Rating: Cautiously Bullish | Target Price Range: RMB 92–104 (12-month base-case fair value) | Current Price: RMB 87.28 (close on 2026-10-23) | Margin of Safety: approx. +5% (measured against the lower bound of base-case fair value) | Time Horizon: 12–18 months
Shengyi Electronics is the steepest riser among the first tier of domestic AI compute PCB players: FY2025 revenue of RMB 9.494 billion (+102.6%), AI compute-related product revenue up +242%, gross margin rising from 14.3% in FY2023 to 34.3% in 2026H1, and a forecast of RMB 1.869–1.952 billion in net profit attributable to the parent for the first three quarters of 2026 (+68%–75%). Industry-wide, output value of multilayer boards with 18+ layers is expected to grow +72.8% in 2026 (Prismark forecast, as cited in the company's interim report). The current price of RMB 87.28 corresponds to approximately 18.8x 2027E consensus PE and roughly 21.5x based on this report's base-case earnings—valuation is not stretched. However, three constraints justify only a "cautious" label: First, the market is effectively pricing in only about 72% of sell-side consensus; whether this discount reflects second-tier status or an opportunity for mispricing reversal cannot be decomposed from price alone. Second, the private placement share unlock in February–April 2027 (22.67 million shares, cost basis RMB 111.58) and the potential share conversion of the controlling shareholder's exchangeable bonds (conversion price RMB 97.44) form two supply gates sitting right near the upper edge of the base-case fair value range. Third, the industry's RMB 80 billion-plus capacity expansion wave in 2027–2028, combined with decelerating AI capex growth, constitutes the major headwind in the second half of the holding period. Base-case fair value is RMB 92–104 (+5% to +19% vs. current price), with a probability-weighted expectation of approximately RMB 99.4 (+13.9%); the bear-case scenario is RMB 37–56 (-36% to -58%). Odds skew positive but downside protection is limited—phased position building is preferable to a single full allocation.
The key storyline over the next 18 months: earnings growth shifts gears from +109% in 2026H1 to approximately +85%–93% for full-year 2026 (this report's estimate: RMB 2.75–2.84 billion), then to a base case of +18%–28% in 2027 (RMB 3.3–3.6 billion), with momentum transitioning from "explosive volume-price growth" to "capacity ramp-up relay." The depreciation from assets transferring to fixed assets at Thailand Phase I (rushing toward trial production in 2026Q3) and Ji'an Phase I (trial production already started at end of June 2026)—with annual depreciation expected to jump from approximately RMB 470 million to RMB 800 million–1 billion—will test net margins (base assumption: a decline from about 19% to 17%–18%).
Primary stock price driver: North American cloud providers' 2027 capex guidance (heavily disclosed during the earnings season from 2026-10-27 to 11-05) and the transmission of NVIDIA's Vera Rubin platform's 2026H2 mass production into high-multilayer board orders—this determines whether 2027E net profit attributable lands at the implied ~RMB 2.9 billion or at the consensus RMB 4.0 billion.
Verifiable expectation gap: Back-solving from the industry peer 2027E trading multiple midpoint of 26x, the current market cap of RMB 75.08 billion implies only about RMB 2.89 billion in 2027E net profit attributable—equivalent to 72% of consensus (RMB 4.0 billion, 9 analysts per iFinD, as of 2026-10-23)—meaning the market has discounted sell-side expectations by roughly 28%. This report must also present the counter-reading: the discount may precisely be the market's rational pricing of second-tier status (the top seat in AI compute accelerator cards/UBB belongs to Victory Giant Technology; the first tier in high-speed switches is WUS Printed Circuit) and the 2028 supply cycle—price data cannot distinguish between the two. The 10-30 Q3 report and November cloud provider guidance are the two-way verification windows.
Verification catalysts: K1 (10-30 Q3 report landing within the forecast range), K6 (cloud providers' 2027 capex guidance), K2 (Rubin order transmission), K4 (Thailand trial production). Falsification conditions: gross margin declining more than 1 percentage point QoQ for two consecutive quarters; 2027E consensus cut by more than 15%; signals of order transfer away from the largest customer.
Key evidence:
On volume, the structural re-rating has genuine support: NVIDIA's annual platform refresh drives step-change increases in PCB layer counts and area per machine (compute boards from 22 to 26+ layers, switch boards from 24 to 32 layers, Rubin Ultra orthogonal backplanes up to 78 layers); media and sell-side extrapolations put the PCB value per VR200 NVL72 rack at approximately USD 116,700, up about 233% from GB300 (rack-level estimate, an extrapolation not confirmed by the original manufacturer); AI PCB's share of global PCB output value is projected to climb from about 10% in 2025 to about 25% by 2030 (broker estimate). At the same time, this report stresses the other side: per Cailian Press statistics, 20 PCB companies announced expansions in 2026 with investment exceeding RMB 80 billion, with new capacity concentrated for release in 2027–2028; Taiwan-based leader Unimicron (2026-10-22) has already warned of "emerging duplicate orders" in AI PCBs; Morgan Stanley (2026-10-16) estimates global AI capex growth at roughly +100% in 2026, about +60% in 2027, and slowing to about +14% in 2028. Structural volume growth does not exempt the industry from the supply cycle—this anchors this report's holding period at 12–18 months (the window before capacity release and capex downshifting), rather than an open-ended growth story.
