DSBJ (002384.SZ) Initiation Coverage: AI Twin Engines Delivering, a Race Between Expectation Discount and Supply Cliff

Rating: Cautiously Bullish | Target Price Range: RMB 115–158 (fair under base case) | Current Price: RMB 114.22 (2026-10-23 close) | Margin of Safety: +0.7% (vs. lower bound of base-case fair value) | Time Horizon: 12 months | Probability-weighted expected price RMB 140.1 (+22.7%)

Market cap approximately RMB 209.2 billion (1831.6 million total A-shares × frozen anchor); the H-share listing on 2026-10-28 will dilute share capital by 6.2%–8.2% (dilution sensitivity covered in the valuation section). All valuations, implied expectations, and odds in this report are anchored to the 2026-10-23 A-share closing price of RMB 114.22.

I. Core Conclusions Summary

DSBJ's AI twin engines (Source Photonics optical modules + Multek AI PCB) delivered their first wave of earnings elasticity in the 2026H1 financials: revenue of RMB 27.80 billion (+63.95%), net profit attributable to shareholders of RMB 2.957 billion (+290.09%), non-recurring-adjusted (recurring) profit of RMB 2.480 billion (+277.52%), of which roughly 32 percentage points of revenue growth came from Source Photonics consolidation — real delivery, but with a consolidation flavor; the AI PCB engine has yet to ramp (releasing from 2026Q3). The current price of RMB 114.22 has retraced 58.6% from the 52-week high of RMB 275.8; the market has already priced in a clear discount to the sell-side consensus 2027 net profit of RMB 18.83 billion (at most multiple tiers, the current price implies RMB 8.4–17.4 billion), while the H-share issuance at a ~35% discount (HKD 86.80, listing 10-28) creates near-term relative-value pressure. This report's base case: 2027 net profit attributable of RMB 10.5–13.5 billion × 20–21.5x exit PE = fair value range of RMB 115–158; the current price sits at the lower edge of the range — assessed as fairly valued to slightly cheap, cautiously bullish. The key constraint is the risk of a price war after the concentrated release of optical module supply in 2027 (bear case RMB 45–60, midpoint -54%) and its knock-on interaction with impairment of the RMB 4.683 billion goodwill. The Q3 report on 10-31 (whether optical module gross margin holds at 38%) and North American cloud providers' 2027 capex guidance in January–February 2027 are the most important validation points over the next two quarters.

II. Investment Opportunity Assessment

Opportunity in one sentence: The profit inflection of the AI optical module + AI PCB twin engines has already shown up in the financials (2026H1 attributable net profit +290%, optical module gross margin 39.33%); the current price's implied pricing of 2027 earnings is 7%–56% below sell-side consensus (depending on exit multiple assumptions) — an opportunity of "delivering but at a discount," not a deep mispricing.

Evidence grade: B+ — performance and segment data are audited/disclosed financial facts, but capacity utilization and 1.6T customer certification, on which growth persistence depends, are management planning figures with no public order-book evidence; the AI PCB engine of the "twin engines" has not yet shown up in the interim report.

Market environment: Tailwinds with signs of a shift to headwinds — the AI compute sector is in its 3rd year of an upcycle (2026 combined capex guidance from the four major cloud providers ~USD 730 billion, +77%), but A-share AI hardware broadly corrected in October (the stock fell over 26% in October), and the discounted H-share issuance further compressed risk appetite; Source Photonics rose from 10th place globally in 2024 to 7th in 2025 (CIC/LightCounting basis) — a rising share trend, though scale remains small (~3%–4%).

Most important sources of change over the next 6–12 months (ranked by impact): ① 2027 optical module long-term contract price negotiations and capacity utilization — determining whether 2027 attributable profit lands in the RMB 5.5–7.0 billion (bear) or RMB 10.5–13.5 billion (base) tier; ② North American cloud providers' 2027 capex guidance (initial indications in early 2026-11, formal in January–February 2027); ③ AI PCB Phase 1 capacity ramp and North American customer certification; ④ execution of H-share IPO proceeds and A/H price convergence after listing.

Primary stock price driver: Optical module volume/price and Source Photonics' earnings midpoint (Q3 optical module revenue QoQ and gross margin are the nearest falsifiable indicators).

Verifiable expectation gap: Market anchor = sell-side consensus 2027E attributable net profit of RMB 18.83 billion (median of 12, ratings 26/26 Buy); this report's base case RMB 10.5–13.5 billion; back-solved from the current price, implied RMB 8.4–17.4 billion (at 12–25x tiers). The expectation gap runs both ways: 7%–56% below consensus, and the consensus sample is crowded and bullish (median target price RMB 234.4 from only 5 firms, published near the June sentiment peak) — convergence could come via a higher stock price, or via a downward revision of consensus estimates.

Validation catalysts: K1 Q3 report (10-31), K2 H-share listing (10-28), K4 optical chip capacity reaching 22 million units/month in 2026Q4, K5 1.6T volume shipments, K7 cloud providers' 2027 capex guidance. Falsification conditions: optical module gross margin falling below 38%, cloud providers' 2027 capex growth guidance <20%, goodwill impairment provision for Source Photonics, substantive US restrictions on optical modules taking effect.

III. Investment Thesis

C1 — The 2026H1 earnings surge is real but carries consolidation flavor: ~32 percentage points of the +63.95% revenue growth come from Source Photonics consolidation; optical modules are the only engine already delivered; AI PCB awaits validation from Q3 onward (confidence 0.80)

• Source / DSBJ 2026 Semi-Annual Report (2026-08-22), Section II Key Accounting Data: "Operating revenue (RMB) 27,797,683,972.88 vs 16,955,163,898.89, +63.95%"; "Net profit attributable to shareholders of the listed company 2,956,898,811.04 vs 758,005,980.61, +290.09%"; "Net profit after non-recurring items 2,479,581,522.26 vs 656,799,943.31, +277.52%"
• Source / Same report, revenue composition: optical module revenue RMB 5,350,999,299.05, gross margin 39.33%; electronic circuits revenue +13.29% YoY, gross margin 18.80%

