip capability like Source Photonics' strategically valuable.

In LED display and precision metal, competition is fragmented and largely Chinese, pricing is competitive, and there is no strong moat; these are scale-and-relationship businesses.

(Market caps are approximate peer-size references that move daily; Chinese photonics names in particular have run hard in 2026.)

Barriers to entry differ by arena. In flexible circuits and optical chips they are high - qualification, process know-how, and capital. In rigid boards, LED, and metal parts they are modest, which is why those segments carry thinner margins. The structural shift shaping the whole landscape is the AI-driven surge in demand for both high-speed PCB and optical interconnect, which is pulling Dongshan, Innolight, Eoptolink, Shennan and WUS up-market simultaneously and rewarding whoever can add capacity fastest.

6. Industry

Demand for Dongshan's products is driven by three cycles that no longer move together. The legacy cycle is the smartphone, where the flexible-circuit business rises and falls with flagship launches and the broader handset market - a mature, cyclical, slow-growth driver. The second is the automotive/EV cycle, where rising electronic content per vehicle and the EV transition drive demand for circuits, displays and metal structures. The third, and the one the market now cares about, is the AI data-center build-out, which drives demand for both high-speed PCB and optical interconnect.

That third cycle is the growth story. The global PCB market is sizeable - roughly USD 80-100 billion depending on the research house (Grand View, Mordor, Fortune Business Insights) - and growing in the low-to-mid single digits, but the data-center/computing slice is the fastest-growing part, projected around 8% CAGR. The optical-transceiver market is smaller but growing far faster: LightCounting and others put the AI-cluster optical-interconnect market in the mid-USD-20-billions for 2026, up sharply year on year, with 2026 widely treated as the first year of 1.6T commercialization (millions of units) and 800G already shipping in the tens of millions. High-speed 800G/1.6T modules alone are a multi-billion-dollar annual market and supply has at times run behind demand.

Dongshan sits in the middle of the supply chain - a Tier-1/Tier-2 component maker selling to brand OEMs and hyperscalers, not to end consumers. In optical chips it is part of a Chinese push toward import substitution: China has historically depended on Western laser-chip suppliers, and in-house Chinese optical-chip capability (Source Photonics, plus domestic chip makers) is a deliberate move to localize a strategically constrained input. That same dynamic creates geopolitical friction - Source Photonics had Taiwan/US ties, and its acquisition by a mainland firm drew tech-security scrutiny.

Regulation that matters: automotive quality regimes for the metal/auto business; the customer-qualification regimes that function like de facto regulation in circuits and optical; and, increasingly, cross-border tech-transfer and export controls that shape where optical capacity can be built and sold (hence the Thailand module capacity). Cyclicality is real - consumer electronics is cyclical and the AI capex cycle, while powerful now, is itself a capex cycle that can turn. The current tailwind is the AI interconnect boom; the headwinds are commoditization in the legacy segments and the risk that hyperscaler capex normalizes.

7. Growth triggers

All points below are drawn from the company's investor-relations activity records and results briefings (the A-share equivalent of earnings calls). Dates given are the record dates.

• Optical-chip and module capacity expansion confirmed and ordered. Management said the aggressive expansion plan is now executing, equipment orders are placed, and equipment is expected to be in place by Q3 2027, with capacity targeted at 10 billion optical chips by 2028 and 20 billion by 2029 (after moving from 3-inch to 4-inch wafers). (FY2025 + Q1 2026 briefing, Apr 28-29 2026; reiterated in the May 2026 IR record.)

Management framed the ambition bluntly: "the world's largest laser company in the future will be Chinese," citing China's roughly 6-month factory-approval timeline versus 12-16 months overseas. (May 2026 IR record.)

• 400G EML mass production targeted by end-2026. The company said it aims to reach 400G EML mass production by the end of 2026 with laser performance matching the industry leaders, and to hold among the largest laser capacity globally. (May 2026 IR record.)

• 1.6T optical modules ramping into volume. Management said 800G EML modules are already in volume delivery and 1.6T modules have passed validation with major AI customers including Nvidia, with 1.6T mass production expected around Q4 2026. (FY2025 + Q1 2026 briefing, Apr 28-29 2026; repeated theme.)

