← back to picks

HQ Taoyuan · Taiwan Reporting TWD Credit synth Aaa/AAA (peak ICR) · modest net debt Valuation 2026-06-02 Damodaran FCFF · 1000-iter MC FCFF · Dark v3

Priced for a permanent AI super-cycle · −88% margin of safety

Overvalued in every scenario — even an aggressive bull is −72%

Intrinsic value NT$98.85/share (after a 15% governance haircut; pre-haircut NT$116.29) vs market ~NT$800 — a −88% margin of safety. The Monte Carlo p95 is only NT$141 (still −82%): every one of 1,000 correlated draws lands below today's price, so P(intrinsic>market)=0%. Ta Liang trades at ~75–93× trailing earnings after a +812% run, on a base year (TTM) that is a cyclical PEAK — a record 19.9% operating margin and NT$6.18B revenue sit at the crest of a sawtooth that printed outright losses as recently as 2023.

SectorPCB / semiconductor equipmentCountry mixChina 68% · Taiwan 20%β / MC σ1.355 levered · ±NT$23/sh (1000 runs)Governance~29% controlling family · 1 indep. dir · 15% haircutQualityDeep-cyclical · 3-yr ROIC ~3% · negative-FCF up-yearsValuation~75–93× P/E · +812% in 12 months

Intrinsic / share

NT$98.85

post 15% gov · pre NT$116.29

Market / share

NT$800.00

2026-06-02 · TWSE · +812% in 12 months

Margin of safety

-87.6%

vs intrinsic

Enterprise value

NT$11.24B

66.6% terminal

Cost of equity / debt

9.73% / 2.44%

β 1.35 · CRP 0.99%

Terminal ROIC / g

10.00% / 2.00%

spread ~44bp (ROIC 10.00% vs WACC 9.56%)

What it sells, where it sells

Operating segments

PCB equipmentHigh-end CCD back-drilling + TM inner-layer copper metrology (±2µm) for AI-server boards; the cyclical core, ~89% back-drill / 11% routing~90%

Semiconductor equipmentAdvanced-packaging metrology/AOI (CoWoS, HBM, silicon photonics) to TSMC/ASE/top-OSATs — small today, the structural growth driver~10%

The ~10% semiconductor-metrology business is the genuine structural story (recurring, higher-margin, AI-anchored) — but it is too small today to carry the valuation, and the ~90% PCB-equipment core is a lumpy, derivative-of-hyperscaler-capex business whose record margin is a high-end mix shift struck at a cyclical peak, not durable pricing power.

Country mix (revenue-weighted CRP input)

🇨🇳China68%

🇹🇼Taiwan20%

🇹🇭Thailand8%

🇰🇷South Korea2%

🇻🇳Vietnam1%

🇺🇸United States1%

End-demand is ~68% mainland-China PCB board-makers, so revenue is doubly geared to the AI-capex cycle AND to China macro/political risk — the country mix adds a real ~1.0% weighted country-risk premium to the cost of capital, not a flat discount bolted on the end.

Background — five things to know before the case

Context the share price doesn't carry on its face. Skim this once and the bull / bear bullets below stop reading as inside baseball.

The base year is a cyclical PEAK, not a steady state. TTM revenue NT$6.18B and a 19.9% operating margin are an all-time high; the same business cycled through peaks in 2018 (~NT$4.0B) and 2021 (NT$4.44B) and a −71% drawdown to a NT$1.29B trough in 2023 — with outright quarterly net losses and an H1'24 operating loss. The 5-year earnings trend is negative (−2.1%/yr); 5-yr revenue +3.5%/yr.
A genuine, corroborated high-end moat — with a commoditized low end. The proprietary TM inner-layer copper-measurement tool (2±2mil, Cpk>1.67, D+4mil) is the spec NVIDIA-class AI-server boards require; management asserts competitors "must" use it. But the standard end faces real mainland red-supply-chain price competition, and equipment demand is a one-time capex pulse per board-maker line.
The 39% gross margin is a mix shift, not pricing power. High-end back-drill/forming machines went from ~20–30% to ~50% of PCB revenue, lifting gross margin from ~23% to 37–39%. Management explicitly avoids raising prices on existing customers to protect relationships.
Capital is reinvested and diluted, not returned. Free cash flow is negative in up-years (FY25 −NT$121M, FY24 −NT$503M) on the Nanjing/Lianshui build-out; the dividend is small and pro-cyclical (~0.5% yield); share count rose ~7% via a zero-coupon convertible + restricted stock. No buyback-and-cancellation.
Two carve-outs route the fastest-growing units below the parent minority. A ~29% controlling family (one independent director) is floating the Nanjing CNC subsidiary on a China/HK exchange (10–25%) and spinning the semiconductor BU toward a separate Taiwan IPO — value-routing that makes the parent's NAV a ceiling for the minority, not a target.