All reported figures are verifiable, but two structural facts must be clarified upfront. First, Source Photonics was only consolidated as of September 30, 2025 (97.48% stake), so the optical module line in the interim report's product breakdown has no prior-year comparative base — the 2025H1 revenue of RMB 16.955 billion contained no optical modules. Excluding optical modules, 2026H1 revenue grew roughly +32.4% YoY (which still includes consolidation increments from GMD precision components; pure organic growth is even lower). Second, the flip side of +290.09% attributable growth is a low base: 2025H1 attributable net profit was only RMB 758 million (2025Q2 alone only ~RMB 300 million). Q2 standalone attributable profit of RMB 1.847 billion (+511.30%) looks striking, but on a recurring basis, Q2 vs Q1 was RMB 1.420 billion vs RMB 1.059 billion (+34%) — a real acceleration, but with a slope far shallower than headline YoY. In addition, H1 non-recurring gains of RMB 477 million accounted for 16.1% of attributable profit, mainly unrealized gains on securities investments (Lianxun Instruments at cost RMB 20 million, book value RMB 482 million at period-end) — not sustainable. The AI PCB engine has not yet been delivered in the interim report: the electronic circuits segment grew +13.29% in revenue with 18.80% gross margin (+1.25pct YoY); per the company, Multek Phase 1 capacity only starts releasing in 2026Q3. If Q3 optical module revenue keeps growing double digits QoQ and electronic circuits gross margin improves sequentially, the "twin engines" narrative completes its second piece; conversely, if optical module gross margin falls below 38%, this thesis is falsified.

C2 — The current price's implied pricing of 2027 earnings is below sell-side consensus, but the "discount magnitude" is highly sensitive to exit multiples, and the gap may close via downward revision of consensus rather than a higher stock price (confidence 0.50)

• Source / 10jqka earnings forecast page (2026-10-25 snapshot): 2026E EPS RMB 4.13 (median of 13, implying RMB 7.56 billion attributable), 2027E EPS RMB 10.28 (12 analysts, implying RMB 18.83 billion attributable), ratings 26 Buy/0 Hold/0 Sell, median target price RMB 234.4 (5 analysts)
• Source / This report's back-solved estimate: market cap RMB 209.21 billion ÷ 12–25x exit PE = implied 2027E attributable net profit RMB 8.37–17.43 billion, 7.4%–55.6% below consensus of RMB 18.83 billion

The back-solved implied expectation is extremely sensitive to the multiple tier: if the market assigns 25x to the optical module supply chain in 2027, the current price implies net profit of only RMB 8.37 billion (56% below consensus); at 12x, it implies RMB 17.43 billion (only 7% below). This report does not presume which tier the market will use, but only points out that at all 12–25x tiers the implied value is below consensus — the market is not pricing the sell-side's 2027 narrative. The factual chain supporting a higher delivery midpoint: the four major cloud providers' 2026 combined capex guidance is ~USD 730 billion (+77%); Morgan Stanley forecasts ~USD 1.2 trillion for the five major cloud providers in 2027; LightCounting expects 1.6T output value to grow at a 57.6% CAGR over 2026–2030; Source Photonics' optical chip monthly capacity is ramping from 9 million units in 2026Q2 toward 22 million in Q4, targeting 300 million units/year in 2027. But three links in this transmission chain are unverified: Morgan Stanley notes over half of 2027 capex flows to memory (optical module unit density may not rise in lockstep), Source Photonics' share gains (10th → 7th → ?) are a trend extrapolation rather than order evidence, and capacity targets are all management plans. The downside shape of this thesis: consensus 2027E ranges from RMB 7.0–29.3 billion — extremely wide dispersion (the low end derived from Guosheng's 7.5/24/40 billion RMB series at the 2026 tier; the high end from Citi's RMB 350 target price implying RMB 29.3 billion). If optical module gross margin weakens in 2026Q3–Q4, consensus itself could be revised down from RMB 18.8 billion to the RMB 12–15 billion tier, at which point the "discount" disappears without the stock necessarily falling — the expectation gap closes via a shrinking denominator.

C3 — Source Photonics' IDM positioning is a supply-chain bargaining chip, not a proven cost advantage: capable of mass-producing 100G/200G EML (rare domestically), but gross margin is below that of peers outsourcing chips (confidence 0.50)

• Source / DSBJ 2026 Semi-Annual Report, Section III Core Competitiveness: "The company is one of the few domestic companies with mass-production capability for 100G and 200G EML optical chips" and "one of the few companies globally with an integrated EML optical chip + optical module layout"
• Source / Wallstreetcn (2026-08, Eoptolink 2026H1 results): Eoptolink gross margin 48.44% (+1.01pct YoY); DSBJ 2026 interim report: optical module gross margin 39.33%

The company's own wording is "one of the few domestically/globally" — it should not be inflated into specific numbers: overseas, Lumentum, Mitsubishi Electric, Sumitomo Electric, and Coherent all develop EML in-house with some vertical integration, and domestically, Yuanjie Semiconductor and Accelink are also competing in 200G EML — Source Photonics sits in the second or third tier of EML in third-party rankings (CLS Gold Minutes Database, 2026-03). Another point to clarify: the company states that "part of the optical chips used in the optical module business are self-developed and self-produced, accounting for no more than 10% of the overall optical module BOM cost" (2026-10-08 IR record) — this is a self-sufficiency ratio (self-developed portion is under 10% of BOM), whereas on an industry basis optical chips account for 40%–60% of optical module cost (21st Century Business Herald, 2026-06), meaning the bulk of Source Photonics' chip costs still must be outsourced. The real value of IDM lies in supply assurance and certification leverage amid the industry-wide EML shortage (demand exceeds supply by ~30%, LightCounting), not in cost: 2026H1 optical module gross margin of 39.33% is below Eoptolink's 48.44% and Innolight's 46.59% (10jqka basis), both of which outsource DSP/chips — scale and product mix are the main reasons. Only if self-sufficiency in self-developed 200G EML rises in the 1.6T generation with a smooth yield ramp can IDM turn from "bargaining chip" into "advantage"; conversely, if CPO/silicon photonics locks in the mainstream ecosystem at the 3.2T generation, EML assets could become a burden of technological transition.