Key evidence:
Delivery at the company level is fact, not narrative: FY2025 revenue +102.6%, net profit attributable +343.8%, gross margin 31.2%; 2026H1 net profit attributable of RMB 1.111 billion (+109.4%), gross margin further rising to 34.3%, with operating cash flow expanding in tandem (FY2025 net operating cash flow of RMB 1.621 billion, cash-to-net-profit ratio 1.10). Two discounts need to be acknowledged: First, "certified and delivered at scale" is the company's own claim without naming platforms, and the same semi-annual report discloses that 1.6T switch support is only at "small-batch mass production" and 1.6T optical module PCBs are still in prototype testing—no product-level revenue breakdown (HDI vs. high-multilayer) is disclosed. Second, single-quarter net profit attributable YoY growth has shifted from +122% in Q1 to +101.6% in Q2 to an estimated +30%–44% in Q3—the growth downshift has already occurred, while consensus implies a Q4 single quarter of approximately RMB 946 million, +18.3% QoQ—a demanding bar. Ji'an Phase I only began trial production at end of June 2026 and Thailand Phase I is still in equipment commissioning; the company has not disclosed the timing correspondence between orders and delivery—"capacity deployment aligning with the Rubin volume window" is currently a synchronicity-based inference, with verification resting on the 10-30 Q3 report's contract liabilities, inventory, and QoQ gross margin.
Key evidence:
Three layers of valuation decomposition: asset value of approximately RMB 10.5 per share (post-placement book net asset per share); an EPV zero-growth layer of approximately RMB 23.8 per share (normalized FCFE of approximately RMB 2.50 per share from operating assets ÷ 10.5% cost of equity, including only operating capacity and excluding new capacity in Ji'an/Thailand/Dongguan); the gap between the current price and EPV—approximately RMB 63 (about 73% of the current price)—is entirely a bet on growth. The growth option is a non-trivial component, so this report does not claim "the market is pricing only zero growth"—richness is determined by the ramp-up scenario. Under the base case (2027E net profit attributable of RMB 3.3–3.6 billion × exit PE of 23–26x, with multiples anchored to the range formed by Victory Giant's 2027E research report multiple of 22.4x and WUS's 26x, taking the mid-to-upper segment), fair value is RMB 92–104 with a midpoint of approximately RMB 98. The counter-argument must be stated clearly: the market's 72% discount to consensus could be mispricing—or a rational discount for second-tier status and the 2028 supply cycle. If leader Victory Giant trades at only about 18–22x 2027E (approximately 18x per stock forum consensus, 22.4x per research reports), then Shengyi's current 21.5x (base-case earnings basis) can hardly be called a significant discount; the sell-side median target of RMB 159 (implying +82%) was formed before October's sharp sell-off and its timeliness is questionable; PB of 11.5x remains at the 77th percentile of the past 5 years. Overall judgment: valuation is not in bubble territory and has moderate upside, but this is not deep mispricing.
Key evidence:
The risk is quantifiable: if 2027 North American cloud capex guidance downshifts in tandem with new capacity release in 2027H2–2028, referencing the 2023 trough (gross margin 14.3%, net profit attributable turning negative), the bear case sees 2027E net profit attributable cut to RMB 2.0–2.4 billion, valuing the stock at RMB 37–56 on a 16–20x cyclical steady-state multiple; an extreme trough (2028 earnings approaching zero, PE failing) values it at approximately RMB 20–26 on a PB of 2.0–2.5x. On the chip supply side, the private placement unlock (currently at an unrealized loss of about 22%) and EB conversion (above RMB 97.44) together represent about 38.93 million shares—approximately 13.9% of the free float in potential supply—concentrated in February–April 2027 within the 12-month holding window. The upper edge of the base-case fair value range must pass through these two gates; this is the main reason this report has cut the exit multiple from the upper segment of the peer midpoint to 23–26x. Furthermore, the company itself is a participant in the RMB 80 billion expansion wave (Ji'an Xin Suan Zhi Lian at RMB 2.257 billion, the Dongguan HDI base, Thailand Phase I), while HVLP4 copper foil scheduling stretching to 2027 and upstream material quotas (per industry media) are hard constraints on new capacity ramp-up—capacity deployment does not equal effective output.
