Delton Technology (Guangzhou) Inc. (1989.HK) - Deep Dive Research Report

Prepared 24 July 2026. Financial magnitudes are deliberately omitted per mandate; the report deals in structure, mix, and mechanism, not valuation.

A note on reporting cadence and "concalls" before we start

Delton is a Chinese company with an unusual listing structure, and getting the reporting picture right up front matters for the rest of this report. The operating entity is 广州广合科技股份有限公司 (Guangzhou Guanghe Technology Co., Ltd.), which has been listed on the Shenzhen Stock Exchange main board as 001389.SZ since April 2024, and which completed a secondary H-share listing in Hong Kong as 1989.HK on 20 March 2026 (offer priced at HK$71.88, raising roughly HK$3.3bn). It is therefore an A+H dual-listed issuer, not a fresh-off-the-boat IPO with no history.

As an A-share issuer, Delton reports on the mainland quarterly cadence (Q1 in late April, half-year in late August, Q3 in late October, annual the following April), with a fiscal year ending 31 December. Chinese A-share issuers do not hold Western-style earnings calls with published English transcripts. What exists instead, and what this report uses as the concall-equivalent evidence base, are the periodic reports, the pre-listing HKEX prospectus (an unusually rich business document), formal investor-relations survey records (投资者关系活动记录表, where dozens of institutions question management on the record), and profit alerts. The six most recent such disclosure events are:

1. H1 2026 profit alert (7 July 2026) - guided net profit up 85.1% to 95.3% YoY
2. Q1 2026 results (29 April 2026) - the most recent full reporting period, within 90 days of today
3. FY2025 annual results (April 2026)
4. Institutional investor survey hosting Goldman Sachs and 17 other institutions (23 December 2025)
5. Q3 2025 results (October 2025)
6. H1 2025 interim results (August 2025), plus the HKEX listing prospectus (March 2026)

Working out the recency explicitly: the most recent formal reporting period is Q1 2026 (released 29 April 2026). The H1 2026 interim report is due by 31 August 2026 and has not yet been released - positively confirmed by the 7 July 2026 profit alert, which pre-announces the half-year range ahead of the audited interim. So the interim is genuinely forthcoming, not merely missing from search.

Section 1: What the company does

Delton makes the printed circuit boards that sit at the heart of computer servers - the large, complex, multi-layer green boards onto which a server's processors, memory, and interconnects are mounted. When a cloud provider or an OEM like Dell builds a rack of AI or general-purpose servers, someone has to fabricate the motherboard, the switch board, the storage backplane, and the accelerator carrier cards. Delton is one of the handful of companies in the world that can fabricate those boards at the layer counts, signal speeds, and reliability that data-centre hardware demands. It does not design the server, own the chips, or assemble the finished machine. It manufactures the substrate everything else bolts onto.

The company was founded in 2002 in Guangzhou. For its first decade and a half it was a general multi-layer PCB maker. The pivotal decision that defines the business today was the deliberate concentration on compute-server PCBs - a shift visible in the numbers: server PCBs rose from roughly 58% of revenue in 2019 to about 69% by 2023, and the broader "compute-scenario" category now sits around 70%. Delton chose to climb the difficulty curve toward high-layer-count, high-speed boards rather than compete in the commoditised low-layer consumer PCB pool, and that choice is why it now ranks, per Frost & Sullivan (2022-2024 revenue basis), as the third-largest compute-server PCB manufacturer globally and the largest China-domiciled one.

Why is this hard? A server motherboard for an AI training node is not the two-layer board inside a toaster. It is an 8-to-20-plus-layer laminate where dozens of copper planes are pressed together with exacting registration, where signal traces carrying 112G+ SerDes must be impedance-controlled to tight tolerances, where the dielectric material (increasingly ultra-low-loss laminates) must survive the thermal cycling of a chip drawing hundreds of watts, and where a single latent defect in one of millions of drilled and plated vias can brick a five-figure server. The know-how is in the process: lamination stack-up control, laser and mechanical drilling, electroplating uniformity, and yield management on boards where the raw material and the customer's downstream value are both high. Qualification by a top-tier server OEM takes years, which is the second reason the field is narrow.

