國巨* Yageo (2327.TT) — Company Deep Dive · 公司深度研究 (grounded 2026-09-17)

headlineThe hikes are real — but the margin ramp missed its own bar漲價是真的，但毛利率的爬升速度沒踩到自己設的門檻

Fundamental momentum: 2Q26 revenue NT$44.46bn (QoQ +16.5%, YoY +35.7%, all-time high) ✓; but gross margin rose only 0.4pp QoQ to 38.5% — landing exactly on this analysis's own pre-committed "base case in doubt" trigger of 38.5% or below, not the 41%-or-higher upgrade bar ✓ 07-29. July and August monthly revenue both printed fresh records (NT$16.13bn, NT$16.33bn — the 5th and 6th straight monthly highs), but month-over-month growth slowed from +5.0% to +1.25% ✓ cnyes/工商時報.

基本面動能：2Q26 營收 NT$444.6億（QoQ +16.5%、YoY +35.7%，歷史新高）✓；但毛利率季增僅 0.4 個百分點至 38.5% — 剛好踩在本研究自己事先設定的「基準情境存疑」下緣（38.5% 或以下），不是 41% 以上的升評門檻 ✓ 07-29。7 月、8 月月營收皆創新高（161.31億、163.32億，連續第5、6個月新高），但月增率由 5.0% 降至 1.25% ✓ 鉅亨/工商時報。

Earnings & estimates: 2Q26 EPS 4.59 (record, +88.4% YoY) landed almost exactly on this analysis's own 07-14 estimate of 4.63 — earnings quality held up even though margin pace didn't ✓. This analysis trims 2026/2027 EPS to 18.97/25.9 (from 19.4/26.5) and the target price to 570 (from 583); the 22× multiple is unchanged — this is an estimate cut, not a re-rating ◆.

獲利與估計：2Q26 EPS 4.59（單季新高，YoY +88.4%）幾乎與本研究 07-14 估計的 4.63 相同 — 獲利品質撐住了，即使毛利率爬升速度沒有 ✓。本研究將 2026／2027 EPS 下修為 18.97／25.9（原 19.4／26.5），目標價下修為 570（原 583）；22 倍的評價倍數未變 — 這是估計下修，不是重新評價 ◆。

Positioning & price: the stock closed limit-down twice in a row — NT$563 on 07-28, then NT$507 on the same day as the record-EPS 07-29 earnings release, before partly recovering. Today's NT$529 close is -56.6% off the 7/1 peak of 1,220 but +4.3% off the 07-29 low ✓. Every source covering the crash cites AI-capex-driven chip-sector selling, a (company-denied) pledge-liquidation rumor, and margin-call/settlement-default technical selling — not a fundamentals reassessment ✓. Because the price fell this far, this analysis's rating mechanically moves from Avoid to Range-bound: even leaving the target price at its OLD level of 583, 529 already implies +10.2% upside, above the -10% Avoid cutoff — the rating flip is a function of the price move, not of this update's own estimate trim ◆.

籌碼與價格：股價連兩根跌停 — 07-28 收 563 元，07-29（財報公布當天）再跌停收 507 元，其後部分收復。今日收盤 529 元，距 7/1 高點 1,220 累計 -56.6%，但較 07-29 低點回升 +4.3% ✓。所有報導這波崩跌的來源都指向 AI 資本支出疑慮引發的晶片股系統性賣壓、遭公司否認的大股東質押斷頭傳聞、以及融資斷頭/違約交割的技術性賣壓 — 沒有一個指向基本面轉壞 ✓。正因股價跌了這麼多，本研究觀點機械性地由迴避轉為區間：即使目標價維持舊版的 583 不變，529 元隱含的漲幅也已達 +10.2%，高於「迴避」的 -10% 下緣 — 觀點翻轉是股價下跌的結果，不是本次估計下修造成的 ◆。

Key considerations主要考量

(1) The AI crowd-out effect is the real mechanism behind this hike cycle (Japan/Korea's high-end capacity pivots to AI, and the standard-part gap spills to Yageo/Walsin) — structurally better than 2018's panic-buying repeat. But the other half of the mechanism is equally real: 2027 capacity from Murata/SEMCO/Taiyo Yuden plus China's Sanhuan/风华 — at which point standard-part ASPs carry a 15–25% giveback risk (TrendForce/vocus). (2) Yageo is no longer "an MLCC company": MLCC is only 19% of revenue; magnetics (25.2%) + tantalum (24.3%, world #1 polymer tantalum, >40% capacity, >30% AI-related) are the real body of the business — the market is pricing an "MLCC super-cycle" onto a company that is 68% non-MLCC, overstating pricing elasticity and understating mix resilience. (3) Split-adjusted, 2026E EPS (18.97) is still close to the 2018 cycle-peak level (~20.1), and the premium the market pays for it has compressed a great deal: at 529, the implied multiple is 27.9× vs the 2018 peak's 16× — down from 39× at the 07-14 price of 764, but still above the historical band. This analysis now judges the remaining premium mostly justified by the structural mix shift (premium 30%→75-80%), with less pure momentum left in it than in July (full math in the valuation section).

(1) AI 排擠效應是本輪漲價的真實機制（日韓高階產能轉向 AI，標準品缺口流向國巨/華新科），結構上優於 2018 的恐慌性重複下單；但機制的另一半 — 2027 年 Murata/SEMCO/太誘新產能 + 中國三環/風華擴產 — 同樣真實，屆時標準品 ASP 有 15–25% 回吐風險（TrendForce/vocus）。(2) 國巨已不是 MLCC 公司：MLCC 僅佔營收 19%，磁性 25.2% + 鉭電容 24.3%（聚合物鉭全球 #1、>40% 產能、>30% AI 相關）才是本體 — 市場用「MLCC 超級循環」定價一家 68% 非 MLCC 的公司，高估了漲價彈性、低估了組合韌性。(3) 分割調整後 2026E EPS（18.97）仍接近 2018 循環頂點水準（約20.1），但市場願意付的溢價已大幅收斂：以 529 計算隱含本益比 27.9 倍，遠低於 07-14 時 764 元對應的 39 倍，但仍高於 2018 頂點的 16 倍。本研究現在認為剩下的溢價，多數可用結構性組合改變（特殊品佔比 30%→75-80%）解釋，動能成分比 7 月時已經少了許多（完整推算見估值章節）。

TP:目標價： This analysis models 2026/2027 EPS at NT$18.97/25.9 (revised down from 19.4/26.5 on the 2Q26 margin miss); 12-month target price NT$570 (was 583), based on 2027E EPS × 22× PER. The past three years' PER band runs roughly 10–20× (2018 peak ~16×); 22× is granted because premium mix and AI structural content run higher than history — but below external houses' 26–40×, since that multiple implies 2028 continues to compound, an unproven option in this analysis's view. The multiple is unchanged from 07-14 — this TP cut is 100% an earnings revision, not a re-rating. 本研究預估 2026、2027 年 EPS 18.97、25.9 元（因 2Q26 毛利率不如預期而由 19.4、26.5 下修），12個月目標價 570 元（原 583），基於 2027E EPS × 22 倍 PER；過去三年 PER 區間約 10–20 倍（2018 頂點 ~16 倍），給予 22 倍係因 premium mix 與 AI 結構性成分高於歷史，但低於外資 26–40 倍 — 該倍數隱含 2028 續強，本研究視為未證實選擇權。倍數與 07-14 相同未變 — 這次目標價下修 100% 來自獲利面修正，不是重新評價。 ◆

Risk factors (two-sided at a range-bound view): downside — a 3Q26 GM print again below 38%, or the July-revenue-style "neither trigger fires" pattern repeating through Q4; upside — a 3Q26 GM print clearing 41%, 2H26 direct-supply hikes landing above 40%, or a full Q4'26 shortage (TrendForce's flagged critical window).

風險因子（區間觀點下為雙向）：下行 — 3Q26 毛利率再度低於 38%，或 7 月營收「兩個門檻都沒踩到」的模式在 Q4 重演；上行 — 3Q26 毛利率站上 41%、2H26 直供合約漲價 >40% 落地、或 Q4'26 全面缺貨（TrendForce 關鍵窗口）。

The 8 numbers that decide this決定這篇論點的 8 個數字

529 ✓ 09-17 close — down 56.6% from the 7/1 peak of 1,220 via a two-day limit-down ON the record-earnings day (07-29); still chips coming off, not fundamentals收盤 — 距 7/1 高點 1,220 累計 -56.6%，財報公布當天（07-29）連兩根跌停造成；跌的仍是籌碼，不是基本面

MLCC 19% ✓ of revenue — the market prices the whole company off one product line; that gap is the core of the variant view僅佔營收 — 市場用單一產品線替全公司定價；這個落差＝變異觀點核心

2Q26 GM 38.5%2Q26 毛利率 38.5% ✓ 07-29 landed exactly on this analysis's own "base case in doubt" trigger (≤38.5%), not the ≥41% upgrade bar剛好踩在本研究自己設的「基準情境存疑」門檻（≤38.5%），不是 ≥41% 的升評門檻

TP 570目標價 570 ◆ = 2027E 25.9 × 22× (was 26.5×22=583) — EPS cut, multiple unchanged; historical PER band 10–20×, 22× still above the upper edge= 2027E 25.9 × 22×（原 26.5×22=583）— EPS 下修、倍數不變；歷史 PER 帶 10–20×，22× 仍在上緣之上

External houses' 26–40× assumes 2027/28 EPS +93%/+84% (GS)外資 26–40× 的前提＝2027/28 EPS +93%/+84%（GS） ✓ — an unproven option— 未證實選擇權

Book-to-bill 1.25–1.31B/B 值 1.25–1.31 ✓ 07 — above the 2018 peak (not re-verified this round); the hike cycle premise is unchanged— 高於 2018 峰值（本輪未重新查證）；漲價循環的前提未變

2027 Japan/Korea + China new capacity2027 日韓＋中國新產能 ✓ — 15–25% standard-part ASP giveback risk; the timing is the whole thesis's fulcrum— 標準品 ASP 15–25% 回吐風險；時點是整篇分水嶺

Weighted scenario value 626 vs current 529加權情境值 626 vs 現價 529 ◆ — a +18.3% gap; wider than the TP-implied +7.75% because it includes the bull tail; TP stays the canonical anchor— 缺口 +18.3%，比目標價隱含的 +7.75% 更寬，因為含多方尾部情境；目標價仍是正典錨點

NT$M except per-share (post-split)NT$百萬，每股例外（分割後）	2022	2023	2024	2025	2026E	2027E
Revenue營業收入	~121,300	107,609	121,667	132,930	184,700 ◆	225,300 ◆
Revenue YoY營收 YoY	+5.7%	-11.3%	+13.1%	+9.3%	+38.9% ◆	+22.0% ◆
Gross margin毛利率	38.01%	33.48%	34.36%	36.21%	39.7% ◆	43.5% ◆
Operating margin營業利益率	23.94%	18.99%	19.22%	22.42%	27.5% ◆	30.5% ◆
Net income (parent)稅後純益(母)	~22,900	17,427	19,356	23,634	39,300 ◆	53,700 ◆
EPS (NT$)	11.06	8.48	9.41	11.51	18.97 ◆	25.9 ◆
PER (yr-avg price / EPS)PER(年均價/EPS)	~11×	~13.6×	~13.8×	~13×	27.9× ✓ live	20.4× ✓ live
Dividend (NT$)股息(元)	2.50	5.48*	5.00	6.00	~10 ◆	—
ROE	—	13.54%	10.07%	14.89%	~20% ◆	~24% ◆