| Metric | FY2023 | FY2024 | FY2025 | 2026Q1 | 2026H1 (Latest) | 2026Q3E (Implied by Forecast) |
|---|---|---|---|---|---|---|
| Operating revenue | 32.73 | 46.87 | 94.94 | 24.11 | 57.84 | — |
| Net profit attributable to parent | -0.25 | 3.32 | 14.73 | 4.45 | 11.11 | 7.58–8.41 (single quarter) |
| Non-GAAP net profit attributable to parent | -0.44 | 3.27 | 14.68 | — | 11.11 | — |
| Gross margin | 14.29% | 22.73% | 31.16% | 35.21% | 34.31% | — |
| Net margin | — | 7.08% | 15.52% | 18.45% | 19.20% | — |
| Net operating cash flow | 4.31 | 3.51 | 16.21 | 6.15 | 9.17 | — |
| Free cash flow (OCF − capex) | -1.87 | -0.74 | -0.19 | -0.38 | -3.51 | — |
| Cash and equivalents (period-end) | — | 4.10 | 7.44 | — | 4.07 (as of 2026-06-30) | — |
| Interest-bearing debt (period-end) | — | 13.53 | 21.97 | — | 26.98 (as of 2026-06-30) | — |
| Debt-to-asset ratio (period-end) | 37.51% | 44.40% | 48.88% | — | 53.76% (as of 2026-06-30) | — |
| Weighted ROE | — | 8.24% | 29.74% | — | 18.1% (half-year basis) | — |
Note: After the net proceeds of RMB 2.52 billion from the September 2026 private placement arrive, cash and the debt-to-asset ratio will improve significantly versus the 2026H1 period-end basis (total share capital increased to 860.26 million shares); balance sheet stock figures are as of the latest period (2026H1) and marked as-of.
Reasons for metric changes (company explanations attached where YoY ≥ ±20%): ① FY2025 revenue +102.6%—the company explains it as "higher proportion of high value-added products and YoY volume growth"; AI compute-related products +242%, server revenue share rising above 60%; our estimate shows average selling price +73% (RMB 5,329/㎡) while volume rose only +17.8%—growth was primarily driven by product mix upgrade; ② FY2025 gross margin +8.4pp—the company attributes this to the higher share of high value-added products; ③ FY2025 capex RMB 1.640 billion (+286%)—directed to Intelligent Computing Center Phase I and Thailand Plant Phase I; ④ FY2025 finance costs +283.6%—"increased financing scale, higher interest expense, and reduced FX gains"; ⑤ 2026H1 inventory +50.1% vs. the start of the year (RMB 2.290 billion)—"increased orders, higher period-end inventory balance," plus proactive stockpiling amid upstream laminate price increases; ⑥ 2026H1 cash -45.3% vs. the start of the year—increased investing cash outflows, since replenished by the placement proceeds in September.
The company forecasts net profit attributable to the parent of RMB 1.869–1.952 billion for the first three quarters of 2026 (+68% to +75%), and RMB 1.864–1.947 billion on a non-GAAP basis (non-recurring items are negligible—earnings quality is high). Back-calculating from the actual 2026H1 net profit attributable of RMB 1.111 billion, Q3 single-quarter profit is RMB 758–841 million, with a midpoint of approximately RMB 799 million: +14% to +26% QoQ from Q2 (RMB 666 million)—momentum continues. However, YoY growth has shifted from +122% in Q1 and +101.6% in Q2 to +30% to +44%—a growth deceleration off a high base is normal, and the market's interpretation is neutral. The company attributes this to "benefiting from strong AI compute demand... orderly capacity release, deepened technology upgrades, and strengthened quality control, with the share of high value-added products steadily rising." Against management guidance: no quantitative guidance, but the 10-24 "Announcement on Recent Production and Operations" confirms "ample orders, stable supply chain, high capacity utilization, with orders for AI servers, 800G/1.6T switches, and other data center products growing steadily, and breakthrough progress on 1.6T optical module products." Peer cross-verification: WUS Printed Circuit forecast net profit attributable of RMB 5.45–5.7 billion for the first three quarters (+100.5% to +109.7%); parent company Shengyi Technology forecast RMB 6.46–6.87 billion (+164% to 181%)—industry chain momentum is corroborated top-down. A quantitative comparison of the Q3 forecast versus sell-side consensus was not obtained (covering institutions have not yet published commentary following the forecast), so no conclusion is drawn on "beat or miss"; the formal Q3 report is due on 10-30, when contract liabilities, inventory QoQ, and gross margin will test order visibility.
The 10-26 intraday price was approximately RMB 82.8, about -5.1% below the 10-23 close of RMB 87.28. Verification findings: this decline was a broad AI hardware sector selloff—active funds' overweight in AI hardware reached 30.4 percentage points, a historical record (21st Century Business Herald, 2026-10-09); the sector entered a crowding-digestion phase in October, compounded by the compute-chain pullback and profit-taking triggered by the OpenAI revenue expectation gap. The company's positive 10-24 announcement (ample orders) failed to halt the decline; no new company-specific negatives emerged. Short-term price has temporarily decoupled from fundamentals, and amplified volatility is a variable the entry pace must tolerate; the valuation anchor remains the 10-23 close of RMB 87.28.
Business model snapshot: Asset-heavy, order-driven custom manufacturing (build-to-order + direct sales), with no subscription or recurring revenue; core assets are high multilayer board and advanced HDI capacity plus 40 years of process know-how in datacom/telecom; customers are concentrated among leading AI compute/communications vendors (top five at 65.2%), and products form an entry barrier through qualification with leading global compute platforms (qualification cycles measured in years). Current pricing power stems from the AI supply-demand gap: FY2025 ASP +73%, and despite multiple rounds of upstream copper clad laminate (CCL) price hikes in 2026H1, gross margin still rose to 34.3%—costs can be passed downstream—but the essence of this pricing power is cyclical supply tightness, not brand premium.