A concrete example of what Delton actually does: an OEM designing a next-generation AI server sends Delton the board design and signal-integrity requirements. Delton's engineering team co-develops the stack-up (how many layers, which laminate, how thick), builds prototypes, submits them for the customer's reliability and signal-integrity qualification, and once approved, ramps to volume fabrication in Guangzhou or, increasingly, Thailand - shipping bare boards to the customer or its EMS partner, who then populates them with silicon. The relationship is engineering-led and multi-year, not a catalogue purchase.

Management frames the strategy as "量产一代、试产一代、研发一代" - mass-producing one generation while pilot-producing the next and developing the one after that. That treadmill is the essence of staying qualified as server architectures turn over.

Section 2: Business segments

Delton is functionally a single-product-family company (PCBs), but its own disclosure organises the book by end-application scenario, and the economics genuinely differ across the three, so they are worth treating as segments.

Compute-scenario PCB (算力场景) - roughly 70% of revenue

This is the core and the growth engine. It covers PCBs for compute servers, split between AI servers (training and inference accelerator platforms) and general-purpose servers, plus the associated data-centre switches, storage backplanes, SSD/HDD boards, and accelerator carrier cards (UBB and OAM module boards for AI clusters). AI-server products now account for over 25% of Delton's server shipments and are the fastest-rising slice.

• Core capability: high-layer-count, high-speed, signal-integrity-critical board fabrication qualified into the AI/HPC supply chain. This is the part of the catalogue that took the longest to build and is hardest to replicate, because it requires both the process control and the multi-year customer qualification.
• Why it stands apart: it faces a different customer set (hyperscalers via OEMs/EMS), different laminate materials (ultra-low-loss), and a different demand cycle (AI capex) from the rest of the book.
• Competitive position: competes directly with Victory Giant, WUS, Shennan Circuits, Unimicron, Tripod, and TTM. Delton wins on being an entrenched, cost-competitive China-based qualified supplier to eight of the world's top ten server makers; it is smaller and less substrate-advanced than the Taiwanese leaders and the AI-PCB darling Victory Giant.
• Role in the group: the margin engine and the entire investment thesis. Management talks about everything else as support.

Industrial-scenario PCB (工业场景)

Boards for 5G communications (AAU modules, RF transceiver units), automotive electronics, industrial control, and security/surveillance. This is the legacy diversification layer - steadier, less explosively growing, and lower-profile in management commentary than compute.

• Core capability: reliability-grade boards for harsher operating environments (automotive thermal/vibration qualification, telecom-grade RF boards).
• Why it stands apart: longer product lifecycles, different certification regimes (automotive), and telecom-capex-driven rather than AI-capex-driven demand.
• Competitive position: a crowded field of Chinese and Taiwanese mid-tier PCB makers; Delton is a participant, not a leader.
• Role in the group: a cash-generative ballast that smooths the compute cycle.

Consumer-scenario PCB (消费场景)

Boards for mini-LED and HD display, 3D smart-printing, and terminal/consumer devices. The smallest and lowest-priority slice, and the one Delton is deliberately de-emphasising relative to compute.

• Core capability: volume manufacturing of display and terminal boards, largely out of the Huangshi facility.
• Why it stands apart: consumer volume economics and a different plant footprint.
• Role in the group: a strategic option and a use for lower-tier capacity, not a focus.

Section 3: Products and business detail

Product catalogue. Within compute, Delton makes server motherboards, data-centre switch boards, HDD storage backplanes, SSD storage boards, optical-module boards, AI switch motherboards, AI UBB boards (the universal baseboards that host multiple AI accelerators), OAM module boards (the OCP Accelerator Module carriers), and stepped-HDI server accelerator cards. Its representative high-end products are described in Chinese disclosure as high-performance-computing server boards, AI-compute server boards, high-performance storage server boards, high-speed switch boards, stepped-HDI server accelerator cards, and 5G communication boards. Within industrial it makes AAU module boards and RF transceiver units for 5G base stations plus automotive boards; within consumer it makes mini-LED boards, HD-display boards, and 3D-smart-printer boards. The unifying technical thread is 8-layer-and-above construction - Delton's book skews heavily to the higher-layer, higher-difficulty end of the PCB spectrum, which is precisely where server demand lives.

What makes them hard. These boards require ultra-low-loss laminate materials, tightly controlled impedance for multi-hundred-gigabit signalling, high-aspect-ratio via drilling and plating, and yield discipline on large-format, high-layer panels. The barrier is not a single machine but the accumulated process recipe and the customer qualification that certifies it. Delton must run the "produce one generation, pilot the next, develop the third" treadmill because server architectures (and the PCB specs they demand) refresh every product cycle.