C4 — The H-share issuance at HKD 86.80 (a discount) creates short-term relative-value pressure and 6.2%–8.2% dilution, but the ~HKD 9.88–13.06 billion raised provides equity ammunition for the USD 1.7 billion expansion plan (confidence 0.70)

• Source / DSBJ Announcement on the Offer Price of Overseas Listed Shares (H Shares) (2026-10-26): "final price of HKD 86.80 per share"; "base offering of 113,783,200 shares (~HKD 9.876 billion), and up to 150,478,100 shares (~HKD 13.061 billion) if the option is exercised in full"; "expected listing on the Main Board of the Hong Kong Stock Exchange on October 28, 2026"
• Source / Eastmoney (2026-10-20): ~64.5% of proceeds for building new facilities and upgrading production lines, ~20% for potential strategic investments or acquisitions, ~5.5% for debt repayment; 20 cornerstone investors (including Alibaba's Aurora SF and Montage Technology) subscribing USD 680 million

At the CFETS rate of 0.8502, HKD 86.80 is approximately RMB 73.80, a ~35.4% discount to the A-share frozen anchor of RMB 114.22; Hong Kong margin subscription was only ~0.39x (final allotment per the 10-27 announcement), with the issuance mainly absorbed by cornerstone investors and international placement — cornerstones accounted for over half of the base offering, showing offshore institutions would only take it at a discount. Relative-value pressure is already materializing: on 10-26 (pricing announcement day), the A-share fell as much as -8.9% intraday to RMB 104.05. But viewed over a longer horizon, this fundraising is necessary: as of 2026-06-30, the company had interest-bearing debt of RMB 23.963 billion, a debt-to-asset ratio of 64.51%, and H1 free cash flow of -RMB 2.869 billion, while the total investment in optical chip and high-speed optical module expansion has been raised from USD 1.2 billion to USD 1.7 billion (~RMB 12.1 billion) — roughly RMB 5.4–7.2 billion of net H-share proceeds will go into capacity, swapping equity for part of the debt financing and marginally improving the financing structure (direct debt repayment is only ~5.5%; a funding gap remains relative to the RMB 12.1 billion expansion plan, and refinancing in 2027 cannot be ruled out). Dilution math: base offering dilutes 6.2%, fully exercised 8.2%; if restated on post-dilution share capital and crediting ~RMB 8.4–11.1 billion of raised cash (fully exercised, roughly +RMB 5.6 per share), the base-case fair range of RMB 108–149 + cash contribution ≈ RMB 114–154, broadly consistent with the RMB 115–158 on the A-share anchor basis. Watch points: whether the H-shares break below the issue price on debut 10-28, fast entry into the Hang Seng Composite Index and Stock Connect arrangements around the 10th trading day after listing, and the expiry of cornerstone investors' 6-month lock-up at end-April 2027.

C5 — Looser supply and falling prices in 2027 are the main downside risk: once the shortage that persists through 2027 eases, double ordering disappears and price cuts accelerate, interacting with goodwill impairment (confidence 0.50)

• Source / LightCounting April 2026 market forecast (#379): the optical module shortage will persist through 2027 and market growth will exceed expectations; but "once suppliers add capacity and the shortage eases, double ordering will disappear and price declines will accelerate" (a conditional warning)
• Source / DSBJ 2026 Semi-Annual Report, Note 27 Goodwill: Source Photonics goodwill net of RMB 2.712 billion (original value RMB 2.799 billion, RMB 87 million already provided); Mudong Optoelectronics RMB 154 million and Suzhou Aifu RMB 115 million goodwill fully impaired (precedent)

The industry evidence chain is accumulating: 800G average prices have fallen from ~USD 480 in 2025 to USD 350–380 as of May 2026; the whole industry is expanding capacity simultaneously (Innolight, Eoptolink, Source Photonics' own USD 1.7 billion, and on the AI PCB side Multek USD 1.0 billion/WUS Thailand/Shengyi Jiangxi), and LightCounting expects price declines to accelerate once the shortage eases. The offsetting facts that must also be presented: industry profitability is still rising — Eoptolink's 2026H1 gross margin was 48.44% (+1.01pct YoY), Innolight's optical module main business 46.59% (up from 39.96% in 2025H1), and LightCounting called 2026Q1 the best quarter ever for the optical module market — volume growth is partially offsetting price declines. Quantified risk elasticity (this report's stress test, not company guidance): if 800G/1.6T price cuts widen to -30% in 2027 and volume growth provides insufficient offset, Source Photonics' net profit has a path of -50% to -85% (net margin squeezed from ~23% to 4%–11%); combined with goodwill impairment testing (Source Photonics acquisition consideration of RMB 4.663 billion vs identifiable net assets of RMB 1.838 billion, with the premium dependent on earnings forecasts being met), a neutral provision of RMB 1.2 billion or a full provision of RMB 2.71 billion would push 2027 attributable profit down to the RMB 2.7–3.9 billion tier (bear case), corresponding to this report's bear-case fair value of RMB 45–60 (-61% to -47%). The goodwill impairment test disclosure in the 2026 annual report (April 2027) and the outcome of 2027 long-term contract negotiations are the validation points for this thesis.

IV. Key Financial Data

4.1 Core Metrics Table (consolidated, RMB 100 millions)

Note: 2026Q1 standalone — revenue RMB 13.138 billion (+52.72%), attributable net profit RMB 1.110 billion (+143.47%), recurring profit RMB 1.059 billion (+166.99%), gross margin 19.33%.

Reasons for metric changes (company explanations provided for all YoY ≥ ±20%): Revenue +63.95% — company explanation: in addition to the Source Photonics/GMD consolidation effects, continued onboarding of new customers and new products in the optical module business; attributable profit +290.09% — Source Photonics consolidation + optical module revenue ramp, plus fair value gains of RMB 491 million (unrealized securities gains); financial expenses +1747.19% (RMB 546 million) — increased financing scale and interest expense (interest expense RMB 257 million) from acquiring Source Photonics and GMD and expanding optical module and AI PCB capacity, plus increased FX losses from currency movements (RMB 303 million); construction in progress +92.4% (RMB 4.514 billion) — active capacity expansion in optical modules (incl. optical chips) and AI PCB; inventory +27.88% (RMB 11.418 billion) — proactive stocking amid AI demand; CapEx +144.5% — same expansion logic.