Cash content of earnings: Book profits largely convert to cash—the ratio of operating cash flow to net profit attributable to shareholders was 1.06 / 1.10 / 0.83 for FY2024/FY2025/2026H1 (FY2025 cash collection ratio ~90%); accounts receivable growth (2026H1 ~+31.6%) was below revenue growth (+53.5%), indicating normal collection quality. However, free cash flow has been persistently negative: FY2023–2026H1 at -187 / -74 / -19 / -351 million yuan respectively—not an earnings quality issue but a choice to expand: capex was 1.640 billion yuan in FY2025 (+286%) and another 1.268 billion in 2026H1, with four major projects (Dongcheng compute center, Ji'an, Thailand, Dongguan HDI) running in parallel; the funding gap is covered by a 2.52 billion yuan private placement in September and borrowings. Non-GAAP and GAAP profit differ minimally (non-recurring items were only 0.4% of attributable profit in FY2025), so earnings are genuine core-business profits.
Return on capital: FY2025 weighted ROE 29.74%, estimated ROIC ~19.3% (2026H1 annualized ~24.8%), significantly above the cost of equity (10.5% used in this report)—expansion is creating value, not destroying it; this is the dividing line between "asset-heavy expansion" and a "capital black hole."
Capex quality: FY2025 capex/depreciation reached 4.3x and remained elevated in 2026H1; the company does not break out maintenance vs. growth capex—judging by the current construction scale, it is growth-led (new site construction), and maintenance spend cannot be independently verified; once assets are placed in service, annual depreciation is expected to jump from ~470 million yuan to 800 million–1 billion yuan, with depreciation front-running revenue during the ramp-up period—this is the main deduction from the 2027 net margin assumption (17%–18%).
Red flags: ① Extremely high and rapidly rising single-customer dependence (largest customer 22.75%→42.90%); the company itself explicitly flags in its annual report risk section "dependence on overseas markets, a single customer, or a single sector"; ② Growth relies on external financing (negative FCF + ~2.7% dilution from private placement); ③ Earnings are deeply tied to the AI capex cycle, and the company's own assessment "risk of significant profit decline or losses" applies—2023 (gross margin 14.3%, attributable net profit -25 million yuan) is what this business model looks like at the cyclical trough.
Track record of words vs. deeds: pragmatic, delivery-oriented. Three sets of promises checked: ① FY2024 annual report pledge to "consolidate the server market while expanding into switches, high-end power supplies, and LEO satellites"→FY2025: server revenue >60% of total, AI compute products +242%, 1.6T switches in small-volume mass production, satellite communication in small-volume delivery (delivered); ② Smart compute center project "complete equipment installation, commissioning, and trial production in 2025"→Phase I trial run in 2025H2 with output meeting expectations (delivered); ③ Thailand plant "trial production in 2026"→2026H1 equipment installation and commissioning underway, racing toward trial production in 2026Q3 (largely delivered, in progress). The 2026-10-17 earnings preannouncement was voluntary disclosure, pending verification by the Q3 report on 10-30.
Shareholder friendliness: generous dividends, but financing and returns run in parallel. FY2024 payout ratio 61.5% (204 million yuan); FY2025 total dividends 741 million yuan (payout ratio 50.3%); cumulative dividends since listing 1.486 billion yuan, 53.1% of cumulative attributable profit; a 50 million yuan buyback completed in 2026-03, and on 2026-10-19 a new 100–200 million yuan buyback announced (cap 179.34 yuan/share; first tranche ~25 million yuan executed on 10-22). The other side: the 2026-09 private placement of 2.529 billion yuan diluted ~2.7% (issue price 111.58 yuan, currently ~22% underwater); the controlling shareholder pledged 25.6 million shares for an exchangeable bond collateral account (exchange period opened from 2026-08, ~16.26 million shares remaining)—a modest shareholder-level supply overhang. Overall verdict: neutral-to-friendly—dividends and buybacks are real, but sustained heavy capex makes external financing a regular occurrence.
Risk signals: Related-party procurement from controlling shareholder Shengyi Technology is rising (FY2025 14.51%→2026H1 16.26%); fairness of related-party pricing amid the CCL price-hike cycle requires ongoing monitoring; after the 2026-05 board reshuffle, the GM/CFO/board secretary are new, while founder Deng Chunhua remains chairman—operational continuity has not been broken; no insider (directors/supervisors/senior management) secondary-market sales in 2026 (most recent: small sale by core technical staff and a 1% reduction by Dongguan Science & Technology Venture Capital in 2025-12).