Manufacturing footprint. Delton runs a multi-plant network with a clear division of labour:

• Guangzhou (headquarters plants): the flagship, positioned at data-centre server, switch, and communications products - the high-value core.
• Thailand (泰国广合): the strategic new node, built specifically to serve the same server/data-centre products from outside China, giving customers a non-China manufacturing option amid supply-chain de-risking. Phase 1 reached production and is in capacity ramp-up through 2026.
• Huangshi (黄石): positioned at intelligent-terminal / consumer products.
• Dongguan (东莞): a machining/support plant supplying the two Guangzhou factories rather than a standalone board fab.

IPO proceeds were earmarked heavily toward this footprint: roughly 52% to upgrading the Guangzhou production base, about 20% to expanding Thailand, 10% to R&D, and the balance to partnerships/acquisitions and working capital. Management flagged a 2026 capacity plan adding more than RMB 2bn of annual sales capacity, with the Thailand Phase 1 ramp alone adding roughly RMB 650m in 2026.

Geographies and milestones. Delton sells to both Chinese server champions and global OEMs, and partners with eight of the world's top ten server manufacturers. The two milestones that reshaped the business are the strategic pivot to server PCBs (mid-2010s onward, lifting server mix from ~58% in 2019 toward ~70%), the April 2024 Shenzhen listing, the Thailand plant reaching production, and the March 2026 Hong Kong listing that funded the next capacity leg.

Section 4: Customers

Who buys. Delton sells directly to two buyer types: end-product server brands/OEMs and EMS (contract manufacturing) providers. Named and reported customers include Dell, Inspur (浪潮信息), Lenovo, Hon Hai/Foxconn, Quanta (广达), and Inventec (英业达) - i.e. both the brand owners and the Taiwanese ODM/EMS houses that build servers for the hyperscalers. Delton states relationships with eight of the top ten global server makers.

Who decides and on what. The buying decision inside these customers is engineering-led: the hardware and supply-chain teams qualify a PCB vendor on signal-integrity performance, reliability/yield, layer-count capability, capacity to ramp, and price. Because a server board sits in the reliability-critical path of an expensive machine, the qualification bar is high and the sales cycle is long (often quarters to years from design engagement to volume). Once a board is designed into a platform and the vendor is qualified, the relationship is sticky for that platform's life.

Why they choose Delton. The specific reasons are: an established qualified position in high-layer server boards, cost-competitiveness relative to Taiwanese incumbents, the ability to co-develop stack-ups, and - increasingly - the Thailand option, which lets multinational customers source server boards from outside mainland China as they de-risk supply chains. The Thailand customer-audit process ("审厂") clearing is itself a selling point management highlights.

Switching costs. Real but platform-specific. Re-qualifying a new PCB vendor for an in-production server board means repeating signal-integrity and reliability testing, which customers avoid mid-lifecycle. The lock-in resets each new server generation, however, so Delton has to re-win designs continually - which is why the "develop the next generation now" treadmill is existential rather than optional.

Concentration. High and worth watching. The top five customers have accounted for roughly 65-70% of main-business revenue across recent reporting periods (disclosed figures around 67.6%, 70.5%, 67.5%, and 65.0%). Multiple of those top-five customers use Delton's AI-server products. This concentration is partly a mark of quality (you cannot be casually qualified into Dell/Inspur/Quanta) and partly a genuine risk: the loss or in-sourcing decision of a single large account would leave a visible hole. Chinese commentary has bluntly described Delton as a "sandwich" - squeezed between powerful upstream laminate suppliers and powerful downstream server customers, with limited pricing power - which is the honest read on what customer concentration means for margins.

Contract structure. Predominantly design-win-driven volume supply against customer purchase orders tied to specific server platforms, rather than multi-year fixed-volume take-or-pay contracts. Revenue predictability therefore tracks the customers' own server build cycles and the AI-capex wave, not a contracted backlog.

Section 5: Competitive landscape

The compute-server PCB business is an oligopoly of a dozen-odd capable fabricators, concentrated in Greater China, and it is being violently reshaped by the AI-capex cycle. Delton sits as the number-three global player by revenue and the leading China-domiciled one, but it is meaningfully smaller than the market-cap leaders, several of which have re-rated enormously on AI-PCB demand.