Where we diverge from the sector

Each row asks: if we'd used the sector median for this one assumption instead of our override, how much would the share price change? Negative = our override is more conservative (worth less); positive = more aggressive (worth more). Skim down the rightmost column to see which bets are doing the work.

Assumption

Our input

What the market prices

Read

01Revenue growth10-yr path off the TTM peak base

~6% CAGR8% Y1-5 (honors the visible 2026-27 order book) fading to 2.0%; lands year-5 ~NT$9.1B, +47% over the peak

~doublingSell-side: FY26 "challenging NT$10B", near-doubling into 2027 → a permanent NT$15-20B level

peak ≠ level

02Operating marginYear-10 through-cycle vs the TTM peak

12.0%Faded from the 19.9% TTM peak; pre-AI through-cycle avg ~7% (sequence +12/+4.5/−3.7/+2.8/+18.1%)

~20% heldThe market extrapolates the 19.9% peak as the new normal

load-bearing

03Sales-to-capitalReinvestment vs Machinery median 1.48

1.5 / 1.8×Heavy reinvestment, faithful to negative-up-year FCF; eases post-Nanjing

~1.48×Damodaran-global Machinery median

≈ in line

04Terminal growthYear 10+ vs the TWD risk-free ceiling

2.0%TWD long-run inflation; held ≤ the 2.65% risk-free (Damodaran ceiling)

2.65%risk-free ceiling (65/35 blend)

NT$0

05Cost of capitalβ anchor: machinery vs the market's semicap identity

9.56% WACCβ 1.355 (Damodaran Machinery, regression-corroborated) · rf 2.65% · 68% China CRP

~12%+ WACCA semicap β (2.12) would push WACC higher → even lower value

generous

Net read

The inputs are, if anything, GENEROUS to the stock — machinery β (not the higher semicap β), a 12% margin above the Machinery sector median, and 8% near-term growth that honors the full order book. Intrinsic still lands NT$99 because the base year is a cyclical peak and the price is ~8× that. There is no reasonable input set that closes the gap.

−88% MoS

Our input What the market prices Generous to the stock Where price detaches from value

The story & the five claims

The 10-year story this DCF is built on, plus the five anchor claims that translate the story into DCF inputs.

The 10-year story

Ta Liang is a real, patent-backed PCB and semiconductor-equipment franchise that the AI-server boom has lifted to an all-time earnings high — and the market has repriced +812% in twelve months as if that high is the new permanent base. It is not. This is a violently cyclical equipment maker: it peaked in 2018 and 2021, lost money in 2023, and its five-year earnings trend is actually negative. The genuine moat — a proprietary ±2µm inner-layer copper-measurement tool that NVIDIA-class boards need — protects share and price at the high end, but equipment demand is a one-time capex pulse per customer line, derivative of hyperscaler AI spending. Our model honors the visible 2026-27 order book (8% near-term growth), then normalizes: operating margin fades from today's 19.9% peak to a through-cycle 12% (the pre-AI average was ~7%), revenue compounds to roughly NT$11B by year ten, and terminal growth is pinned at Taiwan's 2.0% inflation. After a 15% governance haircut for the controlling family and two value-routing carve-outs, intrinsic value is NT$99 a share — against a market price near NT$800. The Monte Carlo's most optimistic draw (NT$141) is still ~82% below price, and even an aggressive bull that triples revenue to NT$14.5B at 18% margins held forever only reaches NT$222. The conclusion is not close: the market is pricing zero cyclicality and permanent sole-source rents on a business whose own history refuses both.

Two debates worth pressure-testing

Is the through-cycle normalization too harsh on a genuine sole-source moat?

Our view: We tested exactly this. An independent council flagged that an earlier 4% growth amputated the visible backlog, so we raised near-term growth to 8% (year-5 revenue +47% over the peak) — intrinsic moved only NT$86 → NT$99. Push further to an aggressive bull (revenue to NT$14.5B, 18% margin held, semicap ROC) and you still land NT$222, −72%. The moat justifies a premium business; it does not justify paying ~8× intrinsic.