| Segment (by end market) | Revenue share | Profit contribution | YoY | Business logic in one sentence |
|---|---|---|---|---|
| Computers/servers (incl. AI compute) | >60% (FY2025 revenue basis) | Primary profit driver: key lift for overall gross margin | AI compute-related +242% | High-multilayer boards for AI server UBB/OAM/accelerator cards; 5-6 order HDI + 30+ layers in mass production; volume-price rise led by price (ASP +73%) |
| Communications network equipment (switches/routers/optical modules/satellite) | Second-largest segment (not separately disclosed; 2026H1 communications boards ~29.5% order of magnitude) | Secondary driver | Steady growth | 800G in volume, 1.6T switches in small batches, 1.6T optical modules in sampling—second growth curve |
| Automotive electronics | No longer separately disclosed (~17% by order intake in 2023) | Small share | Volume orders for domain control/radar PCBs | Strategic cultivation; unit price and margin below server boards |
| Consumer electronics/industrial control & medical/others | <20% and shrinking | Marginal | Intensifying low-end competition | Resources shifting toward AI compute; share keeps declining |
The primary profit driver is clearly AI server boards: the company does not disclose gross margin by application, but overall gross margin rose from 14.3% to 34.3% over three years, and FY2025 electronic components (core business) gross margin was 28.62% (+9.2pp); the company attributes this to "a higher share of high value-added products," mutually corroborating the >60% AI server board revenue share—the gross margin structure gap (AI server boards vs. consumer boards) far exceeds 10pp, essentially a difference in bargaining power under supply-demand gaps and qualification barriers, not a step change in process cost. Capacity structure (supplementary segment dimension): current effective output is constrained by high-end capacity (FY2025 volume of 1.7162 million m² grew only +17.8%; revenue doubling was driven mainly by mix upgrade); capacity ramp schedule—Dongcheng compute center Phase II in use from 2026Q2, Ji'an Phase I in trial production (700,000 m²/year at full run), Thailand Phase I racing toward trial production in 2026Q3, and Dongguan AI HDI base plus Ji'an Xin Suan Zhi Lian (445,900 m²/year; ground broken 2026-09, 36-month build) as 2027–2028 increments—capacity deployment is the main physical variable for revenue growth over the next two years, with material quotas (HVLP4 copper foil, etc.) as the ramp constraint.
Accounting red flags: ① High and rising share of related-party procurement from the controlling shareholder (FY2025 14.51%→2026H1 16.26%; interim report "Financial Risks—risk of high share of related-party procurement")—medium severity: the controlling shareholder is both the largest shareholder (57.88% stake) and the core CCL supplier; pricing fairness during the price-hike cycle should be tracked quarterly; no evidence of abnormal pricing so far; ② No other notable accounting techniques detected (no goodwill, intangibles ~1.6% of assets, non-recurring items only 0.4% of attributable profit, no signs of abnormal R&D capitalization).
Cross-period consistency: ① Blended gross margin 14.3%→22.7%→31.2%→34.3% (FY2023–2026H1), a 20pp jump over three years, consistent with management's period-by-period explanation ("higher share of high value-added products") and internally consistent with the volume-price split of ASP +73% and volume +17.8%—consistent; ② Top-five customer concentration 46.24%→65.21%, largest customer 22.75%→42.90%, consistent with management's statement about "deepening relationships with top-tier customers and successfully entering their core PCB supply chains for key components," but the steep slope of concentration itself constitutes risk (see C4)—consistent; ③ Interest-bearing debt 1.35→2.20→2.70 billion yuan with rising finance costs, directionally aligned with construction in progress +255% and capex +286%, explained as "increased financing scale"—consistent. Overall judgment: based on the FY2023–2026H1 financials extracted here, no notable accounting techniques or cross-period anomalies were found; earnings quality and disclosure credibility are relatively high.
Current market data and multiples: Closing price 87.28 yuan (2026-10-23), total shares 860.26 million (post-placement), market cap 75.08 billion yuan. PE(TTM) 36.6x (TTM attributable profit 2.053 billion yuan)—1-year/3-year percentile of 0% (a low percentile during an earnings explosion reflects "earnings outpacing price"; the percentile is distorted and serves only as a floor reference); PE (static FY2025) 51.0x; PB 11.5x at the 77th percentile of the past 5 years (~8.3x post-placement dilution)—not cheap in absolute terms. Forward basis: 2026E PE ~26.9x (midpoint of this report's base attributable profit of 2.795 billion yuan) / 26.3x (consensus 2.856 billion); 2027E PE ~21.5x (this report's base midpoint of 3.5 billion) / 18.8x (consensus 4.0 billion). This stock sits on a structural demand inflection; the primary valuation anchor is growth-based (forward PE/ramp estimates), with percentiles serving only as downside reference.