The real competitors, by segment relevance:

• Victory Giant Technology (Huizhou) - the standout AI-PCB winner, a direct NVIDIA supplier that reportedly vaulted to ~13.8% of global AI/HPC PCB revenue; the most aggressive capacity-expander and Delton's most dangerous direct rival in AI boards.
• WUS Printed Circuit (Kunshan) - a scale leader in high-end server/HPC PCB with strong AI-driven growth.
• Shennan Circuits - a large diversified Chinese PCB/substrate maker with strong HPC exposure.
• Unimicron, Tripod, Gold Circuit, Compeq, Avary Holding (Zhen Ding) - the Taiwanese incumbents; Unimicron in particular is cited at ~18-22% global AI-server PCB share and dominant in some ASIC substrates.
• TTM Technologies - the largest North American PCB maker, strong in aerospace/defence and data-centre.
• Ibiden, Nippon Mektron - Japanese high-end players (Ibiden more in IC substrates).
• Shengyi Electronics, Fastprint, Suntak - other Chinese fabricators overlapping in server/industrial boards.

Where Delton wins: it is a genuinely qualified, cost-competitive, high-layer server-board supplier with an eight-of-top-ten OEM footprint and a differentiated non-China (Thailand) manufacturing option. Where it is exposed: it is a price-taker sandwiched between concentrated laminate suppliers and concentrated server customers, it is materially smaller than Victory Giant/WUS/Shennan, and it is less advanced in the very highest-end substrate/HDI work than the Taiwanese leaders. The AI wave lifts everyone, so Delton's growth is real but not obviously share-gaining against the fastest movers.

Barriers to entry are high but not impregnable: process know-how, multi-year customer qualification, and heavy capex form a real moat against new entrants, which is why the roster is stable. But among the existing incumbents, competition is intense, capacity is being added aggressively across the board (20+ Chinese PCB firms announced record 2026 expansions), and that collective capacity build is the structural risk to pricing.

Market caps are approximate, move sharply (AI-PCB names re-rated violently through 2025-2026), and are shown only as peer-size reference, not applied to Delton.

Section 6: Industry

Demand drivers. Delton's fortunes are tied almost one-for-one to server and data-centre capex, and above all to the AI build-out. Every AI accelerator needs a UBB/OAM carrier board, every server needs a motherboard and switch/storage boards, and the AI generation of these boards uses more layers, more exotic laminate, and higher board area per unit than the general-purpose generation it replaces. So AI demand hits Delton twice: more units and richer content per unit. Secondary drivers are 5G/telecom capex (industrial segment) and automotive electronification (industrial segment).

Size and trajectory. The high-end and AI-server PCB market is one of the fastest-growing pockets of the electronics supply chain, expanding on the back of hyperscaler AI capex that has been running at record levels. Industry trackers (Valuates, TrendForce) describe multi-year double-digit growth for high-end and AI-training/inference server PCB through the late 2020s, and note that 20-plus Chinese PCB firms announced record capacity investment in H1 2026 to chase it - Victory Giant alone lifting Q1 2026 capex nearly fivefold YoY. The demand backdrop is a boom; the supply response is aggressive.

Position in the supply chain. Delton is a mid-stream fabricator. Upstream sit the copper-clad-laminate (CCL) and prepreg material suppliers (a concentrated set - Shengyi, EMC, and the ultra-low-loss laminate makers) whose material cost and availability Delton largely takes as given. Downstream sit the server OEMs and EMS houses, and beyond them the hyperscalers. Delton's "sandwich" position between two concentrated ends is the defining structural feature of its economics.

Import substitution / localisation. A meaningful tailwind. As global customers de-risk away from single-country sourcing and as China pushes domestic substitution up the compute stack, a China-domiciled leader that also offers Thailand-based capacity is well positioned to capture share from both directions. Delton's status as the top China-domiciled compute-server PCB maker is itself a localisation story.

Regulation and certification. The binding "regulation" here is customer qualification rather than government approval - the OEM audit and signal-integrity/reliability certification that gates supply. Automotive boards add formal reliability standards. Geopolitically, US-China tech tensions and potential tariff/supply-chain restrictions are the policy overhang, and are precisely why the Thailand plant exists.

Cyclicality. PCBs are historically cyclical with the electronics cycle, but the current AI phase has overwhelmed the normal cycle with a secular capex wave. The risk is that this wave itself is a cycle - a digestion phase in AI capex, or an over-build of PCB capacity industry-wide, would expose Delton's price-taker position. Telecom and consumer sub-segments remain conventionally cyclical.