4.2 Latest Results Quick Review (2026H1)

Revenue: RMB 27.80bn, +63.95% YoY, of which optical modules contributed RMB 5.351bn (19.25% of total, purely a consolidation increment with no prior-year base); excluding optical modules, growth was approximately +32.4% YoY (still including GMD consolidation contribution). By quarter: Q1 RMB 13.138bn → Q2 RMB 14.660bn (+75.51% YoY), +11.6% QoQ. Profit: net profit attributable to parent of RMB 2.957bn fell within the July preliminary results range (RMB 2.9–3.0bn); non-GAAP (recurring) profit was RMB 2.480bn, with Q2 recurring profit of RMB 1.420bn, +34% QoQ vs Q1 — the acceleration is real, but the slope is far smaller than the headline +511% (the 2025Q2 base was only about RMB 0.30bn). Earnings quality: non-recurring items of RMB 0.477bn accounted for 16.1% of net profit (mark-to-market gains of RMB 0.491bn on securities investments in Lixun Instruments, at a cost of RMB 0.020bn); gross margin was 20.15% (+6.06pp YoY), driven mainly by mix — optical modules at 39.33% × 19.25% revenue share lifted the blend, while traditional electronic circuits gross margin was 18.80%, up only +1.25pp. Cash flow: OCF of RMB 2.384bn (-4.65%) diverged from net profit, due to AI-related inventory build (inventory of RMB 11.418bn, +27.9%) and receivables expansion; the cash collection ratio was still 98.28%; FCF was -RMB 2.869bn, consumed by CapEx of RMB 5.253bn — a typical expansion-phase profile, but it needs quarterly re-verification. Versus market expectations: at the current price, 27.7x the 2026E consensus net profit of RMB 7.56bn, implying H2 net profit of about RMB 4.6bn, +55% QoQ under a peak-season assumption; management gave no quantitative guidance ("please refer to the periodic reports for H2 operating data"), so delivery hinges on optical modules running at full tilt in Q3/Q4 and the ramp of AI PCB Phase I.

Intraday alert (2026-10-26): the A-shares fell to RMB 104.05 intraday, at one point -8.9% below the 10-23 close of RMB 114.22 (media reported the decline narrowed to about -5% by press time). The trigger was the H-share offering price announcement the same day (HKD 86.80, equivalent to about RMB 73.80, an approximately 35.4% discount to the A-shares), combined with margin subscription of only about 0.39x — a weak offshore demand signal. This is a financing-discount price-comparison effect, not new fundamental negative news; the valuation anchor remains based on the 10-23 close. The share price may continue to fluctuate with the H-share listing debut ahead of the Q3 report on 10-31.

V. Business Model and Earnings Quality

Business model snapshot: an asset-heavy electronics manufacturing platform with four businesses — electronic circuits (No.2 globally in FPC plus high-multilayer/HDI rigid boards; project-based orders, mainly consumer electronics), optical modules (Source Photonics; an IDM with partial in-house EML chip capabilities, AI data-center customers), precision metal structural components (Tier-1 NEV supplier; GMD supplements European capacity), and optoelectronic display (automotive/touch; low-margin mature business). All revenue is project-based orders with no subscription/recurring revenue; weak bargaining power with Apple (annual price-down mechanism), improved bargaining power with North American cloud providers during shortages. In 2025 the strategy pivoted within one year from "consumer electronics + new energy" to "optical modules + AI PCB," driven by external M&A (Source Photonics for RMB 4.663bn plus GMD via debt assumption) and heavy capex (2026H1 CapEx of RMB 5.253bn).

Cash content of earnings: excellent long-term, deteriorating currently — OCF/net profit attributable was 2.63 / 4.59 / 3.83 in 2023–2025 (top-tier among consumer electronics manufacturers, cash collection ratio persistently 98%+), but plunged to 0.81 in 2026H1 (inventory build + expansion); FCF/net profit was 0.87 / 1.10 / 0.67 / -0.97 (turned negative in 2026H1) — a large chunk of book profit is being consumed by expansionary capex. This is not a signal of "fake profits" (the cash collection ratio is still 98.28%), but a classic feature of a capacity arms-race phase: total investment in optical module + AI PCB expansion is about USD 1.7bn + USD 1.0bn, with capex intensity (CapEx/depreciation of 1.68–1.73) entirely growth-related; maintenance capex was not separately disclosed and cannot be split out — if demand falls short in 2027, the return on this capex will be directly exposed.

Recurring earnings test: 2025 net profit of RMB 1.386bn vs recurring profit of RMB 0.966bn — non-recurring items of RMB 0.420bn were 30.3% of net profit (mainly a RMB 0.471bn negative-goodwill gain from the GMD acquisition); core earnings were actually RMB 0.966bn, up only +7.5% YoY. In 2026H1 the ratio fell to 16.1% (securities mark-to-market gains). There is no company-reported Non-GAAP under A-share standards; recurring profit is the yardstick used in this report.

Return on capital: ROIC was about 3.4% in 2025 (dragged down by acquisition loans), recovering to an annualized ~12.3% in 2026H1 — still below the 15% "true moat" threshold, and achieved with the tailwind of the AI shortage boom (39.33% gross margin); if a price war materializes in 2027, ROIC decline pressure will directly test the incremental return on this round of investment of RMB 12.1bn-scale. WACC is about 10% (self-built: risk-free 2.0% + β 1.25 × equity risk premium 5.5%, debt weight 6%) — the current ROIC–WACC spread is positive but thin; the moat verdict is "unverified" rather than "established."

Red flags: ① Largest customer accounts for 46.46% of revenue (51.28% in 2024); top five account for 64.33% — a dual large-customer structure combining Apple dependence and North American cloud concentration; ② Annual dividends suspended for two consecutive years (zero dividend in 2025; the company explained cumulative M&A outlays of RMB 6.749bn exceeded that year's net profit); ③ Controlling shareholder Yuan Yonggang has pledged about 35% of his holdings (company-wide pledge ratio 8.27%, as of 2026-09-18); ④ Nearly one year after the Source Photonics equity transfer agreement, multi-jurisdiction overseas registration remains incomplete (title procedure defect risk).

VI. Management Assessment

Consistency of words and deeds: partially delivered, direction unstable. Three verifiable promise–delivery pairs: ① the 2024 annual report promised "optimize LED operations" → in 2025 LED was no longer separately disclosed and impairments did not expand (delivered); ② the 2025 annual report planned "optical chip capacity release, 800G/1.6T volume shipments, AI PCB new bases in production, advancing A+H" → in 2026H1 optical module revenue was RMB 5.351bn (3.73x the first consolidation quarter's RMB 1.436bn), 1.6T already shipping, H-shares listed on 10-28 (being delivered); ③ however, the "consumer electronics + new energy dual drivers" proposed in the 2024 annual report was switched to "optical modules + AI PCB" within a year — strong execution, weak strategic commitment; the pivot relied on M&A rather than organic evolution. The company consistently provides no quantitative guidance (IR explicitly declined to give H2 figures in Sept 2026), so no guidance-versus-delivery reconciliation is possible.