Peer comparison:
| Company | 2026E PE | 2026E growth | Notes |
|---|---|---|---|
| Shennan WUS (WUS Printed Circuit) | 33.8x | +65.6% (consensus) | Leading domestic player in high-speed switches + AI boards; 2026H1 gross margin 38.7% |
| Victory Giant Technology | 26.6x | ~+100% (sell-side) | #1 global share in AI compute cards/UBB; 2027E sell-side ~22.4x |
| Shennan Circuits | 33.8x | ~+49% (own estimate) | Triad of PCB + substrates + packaging/testing; lower AI purity |
| GCE (2368.TW) | ~32.7x | +110.8% (FactSet median) | Taiwanese core in AI server boards; AI board gross margin 40–50% |
| Shengyi Electronics | 26.3–26.9x | +84%–93% | Fast-rising second-tier player |
Market-implied expectations: Backing out from a peer midpoint 2027E trading multiple of ~26x, the current price implies 2027E attributable profit of ~2.89 billion yuan, or 72% of consensus (4.0 billion); backing out from a 2028 steady-state PE of 18–22x implies 2028E attributable profit of ~3.4–4.2 billion yuan, corresponding to FY2025→2028E attributable profit CAGR of ~32%–41%. Cross-checked against demand math: industry CAGR of 30.3% for 18+ layer multilayer boards 2025–2030 (company citing Prismark) × the company's doubling-style capacity deployment × Q3 2026 run-rate already annualizing to ~3.2 billion yuan—the implied growth falls within the achievable range, near the lower bound; what the current price requires of the company is below sell-side expectations and close to the realized run-rate, neither overdrawn nor paying for upside options. The necessary counter-reading placed side by side: this 72% discount cannot be distinguished from "the market accepts consensus but assigns a lower multiple"—if it is a rational pricing of second-tier status and the 2028 supply cycle, it does not constitute an exploitable gap; the sell-side median target of 159 yuan was formed before the October selloff, so its timeliness is questionable.
Three-tier value (EPV): Asset value ~10.5 yuan/share (deep downside floor); EPV-zero growth ~23.8 yuan/share (normalized FCFE of in-service assets ~2.50 yuan/share: 2026H1 non-GAAP annualized at 2.22 billion yuan, bridged for depreciation and working capital to ~2.15 billion ÷ 860.26 million shares, cost of equity 10.5%—includes only in-service capacity); the gap between current price and EPV is ~63.5 yuan, with the growth option ~73% of the current price—the price is mainly supported by the growth tier, and the zero-growth floor is far below the current price; this is the normal structure of a high-purity growth stock rather than evidence of overvaluation per se—cheapness is adjudicated by ramp scenarios.
Three scenarios and odds (12 months, discounted at 2027E):
| Scenario | Probability | Fair range | Core assumptions | vs. current price |
|---|---|---|---|---|
| Bear | 22% | 37–56 yuan | 2027 cloud capex guidance downgraded + gross margin falls back to 26%–28%; 2027E attributable profit 2.0–2.4 bn × 16–20x cyclical steady state | -36% to -58% |
| Base | 50% | 92–104 yuan | 2027E revenue 18.0–20.4 bn (+28%–45%), attributable profit 3.3–3.6 bn (net margin 17%–18%, depreciation jump already deducted) × exit PE 23–26x (anchored to the upper-middle of the Victory Giant 22.4x / WUS 26x range, incl. a share-unlock supply discount) | +5% to +19% |
| Bull | 28% | 127–160 yuan | Rubin ramp + 1.6T optical module volume production + full utilization at Ji'an/Thailand; 2027E attributable profit 4.2–4.6 bn × 26–30x | +46% to +83% |
Probability-weighted expectation 99.4 yuan (+13.9%). The current price sits slightly below the lower bound of the base-case range and within the bear-bull distribution—the odds are positive but not steep: upside (+46% to +83%) and downside (-36% to -58%) are roughly symmetric in magnitude; the main source of the positive skew is the certainty of base-case delivery (order announcements + preannouncements + peer corroboration). Note on exit multiple anchoring: the base 23–26x is anchored to the range formed by Victory Giant's 2027E sell-side 22.4x and WUS's ~26x 2027E—independent of this company's current price—with the upper-middle taken to reflect the company's higher growth but deducted for second-tier status and share-unlock supply; this company's current 2027E trading multiple (18.8–21.5x) is for reference only, not an anchor. The ramp estimate matches the scenarios: 2027E attributable profit midpoint 3.45 bn × 23–26x ÷ 860.26 million shares = 92.2–104.3 yuan, i.e., the base range in the scenario table—same basis, no second coexisting number.
Verdict: fairly valued, modestly cheap (mild undervaluation on the "fairly valued" side)—solid quality (ROE 29.7%, ROIC ~19.3% > cost of equity, cash-to-profit ratio 1.10) coexists with a non-expensive price (2027E base-basis 21.5x, PEG-like ratio in the lower part of the ~0.5–0.9 range), but the absolute valuation position (PB at the 77th percentile) and the supply-gate factor call for expressing bullishness in a "prudent" manner: target range 92–104 yuan; build positions in batches, reassess above 104 yuan (upper bound of base case), revisit assumptions below 75 yuan (approaching the upper bound of the bear range).