Estimate basis — every E-cell, with its reason估計基礎 — 每個 E 欄位的依據 ◆ this analysis本研究估計

Item項目	Assumption假設	Grounds依據
2026E revenue 184,700 (+38.9%)2026E 營收 184,700（+38.9%）	1H26 actual 82,622 ✓ + Jul/Aug actual 32,463 ✓ + Sept26E ~16,822 (+3% MoM) + 4Q26E 52,7501H26 實際 82,622 ✓ ＋ 7/8月實際 32,463 ✓ ＋ 9月估 ~16,822（月增3%）＋ 4Q26E 52,750	two of three 3Q26 months are now disclosed actuals (Jul 16,131/Aug 16,332 ✓); 3Q guidance repeats "utilization above 90%, steady growth, continued margin expansion" — qualitative, no numeric target ✓ 07-293Q26 三個月中已有兩個月為公告實際數（7月16,131／8月16,332 ✓）；3Q 指引重申「稼動率90%以上、營收穩健成長、毛利率營益率續增」— 僅定性，無數字目標 ✓ 07-29
2026E GM 39.7%2026E 毛利率 39.7%	quarterly path 38.5(A) → 40.0 → 41.5%季路徑 38.5(實際) → 40.0 → 41.5%	2Q26 actual GM printed 38.5%, only +0.4pp QoQ ✓ 07-29 — slower than this analysis's own prior +1.2pp/qtr assumption; 3Q/4Q26E ramp trimmed accordingly (institutional-format T4 scorecard: this catalyst resolved as a miss against the ≥41% upgrade bar, hit the ≤38.5% "base case in doubt" trigger) ◆2Q26 毛利率實際僅季增 0.4 個百分點至 38.5% ✓ 07-29 — 比本研究原假設的每季 +1.2pp 慢；3Q／4Q26E 爬升路徑已相應下修（對照本研究預先設定的門檻：未達 ≥41% 升評門檻，踩到 ≤38.5% 的「基準情境存疑」門檻）◆
2027E revenue 225,300 (+22.0%)2027E 營收 225,300（+22.0%）	full-year hike carryover + AI mix >20%, off the new lower 2026E base漲價全年化＋AI 佔比 >20%，以下修後的新 2026E 基期計算	Japan/Korea new capacity only reaches output in 2027 ✓ → the standard-part gap persists at least through 1H27 ◆ (unchanged thesis, no new 2027-specific evidence this round)日韓新產能 2027 年才投產 ✓ → 標準品缺口至少延續至 1H27 ◆（論點未變，本輪無新的 2027 專屬證據）
2027E GM 43.5%2027E 毛利率 43.5%	trimmed from 44.0%; below GS 46% / MS ~50%自 44.0% 下修；低於 GS 46%／MS 約 50%	the 2026 margin-cadence trim (above) propagated forward one year; still assumes hike momentum stalls mid-2027 as Murata/SEMCO/Taiyo Yuden + China capacity lands ✓ plans, no 4Q27 ASP giveback modeled yet ◆上方 2026 毛利率爬升趨緩的判斷順延一年；仍假設 2027 年中漲價動能熄火（村田／三星電機／太陽誘電＋中國產能落地 ✓ 計畫），尚未計入 4Q27 ASP 回吐 ◆

*2023 dividend incl. stock component (pre-split 19.972 cash + 1.948 stock, ÷4). 2022 rev/NI backed out of company USD disclosure + pre-split EPS 44.22 — secondary ⚠.*2023 股息含股票股利（分割前現金 19.972＋股票 1.948，÷4）。2022 營收/純益由公司美元揭露＋分割前 EPS 44.22 反推 — 次級來源 ⚠。

Jargon decoder名詞解碼

Ten terms to read the rest of this page. Basis: Yageo's own investor deck + web-verified industry frameworks.

看懂本文其餘章節需要的十個術語。依據：國巨自家法說簡報 + 上網查證的產業框架。

10 terms — expand when you hit one you don't know十個術語 — 遇到看不懂的名詞再點開

MLCC 積層陶瓷電容 Multilayer Ceramic Capacitor: Hundreds of stacked BaTiO₃ dielectric + electrode layers acting as an instantaneous current buffer at chip power pins (decoupling) plus filtering — "electronics' rice." Why it matters for Yageo: only 19% of revenue, yet the market is pricing the whole company on this one line — this mismatch is the analysis's core variant view. 上百層 BaTiO₃ 介電材料＋電極堆疊而成，作用是在晶片電源腳位做瞬間電流緩衝（去耦）＋濾波 —「電子業的米」。為何重要：僅佔國巨營收 19%，但市場卻用這條產品線替整家公司定價 — 這個落差正是本文變異觀點的核心。
鉭質電容 Tantalum Capacitor: Sintered tantalum-powder anode with a Ta₂O₅ dielectric — highest capacitance per volume, no MLCC-style DC-bias loss, so it carries bulk rail capacitance in dense power circuits. AI-grade polymer tantalum claims 2× energy density. Why it matters: 24.3% of revenue — Yageo's largest single product line via KEMET, world #1 in polymer tantalum. 燒結鉭粉陽極＋Ta₂O₅ 介電層 — 單位體積電容量最高，沒有 MLCC 的直流偏壓損失，因此在密集電源電路中承擔大宗軌道電容。AI 規格聚合物鉭宣稱能量密度是 2 倍。為何重要：佔營收 24.3% — 透過 KEMET 取得的國巨最大單一產品線，全球聚合物鉭 #1。
磁性元件／VRM Voltage Regulator Module: Energy storage in a magnetic field for DC-DC conversion chokes (one per VRM phase) and EMI suppression / LAN transformers. AI servers multiply VRM phase count as power rises (xPU >1,000W vs a phone's 25W = ~40×). Why it matters: 25.2% of revenue — Yageo's single largest product line, and it benefits directly from AI racks. 利用磁場儲能，做 DC-DC 轉換扼流圈（每個 VRM 相數一顆）＋EMI 抑制／區域網路變壓器。AI 伺服器功耗上升會讓 VRM 相數倍增（xPU >1,000W vs 手機 25W ≈ 40 倍）。為何重要：佔營收 25.2% — 國巨最大單一產品線，直接受惠 AI 機櫃。
NTC 熱敏電阻 NTC Thermistor: A temperature-variable resistor — the standard temperature sensor. Thermal management is AI hardware's binding constraint; liquid cooling is growing ~26% CAGR. Why it matters: 13% of revenue, acquired via the Shibaura (芝浦) deal — a new-TAM leg. 電阻值隨溫度變化的感測元件 — 標準溫度感測器。散熱管理是 AI 硬體的關鍵瓶頸；液冷市場約 26% CAGR 成長。為何重要：佔營收 13%，透過收購芝浦(Shibaura)取得，是新的 TAM（市場）題材。
B/B 值 Book-to-Bill Ratio: New orders ÷ shipments booked in the same period; >1 means orders are growing faster than shipments — an early shortage signal. Yageo/Japan-Korea peers are running 1.25–1.31, above the 2018 cycle peak of ~1.25. Why it matters: the key leading indicator for whether this hike cycle is real, not narrative. 新增訂單金額 ÷ 同期出貨金額；>1 代表訂單成長快於出貨 — 是缺貨的早期訊號。國巨與日韓同業目前 1.25–1.31，高於 2018 循環高點約 1.25。為何重要：判斷本輪漲價循環是否為真、而非話術，最關鍵的領先指標。
處置股 Disposition Stock: A TWSE mechanism that flags stocks with abnormal volatility for enhanced margin/settlement scrutiny (shorter settlement windows, cash-only trading). Yageo was on the list 05-26→06-08 after +127.67% in 30 sessions. Why it matters: explains recent positioning turbulence — not a fundamentals signal. 台灣證交所機制，將波動異常的股票列為「處置股」，加強保證金/交割審查（縮短交割期、需款券現款）。國巨在 30 個交易日 +127.67% 後於 05-26→06-08 被處置。為何重要：解釋了近期籌碼面的亂流 — 不是基本面訊號。
違約交割 Settlement Default: When an investor fails to deliver the settlement payment by T+2 after placing an order — a default. Yageo saw two waves within ~10 days: NT$38.8M on 6/30, then NT$41.31M across 13 brokerages. Why it matters: shows retail/margin positioning got overheated — a trigger for the sharp drop, not a sign the fundamentals turned. 投資人下單後未能於 T+2 交割日完成交割款項，構成違約。國巨十日內出現兩波：6/30 先爆 NT$3,880萬，其後再爆跨 13 家券商的 NT$4,131萬。為何重要：顯示散戶/信用交易籌碼過熱 — 是股價急跌的觸發之一，而非基本面轉壞的訊號。
本益比河流圖 PER Band: A stock's price plotted against several constant-multiple lines (its historical PER trading range) over time, to check today's price against past valuation norms. Why it matters: used in this page's valuation section to show 529 still implies ~27.9× on 2026E EPS vs a 10–20× historical range. 把股價與數條「固定本益比」線疊圖，檢視現價相對於歷史估值區間的位置。為何重要：本文估值章節用它證明現價 529 對 2026E EPS 仍隱含約 27.9 倍，高於過去 10–20 倍的歷史區間。
標準品 vs 特殊品 Standard vs Specialty: Standard parts are commoditized, substitutable, low-margin, and easily undercut by Chinese makers; specialty/premium parts (auto, industrial, defense) carry high qualification barriers and margins. Yageo's premium mix is 75–80%. Why it matters: most of the current pricing elasticity comes from the standard-part gap — conflating the two overstates how durable the hike cycle is. 標準品是規格化、可替代、低毛利、易被中國廠削價的品項；特殊品／高階品（車用、工業、國防）認證門檻高、毛利高。國巨特殊品佔比 75–80%。為何重要：目前的漲價彈性主要來自標準品缺口 — 把兩者混為一談，會高估漲價循環的持續性。
股票分割 Stock Split: Yageo's par value split from NT$10 to NT$2.5 (1 share became 4), effective 2025-08-25 — share count ×4, per-share figures (EPS, price) ÷4, while market cap and fundamentals are unchanged. TWSE's asterisk marker on "國巨*" flags exactly this non-NT$10-par status. Why it matters: every figure in this page is post-split-restated; missing this misreads any EPS/price YoY comparison. 國巨面額由 NT$10 分割為 NT$2.5（1股拆4股），2025-08-25 生效 — 股數變 4 倍、每股數字（EPS、股價）除以 4，市值與基本面不變。TWSE 在「國巨*」上加的星號，正是標示這個「非 NT$10 面額」狀態。為何重要：本頁所有數字皆為分割後 restated；忽略此點會誤讀 EPS/股價的 YoY 比較。

The full case — business & technology · bull vs bear · variant perception · management · competition (click to expand)完整論證 — 事業與技術・多空對決・變異認知・經營層・競爭（點開展開）

The business — where the money comes from本業 — 錢從哪裡來

Finding: MLCC is only 19% of revenue. Consensus talks about Yageo as an "MLCC name" riding the passive-component super-cycle. The 1Q26 deck shows a much more diversified book: magnetics leads at 25.2%, tantalum capacitors 24.3%, MLCC third at 19.0%, resistors 13.5%, sensors 13.0%, other 5.0% ✓ deck. Pricing the whole company off the MLCC headline cuts both ways — it overstates blended ASP leverage on the way up, and understates the floor mix diversification provides on the way down.