Section 7: Growth triggers

Extracted from the periodic disclosures, investor-relations survey records, prospectus, and profit alerts described up front. Delton does not publish English concall transcripts, so each trigger is cited to its dated disclosure.

• Thailand Phase 1 ramp becomes the "second engine" of compute-product growth. Management identifies rising Thailand utilisation, plus key-customer certification and product introduction there, as a direct driver of compute-product sales growth. (H1 2026 profit alert, 7 July 2026)

  The half-year pre-announcement attributes the 85-95% profit surge specifically to "泰国广合产能利用率提升推动算力产品销售增长" - Thailand capacity utilisation lifting compute-product sales.

• Thailand Phase 1 adds ~RMB 650m of capacity in 2026 as it climbs the ramp. (Investor survey / disclosure, reported around 7 July 2026)

• 2026 capacity plan adds more than RMB 2bn of annual sales capacity across the network. (Institutional investor survey hosting Goldman Sachs + 17 institutions, 23 December 2025)

  At that survey management set out a 2026 capacity expansion of "20亿以上" (RMB 2bn-plus) new capacity.

• AI-server products, already >25% of server shipments, positioned as the fastest-growing mix. Rising AI content per board and rising AI-unit share are the structural driver management points to repeatedly. (FY2025 results and prior IR records)

• IPO proceeds fund the next capacity leg - ~52% into Guangzhou upgrades, ~20% into Thailand expansion, ~10% into R&D. (HKEX listing prospectus, March 2026)

• Deepening qualification into eight of the top ten global server makers, with Thailand customer audits ("审厂") clearing to unlock multinational sourcing. (Prospectus and 2026 IR commentary)

Section 8: Key risks

Customer concentration. The top five customers are roughly 65-70% of revenue, and several use Delton's AI-server boards. Mechanism: a single large OEM losing an end-customer, insourcing, or dual-sourcing away from Delton on a major platform would carve a visible hole in the compute segment that industrial/consumer cannot backfill. High-probability moderate drag risk given how normal vendor-share shifts are in this industry; low-probability severe if a top-two account exits.

"Sandwich" pricing power. Delton buys from concentrated laminate suppliers and sells to concentrated, powerful customers. Mechanism: when ultra-low-loss laminate tightens or rises in price, Delton absorbs it; when customers push on price, Delton has limited leverage. Chinese sell-side has explicitly labelled it a margin-squeezed intermediary. High-probability, ongoing margin risk rather than a catastrophe.

AI-capex cycle reversal or digestion. Almost the entire growth story rests on AI/data-centre capex. Mechanism: a hyperscaler capex pause or an AI-training-demand air-pocket would hit compute-scenario volumes hard, and the fastest-growing (AI) slice would decelerate first. Moderate-probability, high-impact - this is the macro tail on the whole thesis.

Industry-wide capacity over-build. More than 20 Chinese PCB firms announced record 2026 expansions; Victory Giant alone raised Q1 2026 capex ~5x. Mechanism: collective capacity coming online into a possible demand plateau compresses pricing across the field, and a price-taker feels it acutely. Moderate-probability, moderate-to-high impact.

Competitive share loss to faster movers. Victory Giant (NVIDIA-direct), WUS, and Shennan are larger and, in some AI pockets, moving faster. Mechanism: if the highest-value NVIDIA/hyperscaler AI-board designs concentrate with rivals, Delton grows with the market but loses relative position and mix quality. Moderate-probability structural risk.

Thailand execution. The second-engine narrative depends on Thailand ramping yield and clearing customer audits on schedule. Mechanism: ramp delays, yield problems, or a customer audit failure would defer the very capacity management is guiding to. Management itself frames Thailand as "in the capacity-climbing phase," which is an implicit acknowledgement that the ramp is not yet complete.

Geopolitics / tariffs. US-China tech tension is the reason Thailand exists; it is also a live risk to the China-based majority of capacity. Mechanism: tariffs or restrictions on China-fabricated boards could force faster, costlier offshoring than planned.

Dual-listing float dynamics. As a newly H-listed, heavily traded name, the H-share free float is dominated by fast institutional and custodian flows (see Section 11), which makes the share register volatile independent of fundamentals.