Could the semiconductor-metrology business re-rate the whole company?

Our view: Possibly — it is the one real bull lever. CoWoS/HBM metrology is recurring and higher-margin, and if it grows from ~10% to the majority of revenue the through-cycle base and margin both rise. But it is ~10% today; we treat it as upside optionality, not the base case, until it crosses ~30-40% of revenue. Even crediting it generously narrows the gap to ~−60%, not zero.

CLAIM 01Revenue grows ~8% Y1-5 (honors the backlog), fading to 2.0%.growth_high: 8.0% · terminal: 2.0% · NT$6.18B → ~NT$11.3BOrder book full to Sep-Oct 2026 with 2027 reserved, so near-term is up — but every prior peak gave back 45-71%. 8% lands year-5 ~NT$9.1B (+47% over the all-time high) while refusing the bull NT$15-20B super-cycle.

CLAIM 02Operating margin normalizes to 12% — not the 19.9% peak.target_op_margin: 12% by Y5The 39% gross / 19.9% operating margin is a high-end mix shift at full capacity, not durable pricing. Through-cycle operating margin has averaged ~7% (and went negative in 2023). 12% is the moat-tilted mid.

CLAIM 03Heavy reinvestment — cash, not accounting earnings.S2C: 1.5× Y1-5 / 1.8× Y6-10FCF is negative in up-years (FY25 −NT$121M, FY24 −NT$503M) because the Nanjing/Lianshui build-out consumes cash. A low sales-to-capital forces the model to fund growth with real reinvestment.

CLAIM 04Terminal ROC 10% — a thin moat premium, not a perpetual peak.override_roc: 10% · g 2.0%Through-cycle ROIC has averaged ~3% (3-yr) and book value is flat over 5 years; the franchise earns excess returns only at peaks. 10% credits a thin durable high-end-moat premium over the ~7.6% terminal WACC.

CLAIM 05Governance haircut 15% — two carve-outs route value below the minority.governance_discount: 15%Nanjing CNC China/HK IPO (10-25% float, proceeds to capex not parent dividends) + a semi-BU spin-off, under a ~29% controlling family with one independent director. The NAV is a ceiling for the minority, not a target.

Where we diverge from sell-side

We treat the base year as a peak; the market treats it as the new normal. At ~75-93× trailing earnings the price extrapolates the 19.9% operating margin and the +95% revenue jump; our model fades both toward a through-cycle mid because the company's own 11-year record is a sawtooth with a negative 5-yr earnings trend.
We honor the backlog but not the super-cycle. Sell-side models FY26 "challenging NT$10B" and a near-doubling into 2027 as a permanent level; we credit the visible order book (8% near-term) and then a cyclical fade.
The 39% gross margin is mix, not pricing. Management avoids raising prices to protect customer relationships — so the margin is contingent on the high-end mix staying ~50%, which a standard-end price war or an AI-capex air-pocket reverses.
Governance is priced in the flows, not waved away. The two carve-outs and the small pro-cyclical dividend make the NAV a ceiling for the minority — a 15% haircut plus a reinvestment drag, where the bull case implicitly assumes all cash reaches the parent shareholder.
The conclusion is robust, so we publish a range, not a decimal. Worth roughly NT$100-220 across base-to-aggressive-bull, versus ~NT$800 — the gap is a canyon, not a rounding error.

Two-sided case — bear anchors

The metrology BU becomes the company. If CoWoS/HBM advanced-packaging metrology scales from ~10% to the majority of revenue at 25-30% operating margins, the through-cycle base and margin both re-rate — the single path that could turn "fairly valued" plausible (still not NT$800).
AI-server PCB demand is a durable multi-year build, not a pulse. If rising layer counts and 1.6T optical keep back-drill/TM demand structurally elevated for years, year-5 revenue beats NT$9B and the cyclical-fade assumption is wrong.
The moat converts to pricing power. If the new TM-4 and sole-source position let Ta Liang actually raise prices (so far it won't), the 19.9% margin holds rather than fades — the biggest single swing factor.
The carve-outs reprice value UP. A China/HK listing of Nanjing at an AI multiple the Taiwan parent never got credit for could surface value — the carve-outs cut both ways.
Momentum keeps working. A +812% darling with thin float and a controlling family can rip another 50% on flows alone; being right on value says nothing about timing.