Shareholder friendliness: favors reinvestment, unfriendly in the current period. Payout ratio 21.65% (2023) → 10.94% (2024) → 0 (2025, triggering the no-dividend clause of the three-year return plan); buybacks are signal-level — the July 2026 plan of RMB 0.2–0.3bn (cap of RMB 367.04/share) had accumulated only about RMB 50mn by early October, less than 0.25% of market cap; refinancing is intensive — a RMB 1.39bn private placement in June 2025 (controlling shareholder fully subscribed 126mn shares, locked until 2028-06-28) and an H-share issue diluting 6.2%–8.2%. Cumulative effect: two years of M&A and expansion outlays far exceeded retained profit, and cash returns to shareholders are suspended. Verdict: neutral-to-"growth first"; the controlling shareholder's placement participation and H-share cornerstone investors (Alibaba, Montage, and other industrial capital) partially offset governance concerns.

Risk signals: the controlling shareholder father-and-son duo hold about 33% combined; Yuan Yonggang has pledged about 35% of his holdings; responses to market rumors (65% US-origin content, FCC restrictions, large orders) have been restrained, with no record of exaggerated promises (positive); on the reported Source Photonics overseas shipping damage (rumored at the USD 27mn level), the company says it was fully insured with minimal impact.

VII. Segment Breakdown

Profit engine has shifted: in 2025, electronic circuits contributed about 80% of group gross profit (63.85% × 17.59%); in 2026H1 it became dual engines of electronic circuits ~42% + optical modules ~38% — with only 19.25% of revenue, optical modules generated gross profit comparable to electronic circuits; the third-largest segment by revenue is now tied for first in profit. The gross margin spread is as wide as 33pp (optical modules 39.33% vs optoelectronic display 5.89%): the former benefits from AI data-center supply shortage during the EML shortage plus IDM structure, the latter is a fully competitive, mature assembly business — if an optical module price war materializes in 2027, these two ends will converge toward the middle, pressuring the group's gross margin center. The +178.63% headline surge in precision components is mainly GMD consolidation (European capacity) + new base ramp-up; organic contribution was not separately disclosed; its gross margin improvement to 12.13% (+3.86pp) is consistent with scale effects. Optoelectronic display continues to shrink (share 14.92% → 9.80%) and is a clearly identified non-core divestiture target for management (a RMB 0.595bn impairment plus RMB 0.164bn disposal loss was taken in 2024).

VIII. Accounting Practices and Cross-Period Consistency

Accounting red flags (by severity):

1. Excessive share of non-recurring items (medium): 30.3% of 2025 net profit — the bargain-purchase gain of RMB 0.471bn recognized on the low-price GMD acquisition (mainly debt assumption) boosted net profit by about 34% one-off; in 2025Q4 alone, recurring profit was -RMB 0.105bn while net profit was +RMB 0.163bn — the core business was actually loss-making/barely breakeven in Q4, masked by the acquisition gain (2025 annual report quarterly data).
2. Impairments concentrated at earnings troughs, big-bath characteristics (medium): the 2024 net profit decline of -44.7% was mainly due to the one-off LED cleanup (asset impairment of RMB 0.595bn + disposal losses of RMB 0.164bn); excluding these, 2024 net profit was about -5% YoY (company's own basis) — technically "bundling bad news into a bad year," but post-cleanup LED did not drag again in 2025, and the impairment quality looks credible in hindsight.
3. Securities mark-to-market gains boosting profit (low): 2026H1 fair-value gains of RMB 0.491bn (Lixun Instruments: RMB 0.020bn cost → RMB 0.482bn book value), market-dependent and unsustainable; 16.6% of H1 net profit.
4. Year-end inventory build causing stockpiling surge (low): 2025 electronic circuits inventory volume +197.83% (company explanation: Q4 stocking for Q1 sales); 2026H1 inventory of RMB 11.418bn (+27.9%), with RMB 0.198bn of asset impairments already taken in H1 — if AI-related stocking meets demand volatility, inventory write-downs are a subsequent watch point.
5. Goodwill impairment testing as a key audit matter (low, but fat tail): goodwill of RMB 4.683bn is 19.07% of net assets; Source Photonics' RMB 2.712bn corresponds to an acquisition consideration of RMB 4.663bn vs identifiable net assets of RMB 1.838bn; goodwill for Mudong Optoelectronics and AI Fu Electronics has historically been fully written off, showing impairment testing is not a formality — this is both positive (discipline) and a reminder (if Source Photonics misses profit forecasts, the write-down will not be gentle).

Cross-period consistency: ① the jump in gross margin 15.18% → 14.02% → 14.09% → 20.15% is driven by consolidation and mix shift; the traditional business actually declined slightly (electronic circuits 18.34% → 17.59% → 18.80%), consistent with management's explanation; ② the sharp divergence between net profit and recurring profit (19.65/10.86/13.86/29.57 vs 16.15/8.99/9.66/24.80) is fully attributable to itemized one-offs, consistent; ③ OCF/net profit plunged from 3.83 to 0.81 — the interim report did not state the reason for the -4.65% OCF change (unexplained by the company), a disclosure gap this report believes requires quarterly tracking; ④ customer concentration 71.04% → 64.33% (marginally down) but still extremely high in absolute terms, and likely to rise again as optical modules ramp, consistent; ⑤ payout ratio 21.65% → 10.94% → 0 moves in sync with M&A funding needs, consistent.

IX. Valuation and Odds

9.1 Current Market Data and Multiples

Current price RMB 114.22 (2026-10-23 close), A-share market cap RMB 209.21bn. PE(TTM) on a net profit basis is 58.4x / 75.0x on a recurring basis — but the TTM earnings structure is mid-transition (Source Photonics was only consolidated in 2025Q4); the historical percentile is contaminated by the old structure and is reference-only, not a headline; PB 8.59x (87th percentile of the past 5 years, but 2026H1 half-year ROE was already 12.90%, annualized ~23%–25%; the high PB is supported by the ROE leap thesis); PS(TTM) 4.1x. Forward multiples are the main anchor: 2026E consensus net profit of RMB 7.56bn implies 27.7x; 2027E consensus of RMB 18.83bn implies 11.1x; PEG (2026E PE ÷ 2026→2027 consensus CAGR of 57.6%) ≈ 0.48.