• Thailand module capacity build-out. The company said it is accelerating module capacity in Thailand, targeting roughly 30 million modules a year, to serve customers wanting supply outside China. (FY2025 + Q1 2026 briefing, Apr 28-29 2026.)

• Laser/optical-chip demand visibility through 2029. Management said customer order indications point to optical-chip demand growth of roughly 100% CAGR over 2027-2029, framing it as based on customer orders rather than analyst models. (May 2026 IR record.)

• Hong Kong H-share listing to fund expansion. The company said the planned H-share listing will fund new and upgraded facilities, M&A, and repayment of acquisition-related debt; the prospectus was first filed November 2025 and refiled May 2026. (Q3 2025 briefing, Oct 22 2025; refiling May 2026.)

• Source Photonics + AI PCB "dual engine" for AI compute. Management repeatedly positioned the combination of its AI PCB capability with Source Photonics' optical-chip capability as the company's core AI-compute hardware platform, with Source Photonics already driving the majority of group profit. (Q3 2025 briefing Oct 22 2025; FY2025 + Q1 2026 briefing Apr 28-29 2026 - repeated.)

• GMD-driven automotive expansion in Europe. Management pointed to the completed GMD France acquisition as the platform to grow the automotive/EV structural-parts business and broaden the European customer base. (FY2025 + Q1 2026 briefing, Apr 28-29 2026.)

8. Key risks

Customer concentration in Apple. This is the dominant risk. With the largest customer at more than half of revenue, any loss of share at Apple, an Apple unit-volume decline, or pricing pressure flows almost directly into Dongshan's results. The mechanism is simple: a single design-out, a shift of flexible-circuit allocation to Zhen Ding or another supplier, or a weak iPhone cycle compresses the largest revenue line at once. This is a high-probability moderate-to-severe drag risk that is structural, not hypothetical - the concentration was flagged as a concern in coverage of the Hong Kong listing.

Acquisition debt and finance costs. The pivot to optical was bought, not built, and the bill is large. Short-term borrowings rose sharply through 2025 and finance costs exploded in Q1 2026 (reported up many-fold year on year), a direct result of debt taken on for Source Photonics and GMD. The mechanism: if the optical ramp slows or interest costs stay high, the acquired growth has to outrun a heavy financing burden, and the balance sheet has less room for error. The H-share listing is partly a response to this; if it is delayed or raises less than hoped, the debt pressure persists.

Goodwill and integration risk. Three foreign acquisitions in a decade (MFLEX, Multek, GMD) plus Source Photonics leave large goodwill on the balance sheet. If any acquired business underperforms - particularly Source Photonics, which now carries the profit story - impairment is a risk, and integrating a US/Taiwan-rooted optical IDM into a Chinese group carries operational and talent-retention risk.

Geopolitical and tech-security friction around optical. Source Photonics had Taiwan/US ties, and its purchase by a mainland company drew tech-security scrutiny. The mechanism: export controls or customer reluctance could restrict where optical chips and modules can be made or sold, which is precisely why the company is building module capacity in Thailand. An escalation could cap Source Photonics' access to Western hyperscaler sockets.

AI capex cyclicality. The entire optical growth case rests on hyperscaler AI spending continuing to rise. Management itself frames demand off customer order indications - which is real visibility, but order books can be cut. If the AI capex cycle normalizes, the high-margin segment that now carries group profit decelerates fastest.

Margin pressure in the legacy base. The traditional circuit, LED and metal businesses are in competitive, capacity-rich markets; group commentary for 2025 noted the circuit base growing while its margin slipped. These segments cannot be relied on to grow profit; they are scale and cash, and they can drag blended margins.

9. Walk the talk

The six investor-relations activity records used here, in order: the July 2024 specific-object survey (record dated 2024-07-17); the FY2024 + Q1 2025 results briefing (Apr 25 2025); the H1 2025 specific-object survey (Sep 16 2025); the Q3 2025 results briefing (Oct 22 2025); the FY2025 + Q1 2026 results briefing (Apr 28-29 2026); and the late-May 2026 IR record. The most recent sits within ~90 days of this report. These are A-share IR records and results briefings rather than English earnings calls, but they are the comparable management-disclosure events.