發現：MLCC 只佔營收 19%。市場共識把國巨當成「MLCC 概念股」，搭被動元件超級循環的順風車。但 1Q26 法說簡報顯示組合遠比想像分散：磁性元件居冠 25.2%、鉭質電容 24.3%、MLCC 排第三 19.0%、電阻 13.5%、傳感器 13.0%、其他 5.0% ✓ deck。用 MLCC 的漲價頭條替全公司定價，其實兩面都會出錯 — 上漲時高估了混合 ASP 的槓桿，下跌時又低估了組合分散帶來的底部支撐。

✓ deck 1Q26 segment mix. AI overlay ≈15% of revenue (up from 4% in <2 years, ~US$600M annualized); channel: distribution 44.6 / OEM 35.0 / EMS 20.4. Forward mix ◆: AI >20% of revenue by 4Q26E; premium/specialty mix 75–80%.1Q26 分部占比。AI overlay 約營收 15%（<2年內從4%上升，約當年化 US$600M）；通路：經銷 44.6／OEM 35.0／EMS 20.4。前瞻組合 ◆：AI 佔比 4Q26E 前達 >20%；特殊品/premium 佔比 75–80%。

✓ deck Auto is only 17.1% of revenue — TDK's 1Q26 auto passives were down 3.4%, so auto is not currently a hot end-market ⚠; the hike story is running through industrial/compute/AI, not autos.車用僅佔營收 17.1% — TDK 1Q26 車用被動元件下滑 3.4%，車用目前並非熱門終端市場 ⚠；漲價故事跑在工業/企業運算/AI，不是車用。

Unit economics單位經濟

~1.1 trillion components/year ÷ ~US$4.26B annual revenue (FY25's NT$132,930M) ≈ US$0.004/unit blended ✓ deck + computed. Mix business: premium parts (auto/industrial/medical/defense/AI) carry ASPs 10–100× commodity. The gross-margin gap vs Murata narrowed slightly to ~3.5pp by 2Q26 (38.5% vs ~42%) ✓ 07-29.

約 1.1 兆顆／年 ÷ 年營收約 US$4.26B（FY25 NT$132,930M 換算）≈ US$0.004／顆（blended）✓ deck + computed。組合生意：特殊品（車用/工業/醫療/國防/AI）ASP 是商規品的 10–100 倍。與 Murata 毛利率差距至 2Q26 略收斂至約 3.5 個百分點（38.5% vs ~42%）✓ 07-29。

90% BOM coverage90% BOM 覆蓋率

Company claim: Yageo parts sit inside "anything that plugs in, switches on, or runs on a battery." ⚠ co. That breadth is why the product-mix chart above matters more than any single line item — the company's fate rides on the blend, not on MLCC alone.

公司宣稱：任何「可插電、可開關、或使用電池」的裝置，都用得到國巨的零件 ⚠ 公司宣稱。正因為這種廣度，上方產品組合圖比任何單一品項都更重要 — 公司命運繫於整體組合，而非 MLCC 單一線。

The AI crowd-out mechanism — why prices are risingAI 排擠機制 — 為什麼會漲價

This is the structural×cyclical hybrid mechanism behind 2026's hikes. AI rack content roughly doubles generation-over-generation (Rubin vs GB200); Japan/Korea's premier makers redirect high-end capacity to AI parts, running utilization 90–95% with book-to-bill 1.25–1.31; that pulls standard-part capacity away from the market, leaving a gap that channel inventory (down to 1–1.5 months vs a 4.5-month norm) can't absorb — hence the 7/1 price hikes of 30–80%. Counter-force — 2027 Japan/Korea + China new capacity and a 15–25% ASP giveback risk — fully argued in the Bear case section below.

這是本輪 2026 年漲價背後「結構性×循環性」混合的機制。AI 機櫃含量每個世代大約翻倍（Rubin vs GB200）；日韓龍頭廠把高階產能轉向 AI 相關產品，稼動率衝上 90-95%、B/B 值 1.25-1.31；這把標準品產能抽走，留下的缺口是通路庫存（已降到 1-1.5 個月，正常水準 4.5 個月）填不了的 — 因此才有 7/1 的 30-80% 漲價。反制力量 — 2027 年日韓＋中國新產能與 15–25% ASP 回吐風險 — 完整論證見空方論點章節。

The AI crowd-out mechanism (top row, ✓ sourced per grounding table) and the counter-force that eventually offsets it (bottom row). Both halves are real; the disagreement in this analysis vs external houses is about timing, not direction.AI 排擠機制（上排，✓ 均有來源，詳資料來源章節）與最終抵銷它的反制力量（下排）。兩邊都是真的；本研究與外資的分歧在於時間點，而非方向。

Bull case多方論點

Real, early-mid-cycle shortage真實且處於中前段的缺貨循環

Book-to-bill exceeds the 2018 peak; channel inventory sits at a decade low; TrendForce's Q4'26 escalation window is still ahead — sequential blowout prints are nearly certain: this analysis's own model runs EPS 3.90→4.63→5.05→5.85. Is this in guidance? Only partially — the company guided just "moderate" for Q2, and the July hike's impact isn't in any company guidance yet. ✓

B/B 值超越 2018 高點；通路庫存在十年低位；TrendForce 的 4Q26 升級窗口還在前方 — 連續的亮眼季報幾乎是確定的：本研究自己的模型就是 EPS 3.90→4.63→5.05→5.85。這寫進公司指引了嗎？只有一部分 — 公司對 Q2 僅指引「小幅季增」，7 月漲價的影響也尚未進入任何公司指引。 ✓

Structural mix結構性組合

Premium mix 75–80%, MLCC only 19% of revenue, tantalum #1 + >30% AI-related, plus a sensor leg. Trough gross margin is now proven higher (33.5% in 2023 vs ~25% pre-2018). ✓

特殊品佔比 75–80%、MLCC 僅佔營收 19%、聚合物鉭全球 #1 且 >30% 與 AI 相關，外加傳感器一條腿。谷底毛利率已被證明墊高（2023 年 33.5% vs 2018 前約 25%）。 ✓

AI content roadmapAI 含量成長藍圖

Rack MLCC dollar content +182% from GB300 to VR200; TDK is guiding 10× growth in AI passives; Yageo's own AI revenue went 4%→15% in under two years. The structural leg is genuinely early. ✓

機櫃 MLCC 金額含量自 GB300 到 VR200 +182%；TDK 對 AI 被動元件喊出 10 倍成長；國巨自身 AI 營收在兩年內從 4% 升到 15%。結構性這條腿確實還在早期。 ✓

Balance sheet + Chen's M&A machine資產負債表＋陳家併購機器

FCF NT$7.75bn/quarter, net debt/EBITDA 0.33×, twA+ stable rating. Chairman Chen's serial-consolidator track record (won the Shibaura bidding war against Minebea; 茂達 platform next) is real optionality. ✓

FCF 每季 NT$77.5 億、淨金融負債/EBITDA 0.33×、twA+ 展望穩定。陳泰銘的連續併購紀錄（芝浦競購戰勝出美蓓亞；下一步茂達平台）是真實的選擇權。 ✓

Unproven optionality未證實選擇權: 2028 EPS 43–57 (external houses), a possible liquid-cooling M&A, and 茂達 power-IC platform synergy — none of these are in this analysis's base case.

未證實選擇權：2028 EPS 43–57（外資）、可能的液冷併購、茂達平台綜效 — 皆不在本研究的基準情境中。

Bear case空方論點

Multiple-compression risk dominates倍數壓縮風險主導

Down to ~27.9× at 529 on what may be cycle-peak-level EPS, vs a 10–20× history — the multiple has compressed a lot since 07-14's ~39×, but is still above the historical band. In 2018, the multiple compressed before peak EPS printed — the stock peaked Jul-2018, but peak quarterly EPS came later; the 2Q26 GM miss (38.5% vs the 41% upgrade bar) is at least consistent with margins topping out earlier than the bull case assumes. Two limit-down sessions on the record-earnings day (07-28/07-29), plus the earlier settlement defaults and disposition-list history (+127.67%/30 sessions in May-June), show the shareholder register is still momentum-heavy (450,000 retail holders as of the 07-14 build). ✓

529 對可能已是循環頂點等級的 EPS 已降至約 27.9 倍，vs 歷史區間 10–20 倍 — 較 07-14 的約 39 倍已大幅壓縮，但仍高於歷史區間。2018 年，倍數在 EPS 見頂之前就先壓縮 — 股價 2018 年 7 月見頂，單季 EPS 高點卻在其後才出現；2Q26 毛利率不如預期（38.5%，未達 41% 升評門檻）至少與「毛利率比多方情境假設更早見頂」的說法一致。財報公布當天連兩根跌停（07-28/07-29），加上先前的違約交割與處置股紀錄（5-6月30個交易日 +127.67%），顯示股東結構仍高度動能化（散戶約 45 萬戶，07-14 版數字）。 ✓

The 2027 capacity wave2027 產能波

Murata's ¥80bn server-MLCC plant, SEMCO's two plants, and Taiyo Yuden Korea all land in 2027; Chinese makers add 4.5B+ pieces/month; TrendForce flags a 15–25% standard-ASP giveback risk for '27-28. ✓

村田 ¥800 億伺服器 MLCC 廠、三星電機兩座新廠、太陽誘電韓國廠全數在 2027 年落地；中國廠商月增 45 億顆以上；TrendForce 標註 '27-28 標準品 ASP 有 15–25% 回吐風險。 ✓

Demand quality需求品質

Consumer units are shrinking (phones -13.9%, PCs -10.4%); OEM pre-buying ahead of the hikes is classic double-ordering fuel; TDK's auto passives OP fell 54% YoY, showing auto is soft. If AI capex plateaus, the crowd-out mechanism reverses fast. ✓

消費性出貨持續萎縮（手機 -13.9%、PC -10.4%）；OEM 在漲價前的預先拉貨是典型的重複下單燃料；TDK 車用被動元件營益 YoY -54%，顯示車用疲軟。若 AI 資本支出趨於平緩，排擠機制會快速逆轉。 ✓

What bulls must be right about多方必須賭對的假設

Hikes must stick through 2027; AI rack units must keep compounding; China must stay locked out of premium sockets; no macro shock (IDC already cites the US-Iran conflict and the memory-cost crisis cutting device demand).

漲價必須撐過 2027；AI 機櫃出貨必須持續複合成長；中國廠必須持續被擋在高階插槽之外；不能有總體衝擊（IDC 已點名美伊衝突與記憶體成本危機正在削減裝置需求）。

Variant perception變異觀點

Consensus (post-rerating) reads Yageo as an MLCC super-cycle name, the strongest in 20 years — buy the leader: FactSet's surveyed consensus TP is 1,275, with individual houses at 1,490–1,515 (GS/Citi/MS). The market may be roughly right about 2026-27 earnings direction — no variant there. The variant is narrower and sits in what the market is overweighting vs what it's ignoring.

市場共識（重新評價後）把國巨讀成「MLCC 超級循環股，20年最強」— 買龍頭：FactSet 調查共識目標價 1,275，外資個別喊到 1,490–1,515（GS／Citi／MS）。市場對 2026-27 獲利方向大致可能是對的 — 那裡沒有變異觀點。變異之處更窄，在於市場過度加權了什麼、又忽略了什麼。

This is not a fight over 2026-27 earnings direction — it's about which slice of the P&L the market is using to price the whole company, and whether 529 leaves any room for anything beyond a clean 2027. It leaves more room than 764 did.這不是在爭 2026-27 的獲利方向 — 而是市場用哪一塊損益表在替全公司定價，以及 529 這個價位，是否還留有任何「乾淨 2027 以外」的空間。它留的空間比 764 時更多。

Honest reverse variant: if Q4'26 tips into a full shortage, consensus 2027 estimates go higher, and 529 will have been the entry point. That is the 30%-weight bull scenario in this analysis's own framework (§ scenarios below) — not the base case, but not dismissed either. This is the same variant the page has carried since 07-13; the price has simply moved a long way toward where it would have to be for the reverse variant to look prescient.