Peer comparison table (snapshot as of 2026-10-25):

9.2 What Expectations Are Already Priced In

Back-solving: market cap RMB 209.21bn ÷ exit multiple tier = implied 2027E net profit — 12x tier: RMB 17.43bn (-7.4% vs consensus); 15x tier: RMB 13.95bn (-25.9%); 18x tier: RMB 11.62bn (-38.3%); 20x tier: RMB 10.46bn (-44.5%); 25x tier: RMB 8.37bn (-55.6%). In one sentence: the current price requires the company to deliver only RMB 8.4–17.4bn in 2027 (corresponding to a conservative-to-neutral optical module capacity absorption rate of roughly 16%–40%), while sell-side consensus is RMB 18.83bn. Whether this discount is justified is tested bidirectionally with demand-chain math: on the demand side, the four major hyperscalers' 2026 capex is about USD 730bn (+77%) → 2027E about USD 1.2 trillion (Morgan Stanley); LightCounting sees 1.6T output value at a 2026–2030 CAGR of 57.6% and +65% for Ethernet optical modules in 2026 — demand is real and steep, supporting profits well above RMB 8.4bn; on the supply side, industry-wide capacity expansion concentrated release in 2027 (including the company's own USD 1.7bn), and LightCounting warns of accelerating price declines once shortages ease — this cannot support the upper end of the RMB 18.8bn consensus. This report's verdict: base-case delivery center of RMB 10.5–13.5bn, falling precisely between the upper half of the price-implied range and the lower half of consensus — the market is applying roughly a 50% haircut, pricing neither a "2026 peak-and-decline" (bear case) nor the sell-side narrative in full. The 52-week high of RMB 275.8 (2026-06-25, market cap ~RMB 505bn) implied about 27x 2027E consensus; the subsequent halving was a violent unwinding of expectations, not fundamental deterioration — the divergence between H1 results +290% and the share price -59% is itself evidence that expectation reset is in progress.

9.3 Three Layers of Value (EPV)

• Asset value (floor): RMB 13.30/share (book value per share, 2026-06-30, including goodwill; RMB 10.75 excluding goodwill) — an independent reference, not to be added to EPV.
• EPV (zero growth): RMB 24.15/share. Inputs (basis = firm): normalized FCFF of RMB 5.86bn/year — annualizing 2026H1 actual recurring net profit of RMB 2.480bn (RMB 4.96bn) as the base period, + income tax RMB 0.87bn + net interest RMB 0.33bn → EBIT RMB 6.50bn → NOPAT RMB 5.52bn + D&A RMB 3.54bn − maintenance CapEx RMB 3.2bn (90% of D&A; the excess within actual 2026H1 CapEx of RMB 5.253bn is expansionary and excluded from steady state) − steady-state working capital change of 0; WACC 10% (range 9.5%–10.5%); minus net debt of RMB 14.37bn (pro forma net debt of about -RMB 3.3bn after the maximum H-share proceeds are received, corresponding to a pro forma EPV of RMB 24–36).
• Growth option (residual): current price RMB 114.22 − EPV RMB 24.15 = RMB 90.07/share, 78.9% of the current price — the market's premium for growth is concentrated in: optical module 2027 capacity ramp (target of 25mn 800G units + 10mn 1.6T units), AI PCB Multek Phase I (USD 1.0bn), and the Apple foldable FPC option.

Interpretation: the growth option share is elevated (78.9% > 15%) — this name does not fit the profile of "the market pays only for zero growth"; the vast majority of the current price is a bet on 2027 capacity absorption and the durability of AI demand. This matches the pricing characteristics of a structural inflection period rather than a bubble (counterevidence to a bubble: the current price is only 11.1x 2027E consensus, PEG 0.48), but the implication is equally blunt: any signal of "2027 supply loosening, 1.6T qualification failure, or hyperscaler capex slowdown" will hit the option layer directly, and a Source Photonics goodwill impairment would pierce both the floor and the EPV simultaneously (RMB 2.712bn of goodwill sits on the optical module asset group). The EPV itself is a conservative construct (only currently commissioned capacity is counted, with Source Photonics utilization running at 90%+); it is not a basis for buying, only a marker of downside reference.

9.4 Three Scenarios and Odds (12-Month Horizon)

Probability-weighted expected value: RMB 140.1 (+22.7%). The current price sits at the lower edge of the base-case range (114.22 vs. lower bound of 115), with an asymmetric upside-skewed distribution — the odds structure is "skewed upward, but with a deep tail" (bear case implies a halving). Anchoring basis for the base-case exit multiple of 20–21.5x: peer TTM P/E of 27.8–40.7x (Innolight/Eoptolink/Shengyi Electronics) discounted 40–50% for share gaps and earnings quality, implying a PEG of ~0.3–0.5 for base-case growth (+43% to +77%) — the anchor is peers, not the company's current price. The main vulnerabilities of this judgment must be explicitly disclosed: if in 2027 the market only assigns 12–15x to this chain (cyclical compression pricing in a high-growth earnings year), the base-case fair value shifts down to RMB 86–110, roughly the current price — at that point "reasonably undervalued" degenerates to "fairly valued," and this report's bullish stance loses its valuation basis; conversely, the bear case's 15–18x is not harsh for a company with earnings declining 60%+. H-share dilution sensitivity: restating the base-case range using post-dilution share count of 1.946–1.982 billion shares yields RMB 108–149, essentially equivalent to the original range after crediting raised cash of ~RMB 5.6 per share — the valuation conclusion is insensitive to the dilution basis.

Reconciliation with external anchors: More pessimistic external coordinates = LightCounting industry view (price declines accelerating after shortage relief, conditional) and the H-share IPO price of HKD 86.80 (RMB 73.80 equivalent — a market-clearing result of financing discount plus cornerstone lock-up, not a fair value judgment) — this report's bear case of RMB 45–60 is more extreme than both, as it additionally layers in the accounting shock of goodwill impairment; no institutional-grade bear-case earnings forecasts are available to cite (ratings 26/26 Buy). More optimistic external coordinates = Citi at RMB 350 (2026-06-24, 2027E net profit RMB 29.3 billion), Guosheng's 2026-07 series of RMB 7.5/24/40 billion — all published near the June stock price peak and not reflecting the October trade disruptions or the H-share discount; this report's bull case of RMB 200–270 is already below their implied levels.