The through-line of management credibility here is that the Yuan brothers do what they say on the big strategic moves, and the big moves have historically worked - while the near-term financial guidance has lately run behind what the market modeled. Through 2024 the story management told was the "consumer electronics plus new-energy-vehicle dual drive": stabilize the Apple-linked circuit base and grow automotive through Tesla and bolt-on deals. That promise was substantially kept - the circuit base held, the auto push continued, and the GMD acquisition in 2025 delivered the European automotive platform they had been signaling.

The defining test is the optical pivot. Across the Q3 2025 and April 2026 briefings, management committed to consolidating Source Photonics, making the "AI PCB plus optical" dual engine the core of the company, and driving optical to a major share of profit. By Q1 2026 that promise had been kept in a way few component-maker pivots are: Source Photonics, consolidated from Q4 2025, was already contributing more than half of group profit. The capacity and product commitments - 800G in volume, 1.6T validated with major AI customers - have so far tracked what management said.

On the optical roadmap, management said 800G EML modules are already in volume delivery and 1.6T has passed validation with major AI customers. (Apr 2026 briefing.)

The outcome - Source Photonics flipping to the dominant profit contributor within two quarters of consolidation - is consistent with that statement, which is a point in management's favor.

Where management has run ahead of itself is on near-term financials and the cost of the pivot. The FY2025 results, while up year on year, came in below sell-side expectations on both revenue and profit, and the same period revealed the scale of the borrowing and finance-cost increase taken on for the acquisitions. Management had emphasized the growth and the strategic logic far more than the leverage it required; the debt and finance-cost surge was the part the optimistic framing under-weighted. This is a recognizable pattern for this management: they are bold, acquisitive operators who consistently deliver the strategic transformation they promise, but who tend to present the upside more vividly than the balance-sheet cost, leaving the market to discover the financing strain after the fact.

The honest read: management that does what it says on strategy and M&A, with a strong long-run record of buying its way into better businesses, but whose near-term financial framing skews optimistic and under-discloses the leverage cost of its ambitions.

10. Shareholder friendliness index

Dongshan pays a small cash dividend but is not run for income; capital is being plowed into acquisitions and capacity, not returned. The company has maintained a modest dividend over the last three years (it has a multi-year history of paying), but the payout is small relative to earnings and is dwarfed by the cash directed at the MFLEX-legacy expansion, GMD, Source Photonics, and the optical-capacity build-out. The trend has been a maintained-to-modestly-growing token dividend rather than a rising income stream, consistent with a company prioritizing reinvestment in a capex-heavy growth phase.

On share count, the direction over three years has been dilution, not retirement. Rather than buying back stock, the company has issued it: a private placement (the RMB ~1.5 billion raise approved in 2024 and completed in mid-2025) added roughly 125 million new shares, and the planned Hong Kong H-share listing will add a further new class of equity. There is also a 2026 employee stock ownership plan (up to ~3.05 million shares). I found no evidence of a meaningful share-repurchase program over the last three years; based on the MoatMap database scope note, the absence of a recent buyback only speaks to the last ~90 days, but the three-year web search likewise surfaced equity issuance, not repurchases. The notable nuance is that the controlling Yuan brothers personally subscribed to the 2025 placement (largely with bank financing), so the dilution to outside holders was paired with insiders putting money in - a conviction signal even as the float grew.

Verdict: Hoards/reinvests capital - the company issues equity and reinvests aggressively for growth rather than returning cash, with only a token dividend and no buyback.

11. Insider activities

A note on sourcing: China A-share insider and major-shareholder disclosures are filed through the Shenzhen Stock Exchange and CNINFO and aggregated by Eastmoney. No automated MoatMap insider block was supplied for this company, so the transactions below are reconstructed from company announcements and reputable financial-news coverage rather than a single regulator feed; treat the list as the material, publicly reported events of the last ~12 months rather than an exhaustive line-by-line register.