Risks to thesis (tail, not bear case)

It keeps ripping (we're early/wrong on timing)High

A momentum darling +812% in a year with thin float and concentrated holders can extend far past intrinsic. This is an AVOID, not a short — there is no cheap borrow and no timing edge; the actionable output is "don't own it."

Semiconductor-metrology re-ratingMed

The one genuine bull lever. If the ~10% CoWoS/HBM business scales to majority revenue at semicap margins, the through-cycle base rises and the gap narrows toward −60% — still overvalued, but the thesis weakens.

Durable AI-capex super-cycleMed

If hyperscaler AI-board demand proves a multi-year structural build rather than a pulse, the cyclical-fade assumption is too harsh and year-5 revenue beats NT$9B.

Cyclical rollover (confirms the thesis)Low

The base risk to the COMPANY (not our call): a NVIDIA platform-transition slip or AI-capex air-pocket de-rates equipment demand, repeating the −45% to −71% drawdowns of 2022 and prior cycles.

Governance / carve-out leakageLow

Predatory Nanjing IPO terms (>25% float, low valuation, related-party placement) would route more value below the parent minority — already partly captured by the 15% haircut.

DilutionLow

Zero-coupon convertible + restricted stock have raised share count ~7%/yr; further equity-linked issuance to fund capacity dilutes per-share value.

10-year forecast

Revenue compounds off the TTM peak as the operating margin fades from 19.9% to a through-cycle 12% by year 5 — the normalization that separates a cyclical peak from a permanent level.

Monte Carlo distribution

Even the 95th-percentile draw (NT$141) sits ~82% below the ~NT$800 price — all 1,000 correlated draws land below market, so P(intrinsic < market) = 100%.

Mean NT$100.99 ± NT$22.79/sh, 1000 iterations (0 failed). P(intrinsic < market NT$800.00) = 100.0%.

Cost of capital build

Risk-free rate	2.65%
Mature-market ERP	4.30%
Levered β	1.35
Weighted CRP	0.99%
Cost of equity	9.73%
Pre-tax cost of debt (synth Aaa/AAA)	3.05%
D / V	~2%
WACC	9.56%

Full year-by-year DCF

Year	Revenue	Op mgn	EBIT	EBIT(1−t)	Reinvest	FCFF	PV
1	NT$6.67B	18.94%	NT$1.26B	NT$933M	NT$330M	NT$603M	NT$551M
2	NT$7.21B	17.20%	NT$1.24B	NT$915M	NT$356M	NT$559M	NT$466M
3	NT$7.79B	15.47%	NT$1.20B	NT$889M	NT$384M	NT$504M	NT$383M
4	NT$8.41B	13.73%	NT$1.15B	NT$852M	NT$415M	NT$437M	NT$303M
5	NT$9.08B	12.00%	NT$1.09B	NT$804M	NT$448M	NT$356M	NT$225M
6	NT$9.70B	12.00%	NT$1.16B	NT$873M	NT$343M	NT$530M	NT$308M
7	NT$10.24B	12.00%	NT$1.23B	NT$937M	NT$302M	NT$636M	NT$340M
8	NT$10.69B	12.00%	NT$1.28B	NT$995M	NT$250M	NT$744M	NT$368M
9	NT$11.03B	12.00%	NT$1.32B	NT$1.04B	NT$190M	NT$853M	NT$393M
10	NT$11.25B	12.00%	NT$1.35B	NT$1.08B	NT$123M	NT$958M	NT$412M

Methodology & flags

Damodaran FCFF DCF, 10y explicit + perpetuity, in TWD. Risk-free 2.65% (local-currency government bond). Synthetic credit Aaa/AAA. CRP 0.99% from revenue-weighted country mix × Damodaran 2026 CRPs. Terminal ROIC faded to 10.00%; 15% governance haircut applied post-DCF. Monte Carlo: 1000 iterations. Engine v1.0.0 · result: /Users/valentin/Documents/notes_jiliac_labs/Finance/Damodaran/valuations/taliang/output/2026-06-02-result.json

R&D cap: ON · Lease cap: OFF · Failure: OFF · ESO: OFF
Governance haircut: 15% applied post-DCF (NT$116.29 > NT$98.85)
Sensitivity tornado: not run

FCFF DCF · dark theme v3. Generated 2026-06-02T10:36:58 from /Users/valentin/Documents/notes_jiliac_labs/Finance/Damodaran/valuations/taliang/output/2026-06-02-result.json.