1. Industry Size: Global PCB output value rose from USD 73.6 billion in 2024 → USD 85.84 billion in 2025 (+16.7%) → USD 101.95 billion in 2026E (+18.8%) — two consecutive years of double-digit acceleration, the strongest expansion since 2010 (basis: Prismark mid-2026 data, as cited in Shengyi Electronics' 2026 semi-annual report and Victory Giant Technology's 2025 annual report; note that circulating versions of the Prismark report contain two figures for the 2030 total — USD 123.3 billion vs. USD 144.6 billion — indicating high forecast variance; this report cites its growth-rate structure rather than a single point total). Mainland China's 2026E output value is USD 58.78 billion, or 57.7% of the global total. Structural segments are the real story: multilayer boards of 18+ layers grew from USD 5.25 billion in 2025 → USD 9.41 billion in 2026E (+72.8%) → USD 19.76 billion in 2030E (5-year CAGR of 30.3%, the fastest across all categories); HDI 2026E +23.0% (CAGR 13.1%); per Goldman Sachs, the AI server PCB market reaches USD 37.5 billion in 2027E (revised up 38% in August) → USD 84.0 billion in 2028E (+124%). Upstream supply chain: global AI high-speed copper-clad laminate (CCL) from USD 2.2 billion in 2025 (+100%) → USD 3.4 billion in 2026E.
2. Quantified Demand Inflection Chain (the basis for the structural judgment): Drivers — NVIDIA's annual platform refresh (GB200→GB300→Vera Rubin in 2026H2→Rubin Ultra in 2027) and switch generational upgrades (800G→1.6T at commercial scale in 2027→3.2T in 2030), with GPU AI server shipments of roughly 1.71 million units in 2026 (CMBI) and AI server shipments +21%–28% (TrendForce, as cited in the company's semi-annual report); Unit consumption — compute boards from 22 layers to 26+, switch boards from 24 to 32 layers, Rubin Ultra's orthogonal backplane at 78 layers, with rack-level PCB value up approximately 233% (VR200 at roughly USD 116,700 per rack, per media and sell-side extrapolations, unconfirmed by the OEM); high-frequency AI server boards command unit prices 8–12x those of traditional servers; Penetration — AI PCBs' share of global PCB output from ~10% in 2025 → ~25% in 2030E (broker estimates); Share — the company's AI computing product revenue grew +242% in 2025, with global ranking rising from 35th in 2024 to 20th–21st in 2025 (third-party ranking basis; there is a gap versus the company's own claim of "leadership in the AI server PCB segment"; both bases are presented side by side: under the third-party basis, Victory Giant ranks first in AI computing cards/UBB, and WUS ranks in the first tier for high-speed switches, with the company just behind). Judgment: on volume, this is an "industry-wide re-rating" (step-change jumps in layer counts/unit prices, category share rising to a quarter within 5 years, and the 78-layer backplane bringing copper interconnect back to PCBs); but on price, 2027–2028 faces dual mean-reversion pressure from an USD-80-billion-scale capacity expansion wave and decelerating capex growth (Morgan Stanley basis: +14% in 2028). Volume growth is structural; price spreads are cyclical — the two must be priced separately.
3. Supply Chain and Value Distribution: Copper foil (HVLP series, Japanese players hold over 50% share, booked out to 2027) → Low-Dk glass fabric/PPO resin → CCL (M6/M8/M9 grades; EMC ranks first globally, Shengyi Technology first among domestic players with ~13–14% global share) → PCB manufacturing → server ODMs (Foxconn Industrial Internet, Quanta, Wiwynn)/network equipment vendors (Arista, Cisco) → North American hyperscaler AI capex (2026 combined spend of the big four: ~USD 750 billion). The thickest margin sits currently in PCB manufacturing (AI server board gross margins of 35%–50%: GCE AI boards 40–50%, WUS PCB business 40.5% in 2026H1, Shengyi Electronics 34.3%, Shengyi Technology CCL segment 29.2%) — driven by certification barriers + capacity scarcity + smooth pass-through of price hikes. Shengyi Electronics vs. upstream: the controlling shareholder's in-group supply guarantee for high-speed substrates is a scarce positioning amid the 2026 M8-grade shortage cycle (related-party purchases only 16.3% of the total, with diversified external sourcing of main materials), giving it neutral-to-strong bargaining power; vs. downstream: certification by leading platforms plus markedly strengthened bargaining power during the shortage, at the cost of orders and receivables highly concentrated among major customers.