9.5 This Report's Earnings Forecasts (Earnings Reconciliation Anchor)

Cross-check: FY2026E midpoint of RMB 7.4 billion is broadly consistent with sell-side consensus of RMB 7.56 billion; FY2027E base case of RMB 10.5–13.5 billion is 28%–44% below consensus of RMB 18.83 billion — the divergence stems from optical module capacity absorption rate (this report uses a demand-chain-supported 25%–30%, while consensus implies 36%–42%). Methodology: H1 sequential extrapolation plus segment volume-price breakdown; greatest uncertainty is the 1.6T Q4 ramp slope and 2027 800G long-term contract negotiations. The 10-31 Q3 report is the first reconciliation checkpoint.

9.6 Valuation Conclusion

Reasonably undervalued. At RMB 114.22, the current price sits at the lower edge of the base-case fair value range (safety margin of only +0.7% on the lower-bound basis, +19.5% on the midpoint basis, +22.7% on the probability-weighted expected value); quality and price must be assessed separately — quality: the AI dual engines are real, but the company is a chaser, not a leader (Source Photonics gross margin 7–9pp below Eoptolink/Innolight, AI PCB 1–2 years behind Shengyi Electronics/WUS, group ROIC just above WACC); price: the market has already priced in a 7%–56% discount to 2027 consensus — pessimism is meaningfully priced in, but a bear-case clearing has not occurred. The conclusion is sensitive to the exit multiple assumption (at the 12–15x tier it reverts to fairly valued), hence the rating is cautiously bullish rather than bullish; in portfolio terms it is better suited as a "verification-driven" gradual accumulation target — until the 10-31 Q3 report and cloud providers' Jan–Feb 2027 guidance land, maintain disciplined tracking of the 38% optical module gross margin line.

X. Industry Panorama and Competitive Landscape

10.1 Industry Space and Demand Inflection Quantification Chain

The market where the company truly competes is the intersection of three tracks: FPC, high-multilayer, and HDI within global PCB (Prismark basis) + global datacom optical modules (LightCounting basis) + NEV precision structural components.

Global PCB: 2024 output value USD 73.57 billion (+5.8%) → 2025 USD 85.15 billion (+15.8%, Prismark website) → 2026E ~USD 95.8 billion (+12.5%, Zheshang Securities citing Prismark) → 2030E USD 123.3 billion (2025–2030 CAGR 7.7%). Mainland China accounts for 56% of global output (USD 41.2 billion in 2024). Structural divergence far exceeds the aggregate: server/data storage PCB at USD 15.675 billion in 2025 (+43.6%) → 2030E USD 34.68 billion (CAGR 17.2%); 2026E output value for 18-layer-and-above multilayer boards +79%, HDI +23%, packaging substrates +28.8% (Prismark) — while the FPC segment (~USD 12.5 billion, 17% share) is near zero growth. Quantified chain of the AI demand inflection: driver = North American Big Four cloud capex (2025 actual ~USD 410 billion → 2026 guidance ~USD 730 billion, +77%; Morgan Stanley 2027E Big Five cloud providers ~USD 1.2 trillion, 2028E ~USD 1.4 trillion) × per-unit usage leap = AI server PCB value of 3–5x traditional (HGX B series USD 175–204 per rack GPU → GB200 NVL72 USD 317–376 → GB300 NVL72 USD 382–501, GF Securities 2025-01 estimate; Rubin NVL144 a further +233%, supply chain teardown 2026-05) × penetration pace = AI penetration concentrated in 2025–2027 (server PCB CAGR 17.2%) × company share = Dongshan No. 3 globally in PCB, an AI PCB chaser (Multek USD 1 billion expansion Phase I released 2026Q3).

Global optical modules: 2025 sales ~USD 23 billion; Ethernet optical modules +93% in 2024 → +82% in 2025 → 2026E +65% (LightCounting 2026-04 forecast, implying total 2026 market of ~USD 38 billion); 1.6T market USD 7.3 billion in 2026 → USD 44.8 billion by 2030 (CAGR 57.6%), with 1.6T output value officially surpassing 800G in 2026. Company chain: Source Photonics No. 7 globally in 2025 (CIC Consulting, ~3% share) → 2026H1 optical module revenue RMB 5.351 billion (annualized ~USD 1.5 billion, estimated ~4% of 2026E Ethernet market) → 2027E capacity targets of 25 million 800G units + 10 million 1.6T units, optical chip monthly capacity of 9 million units in 2026Q2 → planned 22 million units in 2026Q4 → 300 million units/year in 2027 (management guidance) — if capacity is absorbed as scheduled, optical module revenue still has multiple-fold room versus the 2026H1 annualized level; but full-capacity basis requires 60%+ global 1.6T share, unattainable; a realistic absorption ceiling of 20%–40% corresponds to optical module revenue of RMB 25–50 billion — this is precisely the mathematical origin of this report's base case (RMB 38–50 billion) and the bull-bear boundary.

NEV structural components: 2025 precision components revenue of RMB 5.930 billion (+30.6%), customers primarily Tesla, with the GMD acquisition filling in European capacity — this business grows steadily and is not the main line of this report's valuation.

10.2 Industry Chain and Value Distribution

Upstream: PCB copper-clad laminates/high-speed resins (Shengyi Technology, EMC, ITEQ — Shengyi's 2026H1 CCL gross margin 29.16%, +5.47pct; upstream captures the thickest slice during shortages), optical module EML laser chips (Lumentum/Mitsubishi/Sumitomo/Coherent, domestic substitution rate ~5%) and DSP (Broadcom/Marvell). EML is the tightest bottleneck industry-wide (demand exceeds supply by ~30%, LightCounting); upstream holds the bargaining advantage over midstream during shortages — Source Photonics has avoided the tightest link with partially in-house EML, but self-sufficiency is under 10% of BOM. Midstream: PCB/FPC manufacturing (Avary/Dongshan/WUS/SCC/Shengyi Electronics/Victory Giant) and optical module packaging & testing (Innolight/Eoptolink/Coherent/Source Photonics) — AI rigid boards and optical modules carry the thickest gross margins (Source Photonics 39.33%, WUS 38.7%), while consumer electronics FPC and structural components the thinnest (Dongshan precision components 12.13%, optoelectronic display 5.89%). Downstream: Apple (annual price-cut pressure, weak bargaining power) and North American cloud providers (2026 capex +77%, improved bargaining power during shortages, likely weakening after 2027 supply release). Overseas revenue was 83.2% of 2026H1 — the tariff exposure exceeds A-share peers, but Thailand FPC (volume ramp in 2026H2 with new models), France GMD, and Source Photonics overseas plants' global capacity positioning instead constitute an order-taking barrier as trade barriers rise.