誠實的反向變異觀點：若 2026 第四季真的轉為全面缺貨，共識 2027 估計將上修，529 屆時就會是進場點。這正是本研究自己框架中權重 30% 的多方情境（詳下方機率加權情境）— 不是基準情境，但也不能被忽視。這與 07-13 起就存在的反向變異觀點相同；只是股價已經朝著「反向變異觀點會顯得有先見之明」所需的位置，走了一大段路。

Management & capital allocation經營層與資本配置

Founder-chairman Pierre Chen (陳泰銘) is THE serial consolidator in Taiwan's passive-components industry — the acquisitions below are the through-line of the whole company strategy.

創辦人暨董事長陳泰銘，是台灣被動元件產業「連續併購」的代表人物 — 下方的收購紀錄，就是整個公司策略的主軸。

✓ 2000 Philips → 2018 Pulse (US$740M) → 2020 KEMET (~US$1.8B, incl. a US$195M personal sub-loan) → 2022 Chilisin (all-stock) → 2023 Telemecanique (€723M) + Nexensos (€79.4M) → 2025 Shibaura (¥110bn, won a 4-raise bidding war vs Minebea) → 2026 chairs taken at 茂達(6138)/大中(6435), a power-IC/MOSFET adjacency. KEMET's integration verifiably re-based specialty mix from 30%→75%.2000 Philips → 2018 Pulse(US$740M) → 2020 KEMET(~US$1.8B，內含個人 US$195M 次順位貸款) → 2022 奇力新(全股票換股) → 2023 Telemecanique(€723M)+Nexensos(€79.4M) → 2025 芝浦(¥1,100億，贏過美蓓亞的4次加碼競購戰) → 2026 拿下茂達(6138)/大中(6435)董座，切入功率IC/MOSFET鄰接領域。KEMET 整合已驗證把特殊品組合由30%墊高到75%。

✓ Final ¥7,130/share, ~¥110bn total, 87.3% stake, delisted 2026-01-13, 100% owned by 2026-01-15 — dated multi-source timeline.最終 ¥7,130/股，總額約¥1,100億，取得87.3%股權，2026-01-13下市，2026-01-15達100%持股 — 多來源標註日期的時間軸。

Ownership持股結構

Foreign institutions hold 48.58% (07-09; May-08 peak 53.19%, trimmed ~4.6pp into the correction) ⚠ not re-verified this round本輪未重新查證; directors/supervisors ~7.6% ✓. The Chen family holds ~6.8% via the 陳氏傳承/泰銘傳承 succession vehicles (Oct-24/Mar-25 transfers = estate planning, NOT open-market sales) ⚠ derived; daughter 陳少喬 (Wharton, ex-UBS) joined the board in 2025 ✓. No insider open-market selling found in 2025-26 ✓. Retail register is heavy: ~450k shareholders after the run ✓.

外資持股 48.58%（07-09；5/8 高點 53.19%，其後調節約 4.6 個百分點）⚠ 本輪未重新查證；董監約 7.6% ✓。陳家透過「陳氏傳承／泰銘傳承」控股公司持有約 6.8%（2024/10、2025/03 的移轉屬接班規劃，並非市場賣出）⚠ 推算；女兒陳少喬（Wharton、前 UBS）2025 年進入董事會 ✓。2025-26 未發現內部人公開市場賣股 ✓。散戶籌碼沉重：大漲後股東人數約 45 萬 ✓。

Dividend & leverage配息與槓桿

FY2025 cash dividend NT$5.998 (~6.00), payout 52%, paid 2026-07-03 ✓; policy floor: cash ≥50% of any distribution ✓ annual report. Yield at 529 = 1.13% (was 0.79% at 764) ✓ computed. Leverage: net financial debt/EBITDA 0.33× (company, post-Shibaura peak 0.59×) ✓ deck vs ~1.5× gross projected by Taiwan Ratings (twA+/stable — definitions differ ⚠). 2020 GDR raised US$650M for KEMET; no equity raise since; no live buyback found (verify on MOPS ⚠). Shibaura was funded with cash + debt, no new shares ✓.

FY2025 現金股利 NT$5.998（約 6 元），配發率 52%，2026-07-03 發放 ✓；政策下限：現金 ≥ 股利總額 50% ✓ 年報。以 529 計殖利率 1.13%（764 時為 0.79%）✓ 計算。槓桿：淨金融負債／EBITDA 0.33×（公司口徑，併芝浦後高點 0.59×）✓ 法說，vs Taiwan Ratings 預估總負債口徑約 1.5×（twA+／穩定 — 兩者定義不同 ⚠）。2020 年為併 KEMET 發 GDR 募資 US$650M；其後無增資；未發現進行中庫藏股（建議 MOPS 再核 ⚠）。芝浦以現金＋借款支應，未發新股 ✓。

Competitive position競爭地位

China exposure: 21 of 65 plants are in China; Greater China customer mix is 24% (down from ~70% pre-2018 — this is the asymmetry the 2018 comparison misses).

中國曝險：65座廠中21座在中國；大中華客戶佔比24%（相較 2018 前約 70% — 這正是「2018 對比」容易忽略的不對稱之處）。

Product / segment產品/分部	Share claim市占宣稱	Verdict判定
Resistors電阻	#1, >30% capacity產能	⚠ company公司宣稱
MLCC	#3 by capacity >15% (company) vs #5 by revenue 7-9% (aggregators)按產能#3>15%(公司) vs 按營收#5 7-9%(aggregator)	⚠ conflict — both shown來源衝突 — 兩者並陳
Polymer tantalum聚合物鉭	#1, >40% capacity產能	⚠ company公司宣稱
Network transformers網路變壓器	#1	⚠ company公司宣稱
NTC (Shibaura)(芝浦)	#1, 13.5–20%+ (sources conflict)13.5-20%+(來源不一)	⚠ conflict來源衝突

✓ Murata GM ~41-42% vs Yageo 38.1% — the gap this analysis's unit-economics section already flags; Murata remains "the hardest high-end."Murata 毛利率約41-42% vs 國巨38.1% — 本文單位經濟章節已提及的差距；Murata 仍是「最難打的高階」。

Player (ticker)玩家（代號）	Where they're strong強在哪	vs Yageo與國巨的關係
Murata (6981.T)村田製作所 Murata（6981.T）	Hardest high-end MLCC; ¥250bn capex incl. ¥80bn server-MLCC最難打的高階MLCC；資本支出¥2,500億、含¥800億伺服器MLCC	GM ~42% ✓；2027 capacity = bear catalyst2027新產能=空方催化劑
Samsung Electro-Mechanics (009150.KS)三星電機 SEMCO（009150.KS）	₩454B AI-MLCC order; Philippines/Tianjin plants 2027₩4,540億AI-MLCC訂單；菲律賓/天津廠2027	OP +93% YoY — same cycle, higher tier營益YoY +93% — 同循環、更高階 ✓
Taiyo Yuden (6976.T)太陽誘電 Taiyo Yuden（6976.T）	High-end MLCC; new Korea plant not done until 2027高階MLCC；韓國新廠2027年才完工	hiked +6–13% Apr; part of the 2027 capacity wave4月漲價6–13%；2027產能波的一員 ✓
TDK (6762.T)TDK（6762.T）	Broad passives; guiding 10× AI data-center passive sales被動元件廣度；喊AI資料中心被動元件營收10倍	auto passives weak (OP -54% YoY) — echoes Yageo's soft 17.1% auto book車用疲弱（營益YoY -54%）— 呼應國巨17.1%車用部位偏軟 ✓
Walsin Technology (2492.TT)華新科 Walsin（2492.TT）	TW peer; MLCC mix 46.4% of revenue (1Q26)台股同業；MLCC占營收46.4%（1Q26）	higher MLCC beta than Yageo (19%) — a cleaner pure-play echo / leading indicatorMLCC彈性比國巨（19%）更大 — 更純的族群風向球/領先指標 ✓
Sanhuan (300408.SZ) + Fenghua (000636.SZ)三環集團（300408.SZ）＋風華高科（000636.SZ）	Standard-only, <8-10% combined share; +4.5B pcs/mo capacity by 2027僅標準品，合計市占<8-10%；2027年前月產能+45億顆	15-25% underbid; "de-Yageo" qualification attempts post-hike; the 2027-28 ASP-giveback source削價15-25%；漲價後有「去國巨化」認證嘗試；2027-28 ASP回吐的來源 ⚠
TW sector sympathy: 禾伸堂 Holy Stone (3026.TT), 蜜望實 (8043.TT)台股同族群外溢：禾伸堂（3026.TT）、蜜望實（8043.TT）	Rode the same rerating — multiple limit-ups on 6/15 after Citi's upgrade wave同一波重估 — 6/15 Citi升評潮後族群多檔漲停 ✓	sector beta, not direct competitors — gauges how much of the move is theme money族群 beta，非直接競爭 — 可衡量行情中有多少是題材資金

Forecast model財測模型

Coverage-lite quarterly model — the company gives only qualitative guidance ("small sequential increase"), so every E-cell below is this analysis's estimate, basis stated in the table caption.

Coverage-lite 季度模型 — 公司僅給定性指引（「小幅季增」），因此下表每個 E 欄位皆為本研究估計，依據列於表格說明。

	1Q26A	2Q26A	3Q26E	4Q26E	1Q27E	2Q27E
Revenue營收	38,166	44,456 ✓	49,300 ◆	52,750 ◆	51,500 ◆	54,000 ◆
Gross margin毛利率	38.1%	38.5% ✓	40.0% ◆	41.5% ◆	43.0% ◆	43.5% ◆
Operating margin營益率	25.2%	26.4% ✓	28.0% ◆	29.5% ◆	30.0% ◆	30.5% ◆
EPS (NT$)	3.90	4.59 ✓	4.90 ◆	5.58 ◆	5.95 ◆	6.35 ◆

2Q26 is now actual, not estimated ✓ 經濟日報/東森財經/leveragetw · 07-29 — revenue 44,456 (record), gross margin 38.5% (only +0.4pp QoQ), operating margin 26.4%, EPS 4.59 (record, +88.4% YoY), net income NT$9,418M. The GM print landed exactly on this analysis's own pre-committed "base case in doubt" trigger (≤38.5%) from the 07-14 build — not the ≥41%/EPS≥5.0 upgrade trigger. 3Q26E/4Q26E revised down from 50,000/41.5%/28.5%/5.05 and 53,500/43.5%/30.5%/5.85: two of 3Q26's three months are now disclosed actuals (Jul 16,131 ✓ + Aug 16,332 ✓), and the GM ramp is trimmed to the observed +1.5pp/quarter pace (was assumed +3pp/quarter). Management's 3Q26 guidance stays qualitative — utilization above 90%, steady revenue growth, continued margin expansion — with no numeric target ✓ 07-29. 1Q27E/2Q27E carry the same trim forward with no new 2027-specific information.2Q26 現為實際數，不再是估計 ✓ 經濟日報/東森財經/leveragetw · 07-29 — 營收 44,456（新高）、毛利率 38.5%（僅季增0.4個百分點）、營益率 26.4%、EPS 4.59（新高，YoY +88.4%）、稅後純益 NT$94.18億。毛利率剛好踩在本研究 07-14 版事先設定的「基準情境存疑」門檻（≤38.5%），不是 ≥41%／EPS≥5.0 的升評門檻。3Q26E／4Q26E 由原 50,000／41.5%／28.5%／5.05 與 53,500／43.5%／30.5%／5.85 下修：3Q26 三個月中已有兩個月為公告實際數（7月16,131 ✓ ＋ 8月16,332 ✓），毛利率爬升速度下修至實際觀察到的每季 +1.5 個百分點（原假設每季 +3 個百分點）。公司 3Q26 指引仍為定性描述 — 稼動率90%以上、營收穩健成長、毛利率營益率續增 — 無數字目標 ✓ 07-29。1Q27E／2Q27E 沿用同一下修幅度順延，無新的 2027 專屬資訊。