Section 9: Walk the talk

The six disclosure events used are: H1 2025 interim (Aug 2025), Q3 2025 (Oct 2025), the Goldman-hosted institutional survey (23 Dec 2025), the March 2026 prospectus, Q1 2026 results (29 Apr 2026), and the H1 2026 profit alert (7 Jul 2026). The most recent is 17 days old, comfortably within the 90-day window. The honest caveat: this is a young public-disclosure record - Shenzhen-listed only since April 2024, HK-listed since March 2026 - so the credibility read is based on roughly two years of trackable statements plus the prospectus, not a decade of calls.

Within that record, the pattern is consistent delivery and, if anything, conservatism. Through 2025 management's central promises were three: keep growing the server/compute mix, ramp Thailand into a second growth engine, and add capacity in 2026. Each has tracked or beaten.

On growth and mix, the trajectory management pointed to - compute scenario at ~70% and AI-server share climbing past 25% - has been borne out by the reported acceleration, with FY2025 revenue up roughly 47% and net profit up roughly 50%, and Q1 2026 revenue up 71% with profit up 63%.

On Thailand, the December 2025 survey and prior commentary set up Thailand as the second engine and guided a 2026 capacity build of "RMB 2bn-plus." By July 2026 management could point to Thailand actually in production and ramping, and it explicitly credited Thailand utilisation for the H1 2026 profit surge. That is a guided catalyst converting into a cited earnings driver within roughly two quarters - the cleanest walk-the-talk in the file.

Guided (23 Dec 2025 survey): 2026 capacity expansion of "20亿以上" and Thailand as a second engine. Outcome (7 Jul 2026 alert): H1 net profit guided +85-95%, attributed to "泰国广合产能利用率提升推动算力产品销售增长" - Thailand utilisation driving compute sales.

The most telling credibility signal is the shape of the guidance versus the print. Around the IPO, management guided a strong Q1 2026 profit range (RMB 380-400m, +58-66%); the actual came in near the top and slightly above framing (~RMB 393m, +63%). Then the H1 2026 pre-announcement guided +85-95%, an acceleration off Q1, signalling confidence rather than sandbagging. There is no instance in the available record of a promised catalyst being quietly dropped or a guided range being badly missed.

The fair conclusion: on a short but clean track record, this is management that does what it says and has so far under-promised rather than over-promised. The appropriate discount is not credibility but tenure - there simply is not yet a multi-cycle history, and the whole record has coincided with an AI-capex boom that flatters execution. The test that has not happened is how management guides and delivers through a down-cycle.

Section 10: Shareholder friendliness index

Dividends. Delton pays an annual cash dividend and has done so since its 2024 Shenzhen listing. For FY2025 it proposed RMB 6.46 per 10 shares (RMB 0.646/share), totalling about RMB 305.2m, a payout of roughly 30.05% of parent net profit. The most recently paid prior dividend was about RMB 0.48/share (FY2024). The pattern is a modest, growing, ~30%-payout policy - dividends rising in step with the ~50% earnings growth while the company retains the bulk of profit to fund the heavy Guangzhou and Thailand capacity build. That retention is appropriate for a capex-hungry company riding a demand wave; nobody buys this name for yield.

Buybacks and dilution. There is no active buyback programme. MoatMap's buyback feed records zero repurchases in the trailing ~90 days (since 25 April 2026), and a wider search of A-share and HKEX filings surfaces no buyback authorisation or execution over the last three years for either line. Share count has moved the other way: the March 2026 H-share IPO issued roughly 46m new H-shares, diluting the pre-IPO base - a deliberate, capital-raising dilution to fund expansion, not option-driven creep. So over three years the count is growing, via the IPO issuance, with no offsetting buyback.

Verdict: Neutral, tilting to reinvest-for-growth. Delton pays a steady, growing ~30%-payout dividend but issues shares to fund capacity and does not buy back - it returns some capital while prioritising the capex build, which is the right call for its stage but not shareholder-return-led.

Section 11: Insider activities

Per the instructions for this gated venue (HKEX Disclosure of Interests), the MoatMap disclosure block is the canonical source for recent dealings. The critical interpretive point: every transaction in the last 12 months is by a substantial shareholder that is an institution - fund managers, banks, and a custodian - not by a director or officer of Delton. These are H-share free-float movements crossing/uncrossing the ≥5% (or notifiable) HKEX DI thresholds in a newly listed, heavily traded stock. There are no on-market director/officer purchases or sales in the data, so the single strongest signal (an executive buying his own stock) is simply absent.