10.3 Supply-Demand and Competitive Structure

On the supply side, 2026 is in tight balance tending toward shortage: PPE resin shortage once drove a 40% single-month PCB price increase (CCTV Finance), 30% EML gap, Source Photonics capacity utilization 90%+. But the risk of supply-demand loosening in 2027 is real: AI PCB faces an "unprecedented expansion wave" (Multek USD 1 billion, WUS Thailand, Shengyi Electronics Jiangxi, Victory Giant/SCC in parallel), and optical module capacity doubling at Innolight/Eoptolink/Source Photonics — LightCounting explicitly warns that after shortage relief, duplicate orders vanish and price declines accelerate; 800G has already shown "price cuts with volume" (average price USD 480 → 350–380). Concentration is layered: PCB overall market CR10 ~50%, FPC duopoly (Avary >30%, Dongshan No. 2, per Prismark for years running); optical modules highly concentrated (in 2025 Innolight No. 1 at USD 5.33 billion, Eoptolink No. 2 at USD 3.5 billion, with Chinese companies holding seven of the global top ten, per LightCounting), Source Photonics No. 7 at ~3%. Entry barriers: 2–3 year customer qualification, USD 1 billion-scale capital per site, 78-layer HLC / 7-step HDI / EML epitaxial growth processes, and overseas capacity footprint itself becoming a geopolitical barrier. Substitution threat: CPO co-packaged optics entering small-scale commercial use in 2026–2027, high-volume deployment possibly appearing in 2027 (LightCounting), with CPO port share in the 3.2T generation rising significantly by 2029 — a medium-term diversion risk for pluggable EML modules, which Source Photonics hedges with a "silicon photonics + in-house EML dual-route" strategy.

10.4 Cyclicality and Regulation

The industry is in year 3 of the AI hardware upcycle: PCB -14.9% in 2023 (destocking) → +5.8% in 2024 → +15.8% in 2025; optical modules +80% to +93% for two consecutive years, then 2026E +65% — already in the second half of the "high base + expansion wave" phase. Leading indicators: North American cloud providers' quarterly capex (2026Q2 four companies combined USD 171.2 billion, +78.6%), EML gap convergence pace (expected to ease from end-2026 into 2027), upstream material prices, order visibility, and NVIDIA GB300 → Rubin platform shipments. Regulation: US-China tariffs entered a truce extended after two rounds of talks in 2025; from 2026-01-15 the US levies 25% on imported computing chips and derivatives (Section 232, PCB main bodies not yet on the core list but derivatives scope uncertain); on 2026-10-08, rumors of "65% US-origin materials / FCC restricting Chinese optical modules" sent the stock limit-down, with the company clarifying these were unofficial policies — policy tail risk is real and has already been experienced by the market once. China's rare earth/gallium-germanium export controls push up upstream material costs.

10.5 Company Industry Positioning

A scarce A-share dual positioning in "electrical + optical interconnect": FPC No. 2 globally (Prismark), PCB No. 3 globally (2025 revenue, CIC/Prismark basis — note: the company's IR materials describe itself as "a top-3 global PCB supplier, No. 2 global FPC supplier, and a top-10 global optical module supplier," presented alongside third-party figures), optical modules No. 7 globally. Share trends: FPC stable but the segment near zero growth; AI PCB in catch-up phase (1–2 years behind WUS/Victory Giant/Shengyi Electronics); optical modules rising rapidly (No. 10 → No. 7, estimated ~4% in 2026H1). Moat sources: decade-long Apple/Tesla major-customer bindings + global capacity across 14 countries + Source Photonics' partial in-house EML + M&A integration execution (Source Photonics ramped immediately in its first year of consolidation). Weaknesses: low consumer electronics FPC margins, Apple dependence at 46.46%, AI PCB qualification and capacity realization unverified, and depth of North American cloud binding behind Innolight/Eoptolink.

XI. Overall Conclusion and Tracking

Overall assessment: Cautiously bullish, confidence 0.55, 12-month horizon. Dongshan Precision is a stock on the AI compute hardware chain where "results are half delivered and expectations are half discounted": 2026H1 net profit +290% demonstrates M&A integration and optical module ramp capability (though including consolidation effects and non-recurring gains); the current price embeds a 7%–56% discount to 2027 consensus and sits at the lower edge of this report's base-case fair range (RMB 115–158), with a probability-weighted expected value of +22.7% and upside-skewed odds. But the directional conviction is constrained by three factors: H-share listing on 10-28 and initial price-comparison and liquidity disruptions (the intraday -8.9% on 10-26 was a preview), the industry clock of supply loosening in 2027, and the amplifying effect of RMB 4.683 billion goodwill on any earnings downgrade — the -54% bear-case midpoint tail depth means this is not a setup for heavy left-side positioning. Strategic implication: use the 10-31 Q3 report (optical module gross margin ≥38%, sequential revenue) and cloud providers' Jan–Feb 2027 capex guidance as staged verification points; if fundamentals verify while the H-share price-comparison suppression persists, time will be a friend.

Key risk disclosures (see investment thesis C5 and Chapter VIII financial-engineering section): ① 2027 optical module price war exceeding expectations (800G annual decline >25%); ② Source Photonics goodwill impairment (RMB 2.712 billion; 2026/2027 annual reports are the testing windows); ③ substantive US restrictions on optical modules/PCBs (65% origin rule / tightened FCC authorization); ④ financing chain under high leverage (interest-bearing debt RMB 23.963 billion, H1 FCF -RMB 2.869 billion; interest + FX eroding RMB 0.6–1.0 billion annually); ⑤ dual concentration in the largest customer (46.46%) and North American cloud providers.

Tracking discipline: Optical module quarterly revenue and gross margin (falsification line at 38%), AI PCB Phase I assets transferred to production and electronic circuit gross margin, execution of H-share proceeds deployment, 800G/1.6T long-term contract annual price declines, Source Photonics goodwill impairment test disclosures, controlling shareholder pledge ratio (baseline 8.27%) and the unlock-expectation dynamics around 2028-06.

This report uses the 2026-10-23 A-share closing price of RMB 114.22 as the valuation anchor (frozen basis); see the valuation chapter for H-share post-issuance share count and dilution sensitivity. Report date: 2026-10-26.