4. Supply-Demand, Competitive Landscape and Cycle Position: Low overall concentration (43 Chinese companies among the global top 100; China accounts for ~36% of output value), but structural concentration at the high-end AI board level — fewer than 10 companies worldwide can reliably mass-produce 18+ layer high-end boards/advanced HDI. The domestic first tier comprises Victory Giant (leading global share in AI computing cards/UBB, Tier-1 joint development with NVIDIA), WUS (first tier in high-speed switches, core in the Google TPU chain), Shennan (substrate + PCB combination), and Shengyi Electronics (AI server accelerator boards/mid-to-high-end switches, a fast riser), with Taiwan's GCE/Unimicron following. Entry barriers: 30+ layer process capability + 5-6 step HDI + mSAP, 1–2 year platform certification cycles at leading customers, M8/M9 material support, capex of RMB 5–18 billion per site, and reliability track record — a five-fold barrier new entrants struggle to replicate. The current landscape shows K-shaped divergence: "high-end shortage + low-end oversupply" — leading players running at 90%+ utilization with core computing substrates in shortage-driven price hikes, while a price war has already emerged in the mid/low end. Cycle position: mid-to-late stage of the upcycle — leading indicator set: ① leading high-end lines at 90%+ utilization (2026-09); ② multiple rounds of 2026 CCL/high-end board price hikes still passing through; ③ order visibility for core computing substrates booked into 2027 (industry media basis); ④ platform iteration milestones (Rubin mass production in 2026H2, 1.6T commercialization in 2027); ⑤ North American cloud capex and AI server shipments (+21–28%). The supply-side clock is the biggest cyclical variable: WUS RMB 17.7 billion, Victory Giant RMB 20 billion, Shennan's Nantong Phase IV, Shengyi Electronics' four-site expansion, and Taiwan players' concentrated buildout in Thailand — the industry Top 40's capex grew +27% in 2025, with capex/revenue at a historical high of 15.6%; the 18–24 month high-end line build cycle means concentrated capacity release in 2027H2–2028, when high-end supply-demand will move toward balance and price-war risk will materialize (acknowledged in the company's own semi-annual report). Unimicron's warning of "possible duplicate ordering" (2026-10-22) is a classic topping signal from inflated order multipliers and warrants monitoring. Regulation: U.S. Section 232 imposes a 25% tariff on advanced computing chips and certain derivatives effective 2026-01-15; whether PCBs are added to the list is the key exposure (exports account for 60.6% of company revenue). The mainstream response is Southeast Asian capacity (the company's Thailand Phase I sprinting toward trial production in 2026Q3); overseas capacity is becoming a prerequisite for high-end orders.
5. Peer Benchmarking Table (company data as of the 2026H1 statements):
| Company | Revenue Scale | Revenue Growth | Gross Margin | ROE | Rank/Share |
|---|---|---|---|---|---|
| WUS Printed Circuit | 2026H1 RMB 13.689 billion | +61.2% | 38.7% | 17.4% (half-year) | Domestic #1 in high-speed switches |
| Victory Giant Technology | 2026H1 RMB 11.629 billion | +28.8% | 33.2% | High (net margin 24.6%) | Global first tier in AI computing cards/UBB |
| Shennan Circuits | First three quarters of 2026: RMB 25.774 billion | +53.8% | 36.9% (PCB business) | 22.2% (weighted, first three quarters) | Top three domestic by overall scale |
| GCE | Jan–Jul 2026: TWD 53.675 billion | +70.3% | 34.8% (Q1) | N/A | Taiwan's core AI server board maker |
| Shengyi Electronics | 2026H1 RMB 5.784 billion | +53.5% | 34.3% | 18.1% (half-year) | Fast-rising second tier (20th–21st) |
6. Industry Positioning of the Company: The fastest slope among the catch-ups — scale is roughly half of WUS/Victory Giant, and it lags Victory Giant by one to two steps at the AI frontier (computing cards/UBB, 36-layer 14-step HDI, CoWoP), but its 2025–2026 revenue slope and profit improvement rank at the front of the first tier, with a clearly upward share trajectory. A three-fold moat: in-group supply guarantee for high-speed CCL from the controlling shareholder (scarce positioning during the M8-grade shortage cycle), 40 years of data-communications process know-how plus platform certification barriers, and the timing match between four-site capacity elasticity and the platform volume ramp window. Key gaps: single-major-customer dependence (42.9%) above peer average, geopolitical exposure under a high share of overseas revenue, and the risk of returns dilution on new capacity when industry supply normalizes in 2027–2028 — the company is both a beneficiary of the high-end shortage and a participant in the USD-80-billion expansion wave.
Overall assessment: Cautiously bullish (confidence 0.58, 12–18 months). Shengyi Electronics is one of the most solid earnings-delivery names in the AI PCB re-rating cycle (FY2025 revenue +102.6%, 2026Q3 guidance +68%–75%, cash flow/net profit ratio 1.10, ROIC ~19.3%), and the current valuation is not stretched (2027E: 18.8x on consensus, 21.5x on the base case, with implied growth supported by the lower bound of demand math); base-case fair value RMB 92–104, probability-weighted expectation RMB 99.4 (+13.9%). Three reasons for caution: the market's 72%-discount pricing of its position cannot be ruled out as justified; the unlock/share-swap supply gate of ~13.9% of the free float around February–April 2027; and the 2028 supply cycle plus capex deceleration. Strategy: accumulate in batches below RMB 87, with position caps constrained by single-customer and single-industry variables; above RMB 104 (upper bound of the base case), the margin of safety is exhausted — gradually take profits; revisit assumptions below RMB 75.
Risks and Counter-evidence: ① Downgrade in North American cloud providers' 2027 capex guidance (to be verified in the November earnings season) — would directly undermine the C1/C3 demand premises; ② gross margin declining more than 1pp QoQ for two consecutive quarters — a confirmation signal of ramp-related depreciation plus price annual declines exceeding expectations; ③ further jump in the top customer's share or signs of order transfer (verify in the 2026 annual report); ④ actual selling pressure from the 2027-03 private placement unlock and EB share swaps; ⑤ upstream material quotas (HVLP4 copper foil, ABF-type materials) constraining new capacity ramps.
(End)