Scorecard — 2Q26 actual vs this analysis's own prior estimate成績單 — 2Q26 實際數 vs 本研究前估

	Actual實際	This analysis's 07-14 estimate本研究07-14前估	Diff差異	Market survey (ftnn, 07)市場調查(ftnn,07)	Diff差異
Revenue營收	44,456 ✓	44,456*	0%	—	—
Gross margin毛利率	38.5% ✓	39.3%	-0.8pp	39.1%	-0.6pp
EPS (NT$)	4.59 ✓	4.63	-0.9%	4.85	-5.4%

*07-14 revenue "estimate" was the company's own self-reported (自結) actual, already ✓ — the miss is entirely in margin/EPS. This analysis's own estimate missed by less than the market survey's did, on both lines — small consolation on a print that still triggered the pre-committed "base case in doubt" reaction (see catalysts).*07-14 的營收「估計」本就是公司自結實際數，已為 ✓ — 落差全在毛利率／EPS。本研究自己的估計，兩項都比市場調查的估計更接近實際 — 但這只是小小的安慰，這個數字仍觸發了預先設定的「基準情境存疑」反應（見催化劑章節）。

This analysis vs the market本研究 vs 市場

No free TW consensus feed exists for Yageo, so most rows below stay named, dated, single-house estimates — except 2026E, where TradingView's analyst-count consensus (17 analysts, currency-corrected, see §sources) now gives a genuine aggregate to compare against.

國巨沒有免費的台股共識數據源可用，下表大多數列仍是有名有姓、標註日期的單一機構估計 — 除了 2026E 一列，TradingView 的分析師人數共識（17位，經匯率換算，詳見資料來源章節）現在提供了真正的彙整數據可供比對。

	2026E EPS2026E EPS	2027E EPS2027E EPS	2028E EPS	TP	Date日期
This analysis (adj.)本研究(調整後) ◆	18.97 (was前次 19.4)	25.9 (was前次 26.5)	not modeled不建模	570 (was前次 583)	09-17
TradingView consensusTradingView 共識 ⚠ currency-corrected已換匯修正	≈19.86 (17 analysts)(17位分析師)	—	—	excluded (mismatch flag)排除(比對異常標記)	09-17
FactSet survey consensusFactSet 調查共識 ✓ not re-verified this round本輪未重新查證	19.59 median (16.98–21.22)中位數 19.59（16.98–21.22）	—	—	1,275	07 (via ftnn)07（ftnn 轉述）
Goldman Sachs ⚠ not re-verified this round本輪未重新查證	21.2	37.6	57.3	1,490 (26×'28E)	07-06
Citi ⚠ not re-verified this round本輪未重新查證	17.81	31.42	43.24	1,500 (40× PE)	06-09
Morgan Stanley ⚠ not re-verified this round本輪未重新查證	—	bull 45多方情境 45	bull 60多方情境 60	1,515	06-24
宏遠證券 ⚠ not re-verified this round本輪未重新查證	18.69	—	—	1,230	07-07
Unnamed US bank未具名美系 ⚠ not re-verified this round本輪未重新查證	18.2	30.7	42.6	—	06-15

The once-negligible 2026 gap has reopened slightly. In July this analysis's 19.4 sat almost exactly on FactSet's 19.59. Now, after the 2Q26-driven trim, this analysis's 18.97 runs about 4.5% below TradingView's Sept consensus of ≈19.86 (itself only +1.4% above the July FactSet figure — consensus has essentially stopped moving, a sharp slowdown from the +25%/month pace seen through June-July). The 2027-28 divergence is unchanged in kind: external houses (last verified in June-July) assumed hikes of +93%/'27 and +84%/'28 (GS); this analysis still assumes hike momentum stalls mid-2027 as capacity lands, now from a lower 2026 base.

曾經幾乎消失的 2026 分歧，現在又小幅打開了。7 月時本研究的 19.4 幾乎等於 FactSet 的 19.59；經 2Q26 下修後，本研究現在的 18.97 比 TradingView 9 月共識約 19.86 低約 4.5%（該共識本身也只比 7 月 FactSet 高 1.4% — 共識幾乎已經停止上修，比 6-7 月每月 +25% 的速度明顯放緩）。2027-28 的分歧性質未變：外資（最後查證於 6-7 月）假設漲價 +93%/'27、+84%/'28（GS）；本研究仍假設 2027 年中隨產能落地動能熄火，只是現在從更低的 2026 基期起算。

Price-hike log漲價紀錄

Date日期	Event事件	Magnitude漲幅
2026-02	Resistor (R-chip) price hike電阻(R-chip)漲價	+15–20%
2026-03	KEMET tantalum, 3rd hikeKEMET 鉭質電容第三次漲價	— ⚠ magnitude undisclosed未揭露幅度
2026-04	Taiyo Yuden (6976.T) hike太陽誘電 Taiyo Yuden (6976.T) 漲價	+6–13%
2026-07-01	Yageo full capacitor-line hike, first incl. direct OEM/EMS — capacitors ≈40%+ of revenue國巨全電容產品線漲價，首次含直供OEM/EMS — 電容合計約佔營收4成以上	+30–80%

Valuation & target price估值與目標價

At 529 (2026-09-17 close): 35.6× TTM, 27.9× 2026E (this analysis), 20.4× 2027E (this analysis) — down sharply from 60×/39×/28.8× at the 07-14 build's 764, purely on price (this analysis's EPS estimates moved only slightly). Against the last named external estimates (June-July, not re-verified this round), 529 is only ~15.8× Goldman's 2027E EPS of 37.6 and ~17.2× Citi's 31.42 — the multiple compression the bear case was calling for has now happened, at least against the external EPS path. What's priced in: the 2026 hike cycle to date (including the 2Q26 margin miss), a still-modest 2027E EPS path. What's not priced in (either direction): whether 3Q26 GM clears 41% or slips again, the 2027 capacity-wave timing, and consumer pull-in unwind.

在 529（2026-09-17收盤）：TTM 35.6×、2026E(本研究) 27.9×、2027E(本研究) 20.4× — 相較 07-14 建置時 764 元對應的 60×／39×／28.8× 大幅下滑，主因純粹是股價（本研究的 EPS 估計僅小幅調整）。對照最後一次查證的外資估計（6-7月，本輪未重新查證），529 元對 Goldman 2027E EPS 37.6 僅約 15.8 倍、對 Citi 的 31.42 僅約 17.2 倍 — 空方論點原本期待的倍數壓縮，至少對外資的獲利路徑而言已經發生。已反映：2026 漲價循環至今（包含 2Q26 的毛利率不如預期）、仍偏保守的 2027E EPS 路徑。未反映（不論方向）：3Q26 毛利率能否站上 41% 或再度走弱、2027 產能波的時點、消費性拉貨提前透支的回吐。

Actual share price (post-split, IBKR weekly approx.)實際股價（分割後，IBKR週線約值） 22× — this analysis's TP multiple22× — 本研究目標倍數 16× — 2018 cycle-peak multiple16× — 2018 循環頂點倍數 10× — trough-cycle multiple10× — 循環谷底倍數

◆ Historical PER band (computed, IBKR weekly + grounded EPS): 2023 ~13.6× avg / 2024 ~13.8× / 2025 ~13× (range roughly 10–20× through-cycle; 2018 peak ~16×). The price line's spike to 1,045 (2026Q2) crashed through 816→764 by 07-14, then further to a 507 limit-down low on 07-29 (the SAME DAY as the record 2Q26 earnings release), before closing at 529 on 09-17 — still above every drawn multiple line, but the gap has narrowed sharply since 07-14.歷史 PER 河流圖（計算所得，IBKR 週線＋已查證 EPS）：2023 均值約13.6×／2024約13.8×／2025約13×（循環區間約10–20倍；2018頂點約16倍）。股價線衝上1,045（2026Q2）後崩落穿越816→764（07-14），07-29（2Q26 創新高財報公布當天）再跌停至507低點，09-17 收在529 — 仍高於圖上所有倍數線，但缺口較07-14時已明顯收斂。

TP derivation目標價推導 ◆

This analysis models 2026/2027 EPS at NT$18.97/25.9 (revised down from 19.4/26.5 on the 2Q26 margin miss); 12-month target price NT$570 (was 583), based on 2027E EPS × 22× PER; the past three years' PER band runs roughly 10–20× (2018 peak ~16×). 22× is granted — above the historical range's upper edge — because premium mix (75–80% structural) and AI structural content lift the earnings floor above history; but this analysis rejects external houses' 26–40×, since that multiple implies 2028 EPS keeps compounding 50%+, an unproven option, and this stock's history shows the multiple compresses before the earnings peak at every cycle turn. The multiple is unchanged from 07-14 — the TP moved on EPS alone.

本研究預估 2026、2027 年 EPS 為 18.97、25.9 元（因 2Q26 毛利率不如預期而由 19.4、26.5 下修），12個月目標價 570 元（原 583），基於 2027E EPS × 22 倍 PER；過去三年 PER 約 10–20 倍（2018 頂點 ~16 倍）。給予 22 倍（高於區間上緣）係因：premium mix 75-80% 結構性墊高獲利底部 + AI 為多年期需求 — 但拒絕外資 26-40 倍，因該倍數隱含 2028 EPS 續增 50%+，屬未證實選擇權，且本股歷史上循環轉折時倍數皆先於獲利頂點壓縮。倍數與 07-14 相同未變 — 目標價的變動全來自 EPS。

◆ Anchor reconciliation: the TP (25.9 × 22× = 569.8, rounded 570) IS the base-case scenario price — one multiple, one canonical number. The probability-weighted 626 (was 632) is the cross-check including the 20/30% tails, not the target.

◆ 錨點一致化：目標價（25.9 × 22× = 569.8，取整570）即基準情境價位 — 同一倍數、同一個正典數字。機率加權值 626（原632）是含 20%/30% 尾部情境的交叉檢核，不是目標價。

Sensitivity — 2027E EPS × PER → implied price敏感度 — 2027E EPS × 倍數 → 隱含價位 ◆

2027E EPS	14× (bear)(空方)	16× (2018 peak)(2018峰值)	22× (TP)(目標價)	26× (street)(外資)
20	280	320	440	520
25.9	363	414	570 (TP)(目標價)	673
36.5	511	584	803	949

Current 529 = 20.4× base EPS 25.9 (between the 2018-peak and TP columns) or 14.5× bull EPS 36.5 (BELOW the 2018-peak column) — a real change from 07-14: the price no longer requires the bull path to make sense; even the bull EPS only asks for a below-2018-peak multiple now.