The defining insider event of the period is the controllers buying in, not selling. In mid-2025 the company completed a private placement (向特定对象发行) raising roughly RMB 1.5 billion, and the two actual controllers, chairman Yuan Yonggang and his brother Yuan Yongfeng, subscribed personally for approximately 125 million of the new shares - reportedly funding well over RMB 900 million of it with bank loans. Two controlling insiders committing well over a billion renminbi of personal and borrowed capital to new equity, rather than trimming, is the strongest signal in this section. This is a very bullish signal: cluster buying by both controlling brothers, debt-funded, at the moment they were pivoting the company into optical.

On the buy side, the placement subscription is unambiguous: both controlling brothers increased their personal stakes at the same time, in size, using borrowed money - the cluster, the size, and the leverage all point to conviction. Their combined family stake stood at roughly 33% at end-2025 (Yuan Yonggang ~16.5%, Yuan Yongfeng ~13.5%, father Yuan Fugen ~3.2%), up from prior years partly because of this subscription.

On the sell side, there is important context but no material selling in the past 12 months. The Yuan family were heavy sellers in an earlier era - roughly 166 million shares trimmed between 2014 and 2020 for around RMB 3 billion, which halved the founding stake over a decade - but that is outside the 12-month window and has been replaced recently by net buying. The November 2025 pledge release is sometimes misread as a transaction; it is the unwinding of share-pledge collateral, which de-risks the family's position rather than selling shares. Reasons for the historical selling (diversification and the chairman's pre-existing large pledge ratio) are documented, but none of it falls in the assessment window.

Net assessment: over the last 12 months insiders are net buyers, and the buying is concentrated in the two people who matter most - both controlling brothers, in size, with leverage, at a strategic inflection point. There is no reported material insider selling in the window, and the family stake has risen. This is a bullish insider signal, tempered only by the reminder that the same family were aggressive sellers in the 2014-2020 era, so the recent buying represents a genuine reversal of their long-run trimming behavior.

12. Scenarios

Bull case. The AI interconnect boom runs for years, and Dongshan turns out to have bought the right asset at the right time. Source Photonics' in-house optical-chip capability lets it ramp 800G in volume, win 1.6T sockets at Nvidia and the hyperscalers through 2026-2027, and bring 400G EML and the 10-billion-then-20-billion optical-chip capacity online roughly on the schedule management laid out, with the Thailand plant giving customers the non-China supply they want. The high-margin optical business grows from half of profit toward the dominant share of a much larger profit pool, the Hong Kong listing comes through and refinances the acquisition debt, and the finance-cost overhang fades. Meanwhile the legacy circuit base stays anchored at Apple, the AI PCB line rides the same data-center wave, and GMD turns the metal business into a credible automotive growth story with Tesla and European OEMs. Dongshan re-rates from "Apple component maker" to "AI-compute hardware supplier," and the optimistic capacity bets prove conservative against demand that outruns supply.

Base case. Management delivers roughly what it guided, with friction. Source Photonics keeps growing and remains the profit engine, 800G stays in volume and 1.6T ramps, but capacity additions and yield improvements come a little slower and a little more expensively than the boldest framing implied. The Hong Kong listing completes and takes some pressure off the balance sheet, though finance costs remain a visible drag for a while. The Apple-linked circuit base does what it always does - grows modestly, with margin pressure - and the auto and display segments contribute steadily without surprising. The company is genuinely transformed, optical is genuinely the new center of gravity, but the near-term numbers keep coming in respectable rather than spectacular, much as the FY2025 result landed below the most optimistic models.

Bear case. The leverage taken on for the pivot becomes the story. AI capex normalizes or hyperscalers cut optical orders just as Dongshan's huge new optical-chip capacity comes online, leaving expensive, debt-financed fabs underutilized and finance costs grinding against a profit base that suddenly stops growing. Geopolitical friction restricts Source Photonics' access to Western hyperscaler sockets, and the Thailand workaround proves insufficient. The Hong Kong listing is delayed or undersized, so the debt stays. Simultaneously the concentration risk bites at home: an Apple cycle weakens or flexible-circuit share shifts to Zhen Ding, hitting more than half of revenue at once. Goodwill from four acquisitions gets tested, and the same management whose boldness built the company finds that this time the acquisition was timed into a peak rather than a runway.