All entries HKEX Disclosure of Interests filings, dates as shown.

Buys - reading the signal. The buying is institutional portfolio buying, not insider conviction. Fullgoal Fund (富国基金, a large Chinese mutual-fund manager) is actively trading the position both ways - it bought on 18 Jun, 24 Jun, 6 Jul and 9 Jul, and sold on 2 Jul, 16 Jul, 20 Jul and 21 Jul - which reads as active fund management and threshold-crossing disclosure, not accumulation. JPMorgan's 9 Jun purchase of ~399k shares (~HK$71m) and Fullgoal's net activity are meaningful third-party sponsorship, but none of this is management putting personal money to work. There is no director or officer open-market buy to flag as a bullish signal.

Sells - the why. Fullgoal's sells and UBS's 17 Jul disposal sit inside two-way institutional trading of a stock that ran from a HK$71.88 offer to nearly HK$200 and back toward HK$117 within four months - the sells are profit-taking and position-management in a volatile post-IPO name, not disclosed insider distress; no reason is given in the filings and none is inferable beyond ordinary trading. The large BNY Mellon and Morgan Stanley "Other" entries are almost certainly depositary, custody, securities-lending, or prime-brokerage flows (the "Other" category the block flags as deemed-interest/corporate-action moves), not directional bets.

Net assessment. Activity is broad among institutions and entirely absent among true insiders (directors/officers). The register is dominated by fast money churning a hot, volatile new listing. There is genuine institutional interest (Fullgoal, JPMorgan, UBS, Morgan Stanley all notifiable holders), but the two-way churn nets to noise and the absence of any management purchase means the highest-signal indicator is unavailable. Plain read: neutral - institutional sponsorship is real, but there is no insider-conviction signal to lean on, and the flow is as much about post-IPO trading dynamics as about the business.

Section 12: Scenarios

Bull case. The AI-capex wave keeps running, and the content-per-server story compounds: each new accelerator generation demands more layers, richer laminate, and bigger boards, so Delton's compute revenue grows faster than unit volumes alone. Thailand ramps cleanly through 2026 and beyond, clears every major customer audit, and becomes the de-risked-sourcing plant that multinational OEMs increasingly insist on - turning geopolitics from a threat into a moat and letting Delton win designs its China-only rivals cannot. The 2026 capacity adds fill quickly, AI-server mix pushes well past 25%, and Delton climbs from third-largest toward genuine parity with the AI-PCB leaders while diversifying its top-five customer base enough to blunt the concentration risk. Management keeps under-promising and over-delivering, and the ~30% dividend grows alongside earnings. In this world Delton is a scaled, geographically hedged, AI-levered PCB compounder.

Base case. AI and data-centre demand stay strong but normalise off the euphoric 2026 pace. Delton delivers roughly what it has guided: Thailand ramps on schedule and contributes its planned incremental capacity, the 2026 build lands, compute stays around 70% of the book with AI mix grinding higher, and the company grows nicely but remains a price-taker sandwiched between laminate suppliers and OEMs, so margins are respectable rather than expanding dramatically. Customer concentration persists but no top-five account is lost. Industry-wide capacity additions cap pricing but demand absorbs them. The dividend holds its ~30% payout and grows with profits; no buyback appears. Delton stays the clear China-domiciled leader and a solid number-three globally - executing well, cyclically exposed, not obviously share-gaining against Victory Giant and WUS.

Bear case. AI capex hits a digestion phase just as the 20-plus Chinese PCB firms' record 2026 capacity all comes online, and the field swings from tight to oversupplied. As a price-taker with 65-70% top-five concentration, Delton feels the pricing squeeze immediately and disproportionately. One large OEM dual-sources or insources a key platform, carving a visible hole in compute revenue. Thailand's ramp stumbles - yield or audit delays push out the second-engine capacity the whole guidance rested on - while the geopolitical overhang forces costlier, faster offshoring than planned. The higher-value NVIDIA/hyperscaler AI designs concentrate with faster-moving rivals, leaving Delton growing with a decelerating market but losing relative position and mix quality. The post-IPO institutional float, which trades this stock hard in both directions, amplifies the de-rating. In this world Delton is a solid operator caught in a cyclical down-leg it has no pricing power to defend against, and its short public track record gets its first real stress test.