現價 529 = 基準 EPS 25.9 的 20.4 倍（介於2018峰值與目標價欄位之間），或多方 EPS 36.5 的 14.5 倍（低於2018峰值欄位）— 這是與07-14相比的實質變化：這個價位已經不需要多方路徑才說得通；就連多方 EPS 現在對應的倍數也低於2018峰值。

✓ Split-adjusted, 2026E EPS (~18.97) roughly equals the 2018 cycle-peak EPS (80.3÷4=20.1); after the -56.6% break, the current price (529, 09-17 close) is now only 1.6× the 2018 peak price — down sharply from 2.3× at the 07-14 build's 764 and 2.5× at the initiation's 816. Most of what this analysis called "half momentum" in July has now unwound through price alone.分割調整後，2026E EPS（約18.97）大致等於 2018 循環頂點 EPS（80.3÷4=20.1）；跌了 56.6% 之後，現價（529，09-17收盤）現在只是 2018 頂點價格的 1.6 倍 — 較 07-14 建置時 764 元對應的 2.3 倍、初次評估時 816 元對應的 2.5 倍，都明顯下降。本研究 7 月時認為「一半是動能」的部分，多數已經單純透過股價下跌而消退。

✓ TV P/B cross-check: 6.37× at 529 (TV `price_book_ratio`, upgraded from the 07-14 build's ⚠ aggregator-derived ~9.3×; implies BVPS ≈NT$83.1, consistent with the prior ~82 estimate) vs ROE ~20-24%E — P/B has fallen roughly in proportion to price, and still sits above the ROE-implied "fair" P/B for a mid-20s ROE business, so it is not yet flagging as cheap on this cut either.P/B 交叉檢查：529 元下 6.37×（TV `price_book_ratio`，取代07-14版⚠ aggregator推算的約9.3×；隱含BVPS約NT$83.1，與先前約82的推算一致）vs ROE ~20-24%E — P/B 大致隨股價同比例下降，但仍高於 ROE 中段水準所對應的「合理」P/B，這個角度尚未出現便宜訊號。

Probability-weighted scenarios機率加權情境

Top: the three scenarios by probability weight (base = the TP price, revised 583→570). Bottom: the same three scenario prices plus the probability-weighted value (~626, was 632) and the current price (529, was 764) on a shared value axis — the accent dot and the red dot have moved much closer together than in the 07-14 build.上：三個情境按機率權重呈現（基準情境＝目標價，自583下修為570）。下：同樣三個情境價位，加上機率加權值（約626，原632）與現價（529，原764），畫在同一條數值軸上 — 靛色點與紅色點的距離，比07-14版拉近了許多。

Scenario情境	Prob.機率	2027E EPS	PER	Price價位	Trigger觸發條件
Bear空方	20%	20	14×	280	Hikes reverse by 1H27: China capacity + consumer collapse + AI digestion1H27前漲價反轉：中國產能+消費崩跌+AI消化
Base基準	50%	25.9	22×	570 (= TP, was 583)(=目標價，原583)	Hikes stick into 1H27 then plateau as capacity lands漲價撐到1H27後隨產能落地趨緩
Bull多方	30%	~36.5	26×	949	Q4'26 full shortage; 2028 visibly growing on Rubin content4Q26全面缺貨；2028在Rubin含量下明顯成長

Asymmetry check不對稱性檢驗 ◆

The picture at 529 is materially different from 764. To double (→1,058): this analysis's own bull path (EPS 36.5) now only needs ~29× — still above this analysis's 22× TP multiple, but no longer an unprecedented ask. More strikingly, Goldman's own 2028E EPS of 57.3 (last verified July, not re-checked this round) gets there at just ~18.5×, fully inside the 10–20× historical band. That doesn't make a double likely — it still requires an external house's EPS path to be right — but doubling from here no longer requires a multiple this stock has never sustained. To fall 50% (→265): needs bear-case EPS (20) at ~13.2×, BELOW even this analysis's own bear-scenario multiple of 14× (20×14=280) — a further 50% drop now needs something worse than the bear case already models, not just the bear case coming true. Asymmetry has moved from clearly negative at 764 toward roughly balanced at 529 — which is exactly why the rating moved from Avoid to Range-bound rather than staying put.

529 元的處境與 764 元時已有實質差異。要翻倍（→1,058）：本研究自己的多方路徑（EPS 36.5）現在只需要約 29 倍 — 仍高於本研究目標價所用的 22 倍，但已不再是「前所未見」的要求。更值得一提的是，Goldman 自己的 2028E EPS 57.3（最後查證於 7 月，本輪未再查證）只需約 18.5 倍就能到達 — 完全落在 10–20 倍的歷史區間內。這不代表翻倍很可能發生（仍需要外資的 EPS 路徑成立），但從這裡翻倍，已經不再需要一個本股歷史上從未維持過的倍數。要下跌 50%（→265）：需要空方情境 EPS（20）搭配約 13.2 倍 — 低於本研究自己空方情境所用的 14 倍（20×14=280）— 也就是再跌 50% 現在需要比空方情境本身更糟的狀況，不只是空方情境成真。不對稱性已從 764 元時明顯偏負，走向 529 元時大致平衡 — 這正是觀點由迴避轉為區間，而不是維持原地不動的原因。

Catalysts (12mo, dated)催化劑（12個月，已排日期）

Timing時間	Event事件	Status狀態	Threshold → reaction (pre-committed)門檻 → 反應（預先承諾） ◆
Jul–Aug 2026	2Q26 full P&L (per the 45-day disclosure rule)2Q26 完整損益表（依45天揭露規則）	resolved: miss已發生：未達標	◆ GM ≥41% or EPS ≥5.0 → this analysis was too conservative, TP revision review; GM ≤38.5% → hike pass-through is lagging, base-case path in question. Actual: GM 38.5%, EPS 4.59 ✓ 07-29 — landed exactly on the ≤38.5% trigger, not the upside one. Reaction executed: 3Q/4Q26E margin ramp trimmed, 2027E EPS cut 26.5→25.9, TP cut 583→570.GM ≥41% 或 EPS ≥5.0 → 本研究過保守，TP 上修檢討；GM ≤38.5% → 漲價傳導遲緩，基準情境路徑存疑。實際：毛利率 38.5%、EPS 4.59 ✓ 07-29 — 剛好踩在 ≤38.5% 的門檻上，不是升評門檻。反應已執行：3Q／4Q26E 毛利率爬升路徑下修、2027E EPS 由 26.5 下修至 25.9、目標價由 583 下修至 570。
~Aug 10	July monthly revenue — first full month reflecting the 7/1 hikes7月營收 — 首月完整反映7/1漲價	resolved: neither trigger fired已發生：兩個門檻皆未觸發	◆ ≥ NT$16.7bn → 3Q model on track; < NT$15.4bn (below June) → 3Q revenue estimate cut. Actual: NT$16.13bn ✓ 08-10 — above June's 15.36bn (downside trigger did not fire) but below the 16.7bn upside bar either. August followed at NT$16.33bn (+1.25% MoM, decelerating from July's +5.0%) ✓ 09-08 — directionally consistent with the GM-deceleration read, though no threshold was cleanly crossed. A gap the original design didn't anticipate.≥ NT$16.7bn → 3Q 模型在軌；< NT$15.4bn（低於 6 月）→ 3Q 營收估下修。實際：NT$161.31億 ✓ 08-10 — 高於 6 月的 153.59億（下行門檻未觸發），但也未達 16.7bn 的上行門檻。8 月接著為 NT$163.32億（月增1.25%，較7月的月增5.0%放緩）✓ 09-08 — 方向上與毛利率爬升趨緩的判斷一致，但沒有明確踩到任何門檻。這是原始門檻設計沒有預料到的中間地帶。
3Q26	Direct-customer contract reset +30–40% (MS estimate)直供客戶合約重議 +30-40%（MS估）	speculative推測	◆ lands ≥ +30% → bull-scenario probability revised up落地 ≥ +30% → 多方情境機率上調
4Q26	TrendForce's critical window for full shortageTrendForce 全面缺貨關鍵窗口	speculative推測	◆ full shortage confirmed → bull scenario triggers, TP framework re-rated; fails to materialize → cycle-top signal, bear weight revised up全面缺貨成真 → 多方情境觸發、TP 架構重估；未兌現 → 循環見頂訊號，空方權重上調
2H26	VR200 / Vera Rubin rack rampVR200/Vera Rubin 機櫃放量	roadmap, timing risk已排定藍圖，時間點有風險	◆ on-schedule ramp → AI mix >20% assumption confirmed; slippage → 2027E revenue estimate cut如期放量 → AI 佔比 >20% 假設確認；遞延 → 2027E 營收估下修
1H27	Murata/SEMCO/TY new capacity, first output (bear catalyst)村田/SEMCO/太誘新產能首次投產（空方催化劑）	planned已排定計畫	◆ on-schedule → base case confirmed; slips ≥2 quarters → 2027E EPS revised up如期投產 → 基準情境確認；遞延 ≥2 季 → 2027E EPS 上修
Ongoing持續	Monthly revenue prints ~10th; possible liquid-cooling M&A announcement每月約10日公布月營收；可能的液冷併購公告	speculative推測	◆ watch — informational (no preset threshold)觀察 — 資訊性（無預設門檻）

Playbook操作框架

Playbook (not investment advice)操作框架（非投資建議）

If not holding未持有	◆ Range-bound, not a buy signal (was Avoid) — 529 vs the 570 target implies about +7.75%, inside the -10%~+15% Range-bound band. The price condition of the entry zone (formerly <583, now <570) is now met, but the evidence condition (2Q26 GM ≥40%) is not — actual GM printed 38.5%. This is "no longer clearly overpriced," not "cheap with confirmation."區間，非買進訊號（前次為迴避） — 529 對目標價 570 隱含約 +7.75%，落在 -10%~+15% 的區間帶內。進場區的價格條件（原 <583，現 <570）現已滿足，但證據條件（2Q26 毛利率 ≥40%）未滿足 — 實際毛利率為 38.5%。這是「不再明顯偏貴」，不是「便宜且已獲證實」。
If holding已持有	◆ The author held 2327.TT at the 07-14 build (764) and exited in August 2026, before the price fell further to today's 529 — the Avoid call is vindicated by the price action. For a reader still holding: the framework's default trim-to-bear-survivable-size guidance is no longer as urgent as it was at 764 (bear scenario 280 is now -47% away, not -63%), but the evidence bar for actively adding hasn't cleared either — the same watch-and-wait applies as for new money.作者已於 07-14 建置時（764）持有 2327.TT，並於 2026 年 8 月出清，早於股價進一步下跌至今日的 529 — 迴避的判斷已由股價走勢印證。對仍持有的讀者而言：本框架原先「減碼至能承受空方情境」的預設指引，急迫性已不如 764 元時（空方情境280 現在距現價 -47%，不是 -63%），但加碼的證據門檻同樣尚未通過 — 適用與未持有相同的觀望邏輯。
Entry zone進場區	◆ Price condition MET: 529 < 570 (the new base-case price). Evidence condition NOT met: 2Q26 gross margin printed 38.5%, below the ≥40% bar. Next checkpoint: 3Q26 results (~November) — a GM print clearing 41% would complete the evidence condition and warrant an upgrade review; a repeat sub-38% print would reopen the Avoid case. Price alone is not a reason to add.價格條件已滿足：529 < 570（新基準情境價）。證據條件未滿足：2Q26 毛利率為 38.5%，未達 ≥40% 門檻。下一個檢查點：3Q26 財報（約11月）— 若毛利率站上 41%，證據條件即完成，值得檢討升評；若再度低於 38%，則重新開啟迴避的論述。單靠價格本身不構成加碼理由。
Why not more aggressive為何不更激進	◆ Not a buy: the evidence condition above hasn't cleared, and the July-revenue catalyst landed in an ambiguous middle zone rather than confirming the bull case. Not a short: the bull scenario (949) still sits +79% above the current price, and July/August revenue kept printing fresh records even as growth decelerated — the fundamentals have not broken, only the margin RAMP slowed. Range-bound means exactly that: neither side of the trade is compelling right now.不買進：上述證據條件尚未通過，7月營收這個催化劑也落在模糊的中間地帶，沒有確認多方論點。不放空：多方情境（949）仍高於現價 +79%，且7、8月營收持續創新高，只是成長速度放緩 — 基本面沒有壞，只是毛利率的爬升速度變慢。區間的意思正是如此：現在買進或放空都沒有足夠說服力。
Invalidation失效條件	◆ (1) 3Q26 GM ≥41% confirming faster pass-through → review for an upgrade toward Buy; (2) 3Q26 GM below 38% again, or Q4'26 shortage signal fails to appear → the Avoid case reopens.(1) 3Q26 毛利率 ≥41%，確認漲價傳導加速 → 檢討升評至買進；(2) 3Q26 毛利率再度低於 38%，或 Q4'26 缺貨訊號未出現 → 重新開啟迴避的論述。

Data provenance — every number, sourced資料來源 — 每個數字都有出處

54 claims: 44 ✓ verified · 9 ⚠ flagged · 1 ✗→ corrected — expand the ledger54 項查證：44 ✓ 已查證 · 9 ⚠ 未核實 · 1 ✗→ 已修正 — 點開展開逐項表

Claim主張	Verdict判定	Value · source · as-of數值 · 來源 · 日期
Price & valuation · live 2026-09-17 (superseding 07-14 figures below)價格與估值 · 即時 2026-09-17（取代下方07-14數字）
1. Close / -1.31% / prior close 536收盤/-1.31%/前收536	✓ IBKR	529 · 2026-09-17 close (ONE_DAY final bar, conid 38768800); TV batch agrees2026-09-17收盤(ONE_DAY final bar，conid 38768800)；TV批次一致
2. 52wk range 154–1,220; YTD +129%52週區間154–1,220；YTD+129%	✓ IBKR	09-17
3. Share count股數	✓ 工商時報 (07-14 build, not re-verified — TV market cap implies a ~0.6% higher count, unreconciled)(07-14版，本輪未重新查證 — TV市值隱含股數高約0.6%，未核對)	2,072,965,820 · 08-25
4. Market cap市值	✓ TV batch (used directly, not recomputed off #3)批次(直接採用，未以#3反推)	NT$1,103.3億×10 (1.10兆) · 09-17
5. TTM EPS 14.85TTM EPS 14.85	✓ TV (`earnings_per_share_diluted_ttm`), cross-checked by quarterly sum ≈14.86(`earnings_per_share_diluted_ttm`)，季度加總交叉核對≈14.86	09-17
6. TTM PER 35.6×	✓ recomputed live; matches TV's own `price_earnings_ttm` exactly現價重算；與TV自報`price_earnings_ttm`完全一致	09-17
7. BVPS ~83.1 / PBR 6.37×BVPS約83.1/PBR 6.37×	✓ TV `price_book_ratio` (upgraded from 07-14's ⚠ aggregator estimate)`price_book_ratio`(取代07-14版⚠ aggregator推算)	09-17
8. Dividend, yield股息、殖利率	✓ 鉅亨/AGM (dividend fact unchanged; yield recomputed at 529)(股息事實未變；殖利率以529重算)	5.998 · 07-03 · 1.13%
9. 1-for-4 split, "國巨" marker1拆4，「國巨」標記	✓ 工商時報/sinotrade	2025-08-25
Financials & guidance財務與指引
10. FY23/24/25 revenue營收	✓ multi-source多方來源	107,609/121,667/132,930
11. FY23/24/25 gross margin毛利率	✓ multi-source多方來源	33.48/34.36/36.21%
12. FY25 NI, EPS	✓ cnyes/sinotrade	23,634 (+22.1%) · 11.51
13. 1Q26 results (rev/GM/OM/EBITDA/NI/EPS)業績(營收/毛利率/營益率/EBITDA/純益/EPS)	✓ company deck公司法說簡報	38,166 · 38.1% · 25.2% · 31.7% · 8,001 · 3.90
14. 2Q26 self-reported revenue自結營收	✓ 工商時報	44,456 (QoQ+16.5%,YoY+35.7%) · 07-08
15. Monthly 2026 (Jun record, 1H sum)2026月營收(6月新高,上半年合計)	✓ cnyes	15,359(+38.9%) · 82,622(+29.4%) · 07-10
16. Guidance (Q2 moderate; utilization; B/B>1)財測(Q2小幅增;稼動率;B/B>1)	✓ call notes/AGM法說會/AGM	fugle/stockfeel/CNA
17. AI mix ~15%AI占比約15%	✓ AI Summit deckAI峰會簡報	4%→15% in於 <2yrs
18. FCF by quarter各季自由現金流	✓ deck法說簡報	3,600→6,847→8,742→7,752
19. Net debt/EBITDA淨負債/EBITDA	⚠ definition conflict定義衝突	59%→33%(co.) vs 1.3-1.6×(Taiwan Ratings)
20. 2026E/2027E estimates (revised 09-17 on the 2Q26 GM miss)估計(09-17因2Q26毛利率不如預期下修)	◆ judgment判斷	EPS 18.97/25.9 (was 19.4/26.5), rev 184.7B/225.3B (was 186.1B/227B)
21. FY2020-21 EPS (pre-split)FY2020-21 EPS(分割前)	⚠ aggregator-only僅aggregator	25.52/46.46
Business & mix (deck-verified)本業與組合(法說簡報查證)
22. Product mix產品組合	✓ deck	25.2/24.3/19.0/13.5/13.0/5.0
23. End-market mix終端市場組合	✓ deck	29.0/24.9/17.1/13.0/11.7/4.3
24. Channel / region mix通路/地區組合	✓ deck	44.6/35.0/20.4; 49.6/29.9/20.5
25. Unit economics單位經濟	✓ deck+computed法說簡報+計算	1.1T units → ~$0.004/unit
26. Share claims (R/MLCC/Ta/transformers)市占宣稱(電阻/MLCC/鉭/變壓器)	⚠ company-asserted公司宣稱	#1>30% / #3>15% / #1>40% / #1
27. MLCC #5 by revenue (aggregators)MLCC按營收#5(aggregator)	⚠ conflict w/ #26與#26衝突	7-9%
28. Premium/specialty mix特殊品占比	✓ slides+call簡報+法說會	75-80%
29. GM vs Murata (2Q26)毛利率對比Murata(2Q26)	✓ compiled彙整	38.5% vs ~41-42% (was 38.1% vs ~41-42% at 1Q26)(1Q26時為38.1% vs ~41-42%)
30. 90% BOM coverage覆蓋率	⚠ company claim公司宣稱	deck
Industry & cycle產業與循環
31. B/B Murata/SEMCO/TY vs 2018 peakMurata/SEMCO/太誘 vs 2018峰值	✓ TrendForce	1.30/1.31/1.25 vs 1.25 · 07-06
32. Channel inventory / lead times通路庫存/交期	✓ FindChips/773grp	1-1.5mo vs 4.5 norm常態
33. Spot / AI-cap price rises現貨/AI高容漲幅	✓ TrendForce	+15-20% / +50-60% · 06-25
34. Yageo 7/1 hike漲價	✓ 工商時報/TrendForce	+30-80%
35. Hike log (Feb R-chip, Mar KEMET, Apr TY)漲價紀錄(2月電阻,3月KEMET,4月太誘)	✓ dated log已標日期	+15-20% / — / +6-13%
36. Yageo AI-capacity share '25-28產能占比'25-28	⚠ broker-derived外資推算	9→14→24→32% · TechNews 07-06
37. Utilization high-end vs std稼動率高階vs標準	✓ TrendForce/efrontrade	90-95% vs 60-70%
38. MLCC content ladder (server→Rubin→rack)MLCC含量階梯(伺服器→Rubin→機櫃)	✓ TrendForce/MS	2,200→12,000; rack $1,530→$4,320(+182%)
39. Phones/PC units 2026手機/PC出貨2026	✓ IDC/Gartner	-13.9% / -10.4~11.3%
40. AI servers units/value share伺服器出貨/價值占比	✓ TrendForce	+28% units出貨; 74% of value價值占比
41. CN makers share / giveback risk中國廠市占/回吐風險	⚠ scenario-flagged情境標註	<8-10%; 15-25% risk風險
42. EV MLCC multiplier倍數	✓ Murata-attributed	3-6× (up to最高達 20×+)
Corporate governance & ownership公司治理與籌碼
43. Shibaura bid sequence/timeline競購序列/時間軸	✓ multi-source多方來源	4,300→7,130 · delisted下市 01-13
44. Ownership (foreign/insider/family)持股(外資/董監/家族)	⚠ family% derived家族%為推算	48.58% / ~7.6% / ~6.8%
45. Disposition / settlement default / holder count處置/違約交割/股東人數	✓ ETtoday/NOWnews/LTN/ftnn	+127.67%/30d · 6/30 NT$38.8M→NT$41.31M×13 · 45萬 holders股東
46. Sell-side TPs & survey consensus外資目標價與調查共識	✓ all named+dated皆具名標日	FactSet consensus TP 1,275 · 2026E median 19.59 (16.98–21.22, prior 18.89; via ftnn, 07) / GS 1,490 / MS 1,515 / Citi 495→1,500 / 宏遠 1,230 / JPM 1,010 / Nomura 850 / HSBC 650FactSet 共識TP 1,275 · 2026E中位數 19.59（16.98–21.22，前版18.89；ftnn轉述，07月）/ GS 1,490 / MS 1,515 / Citi 495→1,500 / 宏遠 1,230 / JPM 1,010 / Nomura 850 / HSBC 650
47. -33%/6-session break, no fundamental news6交易日-33%,無基本面消息	✓ LTN/Yahoo	07-13
Added 2026-09-17 (claims 48-54)2026-09-17新增(48-54筆)
48. 2Q26 actual results (rev/GM/OM/NI/EPS)2Q26實際數(營收/毛利率/營益率/純益/EPS)	✓ 經濟日報/東森財經/leveragetw	44,456 · 38.5% · 26.4% · 9,418 · 4.59 · 07-29
49. July/August 2026 monthly revenue7月/8月2026月營收	✓ cnyes/工商時報	16,131(+51.5%YoY)·08-10 / 16,332(+51.8%YoY)·09-08
50. Two-day limit-down, 07-28/07-29 (563→507)連兩日跌停，07-28/07-29(563→507)	✓ leveragetw/businesstoday/ETtoday	-58% off the 7/1 ATH距7/1高點 · 07-28/07-29
51. 3Q26 guidance (utilization >90%, steady growth, margins expanding — qualitative)3Q26指引(稼動率>90%、營收穩健成長、毛利率續增—定性)	✓ leveragetw/口袋學堂/CMoney	07-29 法說會法說會
52. TV comps batch (Murata/TDK/Walsin close & multiples)TV同業批次(村田/TDK/華新科收盤與倍數)	✓ TV	Murata ¥7,413/51.0× · TDK ¥2,809.5/22.7× · Walsin NT$297 · 09-17
53. TV consensus 2026E EPS ≈19.86, 17 analysts "Strong Buy"TV共識2026E EPS約19.86，17位分析師「強力買進」	⚠ currency-corrected, see methodology已換匯修正，詳方法說明	09-17
54. USD/TWD 31.80 (used only to de-mangle claim #53)USD/TWD 31.80(僅用於還原第53筆)	✓ tradingeconomics/investing.com	09-16 close收盤
Judgment (not the 54 enumerated claims)判斷(不計入54筆列舉)
TP formula (2027E×22×)目標價公式(2027E×22×)	◆	570 (was原 583)
Scenario probabilities情境機率	◆	20/50/30 → weighted加權 626 (was原 632)
Research view (Avoid→Range-bound)研究觀點(迴避→區間)	◆	mechanical consequence of TP-vs-price, see mastheadTP對現價的機械結果，詳頂部摘要 · 09-17
Playbook / entry zone / thresholds (decision layer)操作框架/進場區/門檻(decision layer)	◆	all zones/thresholds = this analysis's judgment · 09-17所有區間/門檻皆為本研究判斷 · 09-17

Counts: 54 claims enumerated (47 from 07-14 + 7 added 09-17) → 44 ✓ verified · 9 ⚠ flagged-partial (BVPS/PBR was one of these, now upgraded to ✓ at #7; TV consensus currency-correction #53 added as a new ⚠) · 1 ◆ judgment (2026E/27E estimates) within the 54, plus 4 further ◆ judgment calls outside the enumerated set (TP formula, scenario probabilities, research view, decision layer) — 5 ◆ total referenced in this analysis.統計：54筆列舉主張（07-14版47筆＋09-17新增7筆）→ 44✓已查證 · 9⚠部分標註（BVPS/PBR原為其中一筆，現已升級為#7的✓；TV共識換匯修正#53為新增的⚠）· 54筆內1◆判斷(2026E/27E估計)，另有4項◆判斷不計入54筆列舉(目標價公式、情境機率、研究觀點、操作框架決策層) — 全文共5◆判斷。

Dropped (4, couldn't verify → excluded): a 36kr $12k-40k/rack MLCC-content figure (8× conflict with the TrendForce/MS-cited set); a "15,000-25,000 MLCC per AI motherboard" claim (untraceable); Shibaura EPS accretion % (never disclosed); KEMET synergy target (never quantified). Separately (not dropped, verified as absent): no 2025-26 buyback found on MOPS — stated in the management section as "no live buyback" with a verify note.剔除(4筆，無法查證→排除)：36kr 的 $12k-40k/機櫃 MLCC含量數字(與TrendForce/MS引用的資料組相差8倍)；「每片AI主機板15,000-25,000顆MLCC」(無法溯源)；芝浦EPS增厚%(從未揭露)；KEMET綜效目標(從未量化)。另外(非剔除，而是查證為「不存在」)：MOPS上未找到2025-26庫藏股 — 已在經營層章節註明「無現行庫藏股」並附查證說明。

Update log更新紀錄

✗→ 09-17 Margin growth slowed down; the price fell even further — the view moves from avoid to no strong edge毛利率成長變慢了，股價卻跌得更多——研究觀點從迴避改為看不出明顯方向

What happened: the July 29 earnings report showed profit margin catching up more slowly than this analysis had assumed, and the stock price then fell by more than half from its July peak — including two days where the stock hit its daily down-limit, one of them the SAME day as the earnings report. What we did: trimmed the profit forecast a little and lowered the fair-price estimate to match, but because the price fell so much further than the forecast did, the stock is no longer priced for certain bad news either. What stayed the same: the belief that this round of price hikes is real, and the belief that 2027-28 is where this analysis and outside brokers genuinely disagree.發生了什麼：7 月 29 日的財報顯示，毛利率追趕的速度比本研究原先假設的慢，接著股價從 7 月的高點腰斬超過一半——其中兩天股價觸及當日跌停，一天還恰好是財報公布當天。我們做了什麼：把獲利預估小幅下修，合理價格也跟著調降，但因為股價下跌的幅度遠超過預估下修的幅度，這支股票現在也不再是「篤定看壞」的價位。什麼沒有改變：本研究仍然相信這輪漲價是真的，也仍然相信 2027-28 年是本研究與外資真正分歧的地方。

1. What happened1. 發生了什麼

Second-quarter revenue hit a record, up 16.5% from the prior quarter, but profit margin rose only 0.4 percentage points to 38.5% — this analysis had expected about a full percentage point more. Profit per share of 4.59 was still a record and almost matched this analysis's own estimate, so the miss showed up in the margin line, not the bottom line. ✓ 07-29 The stock then closed at its daily down-limit twice in a row, including the earnings day itself, before settling around today's price — every source covering the move blamed a broad sell-off in chip stocks and forced selling from margin accounts, not a change in the business. ✓

第二季營收創新高，比前一季成長 16.5%，但毛利率只季增 0.4 個百分點到 38.5%——本研究原先預期會多漲將近一個百分點。每股純益 4.59 元同樣創新高，幾乎與本研究自己的估計相同，所以這次沒踩到的地方是毛利率，不是最終獲利數字。✓ 07-29 股價隨後連兩天觸及跌停，其中一天就是財報公布當天，之後才回穩到今天的價位——所有報導這次下跌的來源，都把原因指向晶片股的全面性賣壓與融資戶的斷頭賣壓，不是公司本身出了問題。✓

2. What we downgraded2. 我們調降了什麼

Estimates: because the price-hike pass-through is running slower than modeled, this analysis trims 2026/2027 profit-per-share estimates from 19.4/26.5 to 18.97/25.9 and the fair price from 583 to 570. The multiple used to set the fair price (22× next year's profit) has NOT changed — this is purely a profit-estimate cut, not a decision that the stock deserves a lower valuation multiple.

預估：因為漲價傳導到毛利率的速度比模型假設得慢，本研究把 2026／2027 年每股獲利預估由 19.4／26.5 下修為 18.97／25.9，合理價格也由 583 下修為 570。用來算合理價格的倍數（明年獲利的 22 倍）並沒有變——這純粹是獲利預估下修，不是判斷這支股票該用更低的倍數評價。

3. What did not change3. 什麼沒有改變

Fundamentals: July and August revenue both printed fresh monthly records (though the pace of growth cooled), management's outlook for the current quarter is unchanged (utilization rising above 90%, revenue growing, margins still expanding), and the structural argument — that this is not just an MLCC company, and that AI demand is genuinely early — is untouched. Nothing this round says the business broke; it says the ramp is happening on a slower clock than this analysis first assumed.

基本面：7 月與 8 月營收都創了新的單月紀錄（雖然成長速度已經放緩），公司對本季的展望也沒有變（稼動率升破九成、營收持續成長、毛利率仍在擴張），而「國巨不只是一家 MLCC 公司」、「AI 需求真的還在早期」這兩個結構性論點也都沒有變。這一輪更新沒有任何一項顯示公司出了問題；只是爬升的速度比本研究原先假設的更慢。

4. The decision4. 我們的決定

The stock has fallen so far since the last update that even the OLD fair price (583) would already put it inside the "no strong edge" band rather than the "avoid" band. That is the main reason the view changes — not this update's own profit-estimate trim. The price has now met one of the two conditions this analysis set in July for getting more constructive (the price is below the fair-price line); it has NOT met the other one (profit margin needed to clear 40% in the second-quarter report, and it printed 38.5%). One condition met and one not met is exactly why the honest call is "no strong edge," not "buy."

股價自上次更新以來跌得夠深，即使用舊版合理價格（583）來算，現在也已經落在「看不出方向」的區間，而不是「迴避」的區間。這是觀點改變的主要原因——不是這次獲利預估下修造成的。本研究 7 月設下的兩個轉為積極的條件，價格已經滿足其中一個（股價已經低於合理價格）；另一個沒有滿足（第二季毛利率需要站上 40%，實際是 38.5%）。滿足一個、沒滿足另一個，正是為什麼誠實的結論是「看不出方向」，而不是「買進」。

One more note: the author of this page held this stock at the last update's price of 764 and sold the position in August 2026, before the price fell further to today's level — the original avoid call has since been borne out by the price action. Also, the outside analyst consensus for this year's profit (from a data provider called TradingView, after correcting a currency mix-up in its raw numbers) has moved from 19.59 in July to about 19.86 now — almost no change, a sharp slowdown from the roughly 25%-per-month pace of upward revisions seen through June and July.

補充一點：本頁作者在上次更新的價位（764）持有這支股票，已於 2026 年 8 月出清，早於股價進一步下跌至今日的水準——當時的迴避判斷，已被之後的股價走勢印證。另外，外資對今年獲利的共識（來自一家叫 TradingView 的資料商，經修正其原始數字中的匯率換算問題後）由 7 月的 19.59 移動到現在約 19.86——幾乎沒有變化，比 6、7 月時每月上修約 25% 的速度明顯放緩許多。

Snapshot: view Avoid → Range-bound; fair price 583 → 570; close 764 (07-14) → 529 (09-17); implied gap -24% → +7.75%.

數據快照：觀點迴避 → 區間；合理價 583 → 570；收盤 764（07-14）→ 529（09-17）；隱含空間 -24% → +7.75%。

Earlier entry (2026-07-14 numbers refresh + decision layer added) — expand較早的紀錄（2026-07-14 數字更新＋決策層補建）— 點開展開

✓ 2026-07-14 — Numbers refreshed + decision layer added2026-07-14 — 數字更新＋決策層補建

Numbers: price 816→764 (intraday 09:15, -37% from the 7/1 peak; ✓ IBKR), all live multiples recomputed (TTM 60.2×, 2026E 39.4×, ~24× consensus 27E); no new fundamental news — the slide extends the positioning unwind. Estimates unchanged (2026E/27E EPS 19.4/26.5).

數字：股價 816→764（盤中09:15，距7/1高點-37%；✓ IBKR），所有即時倍數重算（TTM 60.2×、2026E 39.4×、共識27E約24×）；無新增基本面消息 — 這波下跌延續籌碼出清。估計不變（2026E/27E EPS 19.4/26.5）。

TP restated 585→583 ◆ — arithmetic (26.5×22=583) + anchor unification (base scenario aligned to the TP's 22×, 557→583; bull 950→949; weighted 620→632). Not a view change: Avoid maintained; neither EPS nor the multiple moved.

目標價重列 585→583 ◆ — 算式（26.5×22=583）＋情境錨定一致化（基準情境對齊目標價的22倍，557→583；多方950→949；加權620→632）。並非觀點改變：迴避維持，EPS 與倍數皆未變動。

Format upgrade (house format 2026-07-14): added the 8-numbers decision box, sensitivity grid, threshold→reaction catalysts, playbook (incl. if-holding guidance), fast path, and a first de-dup pass. ◆ framing = judgment.

格式升級（house format 2026-07-14）：新增關鍵數字盒、敏感度表、門檻→反應催化劑欄、操作框架（含若已持有指引）、快速路徑，並做了首輪去重。◆ 標記＝判斷。

Date日期	Action動作	View觀點	TP	Close收盤
2026-09-17	2Q26 miss scored, model rolled forward2Q26未達標計分，模型順延	Range-bound (was Avoid)區間（前次迴避）	570 (prior前次 583)	529 close收盤
2026-07-14	Update + decision layer數字更新＋決策層補建	Avoid (maint.)迴避（維持）	583 (prior前次 585)	764 intraday盤中
2026-07-13	Initial coverage初次評估	Avoid迴避	585	816

Append-only ledger. Revision table (§model) now available from the 09-17 update onward, scored against the 07-14 snapshot.附加式(append-only)紀錄。修正對照表（見財測模型章節）自 09-17 起可用，對照 07-14 版模型計分。