Winbond (2344.TW): Real Earnings Inflection, the 10x Rally Has Priced In the Cyclical Peak

The 2026Q1 results booked across-the-board price increases in niche DRAM, NOR and SLC NAND, confirming the earnings inflection. The stock is up ~+880% over the past year (nearly 10x), ~+170% over six months, ~+115% over three months and ~+46% over one month; in June 2026 alone it ran from NT$145 to a NT$233.5 high, then pulled back to NT$206.5, tracing a parabola. The imminent listing of a 2x-leveraged memory ETF adds evidence of elevated sentiment. The advance includes a material liquidity premium beyond the improvement in earnings.

The three majors and China's CXMT are shifting from DDR4 toward HBM/DDR5, leaving Winbond and Nanya with greater pricing power. Four investment conclusions follow: 1) reported earnings confirm the current inflection; 2) concentrated DDR4 supply supports pricing power; 3) CUBE adds long-term option value; and 4) core memory and Nuvoton require separate valuation treatment.

Thesis 1: Across-the-board price increases are already in the accounts; the earnings inflection is real

Reported earnings confirm the inflection.

The impact path is "industry shortage → niche ASP jump → Winbond revenue and gross margin rise together → operating leverage amplifies to net profit." The current-period reads are all in place: 2026Q1 consolidated revenue NT$38.25B, core memory revenue NT$29.92B at 56.6% gross margin and 42.5% operating margin; DRAM revenue +93% QoQ with ASP up more than 50%; Flash revenue +23% QoQ with ASP up more than 30%. Two points deserve emphasis. First, the revenue increment comes mainly from price (ASP), with volume secondary; the typical shape of niche memory in an upcycle, where capacity is short-term rigid and price increases pass straight through to gross margin, with the single-quarter gross-margin +11pp being the direct expression of operating leverage. Second, the company basis and the third-party basis corroborate each other: in TrendForce's global DRAM supplier ranking, Winbond's 1Q26 DRAM revenue was ~US$568M, +91.4% QoQ, almost matching the earnings-call DRAM +93%; peer Nanya posted 1Q26 revenue of US$1.55B, +60% QoQ, same direction and comparable magnitude. Monthly revenue extends the inflection into Q2: May 2026 consolidated revenue NT$20.0B (+182% YoY), April NT$19.245B (+182% YoY), February NT$11.97B (+88% YoY); the first two months of Q2 already reach NT$39.2B, already above all of Q1's NT$38.25B, meaning Q2 single-quarter revenue will step up sharply versus Q1. Financial statements and monthly revenue jointly confirm the earnings inflection. Any quarter in which core-memory gross margin turns down QoQ, or two consecutive months in which monthly-revenue YoY momentum converges below double digits, would undermine the current-period "price pass-through" logic.

Thesis 2: The concentrated DDR4 supply structure distinguishes Winbond from the majors

Nanya and Winbond hold greater pricing power as larger suppliers shift capacity.

This is the biggest difference between Winbond and a major like Micron, and the bull camp's core assumption. The three majors (Samsung/SK hynix/Micron) shift wafer capacity wholesale to HBM and enterprise DDR5, and China's CXMT is likewise migrating fast from DDR4 to DDR5/HBM, so the supply of mature-node niche DRAM like DDR4/DDR3 shrinks passively with new capacity additions at zero. Some industry analysts expect that by end-2026 global DDR4 capacity could fall to roughly a quarter of early-2025 levels, at which point supply becomes highly concentrated in Nanya and Winbond. When a category that still carries real downstream demand (TVs, networking, industrial, consumer, embedded, some servers) is left with only two rational players on the supply side, the pricing logic switches from "cost-plus" to "scarcity pricing," and gross margin is now set by the buyer's substitution cost and supply-disruption risk rather than cash cost. Lock-in behavior corroborates this logic: Nanya and Winbond are extending LTAs from one year to at least two, with some frameworks stretching to 2030, using a "lock volume, float price" structure; this gives the two a smooth baseline of profit and utilization while capping their upside margin elasticity when spot surges, an arrangement that trades upside for certainty. Relative to Micron, whose 2026 earnings are closer to an absolute peak, Winbond's high-margin plateau may extend into 2027, which is the basis on which the market is willing to value its 2027 EPS. Any major or CXMT re-allocating capacity back into niche DDR4, or downstream demand migrating to DDR5 faster than supply exits, would dissolve "scarcity pricing."

Thesis 3: CUBE is a long-term option, outside the current valuation

CUBE adds long-term option value.

Winbond operates outside the mainstream HBM market (only SK hynix, Micron, and Samsung can make HBM); its path into the AI-memory narrative is CUBE, Winbond's 3D custom-memory platform. Beyond its NT$40B full-year 2026 capital spending, the company approved an additional NT$7.3B, of which CUBE line equipment exceeds NT$5B, and that increment is itself management's bet on this path. Technically, CUBE targets edge-AI scenarios; single die of 256Mb-8Gb can be 3D-stacked, positioned as low-cost, low-power custom high-bandwidth memory and described by several industry outlets as an edge-side HBM alternative; it competes for the new bandwidth and power demand of edge inference outside the head-on datacenter HBM contest. CUBE gives Winbond a differentiated AI exposure independent of the three HBM giants, but current revenue contribution is minimal, so it is a call option included only in the bull-case valuation. Persistent absence of a production-grade lead-customer design-win and an announced shipment milestone would reduce the option's time value.

Thesis 4: Separate the core memory business from consolidated Nuvoton

Core memory and Nuvoton require separate valuation treatment.

This is where the consolidated number most easily misleads when reading the financials. Core memory 1Q26 revenue NT$29.92B at 56.6% gross margin; consolidated revenue NT$38.25B, the ~NT$8.3B gap coming from Nuvoton (MCU) plus logic/foundry/specialty business. Nuvoton is consensus-estimated to be loss-making in 2026 (about −NT$1.27 EPS), and it dilutes Winbond's headline profitability on two levels: one, it pulls consolidated gross margin down from the core memory 56.6% to 53.4%; two, minority interest (Winbond holds only ~61.6%) dilutes again at the attributable-to-parent level. The core memory 56.6% gross margin and 42.5% operating margin therefore provide the clearer measure of Winbond's memory franchise. A widening Nuvoton loss or an impairment would amplify consolidated earnings volatility, making consolidated EPS more fragile than the core.

2.1 Product lines and segments (2026Q1)

Winbond's consolidated statement comprises two blocks: the core memory business (78% of consolidated revenue) and non-memory (~22%, mainly consolidated Nuvoton). Core memory 1Q26 revenue NT$29.92B at 56.6% gross margin and 42.5% operating margin is the absolute bulk of the company's value. Its internal structure:

• Niche DRAM (~60% of memory revenue, ~NT$17.95B): DDR4/DDR3/LPDDR4, specialty, consumer, and niche applications on its own 16nm process; end-markets cover TVs, servers, networking, industrial, embedded. This is the most elastic part this period, with DRAM ASP up more than 50% in a single quarter directly driving the 1Q26 gross-margin jump.
• Flash (~40%, ~NT$12B): NOR Flash world #1 (23% share), the company's steadiest moat; NOR is a highly fragmented market with high spec-qualification barriers, low unit price but strong stickiness, a perennial cash-flow contributor; Serial/SLC NAND world #3 (10% share), with 24nm SLC NAND rising as a share of shipments and carrying better margin than NOR, a source of structural improvement.
• Custom memory CUBE: Winbond's 3D custom-memory platform for edge-AI high-bandwidth applications, with mass-production contribution still small.

Non-memory ~NT$8.3B: consolidated Nuvoton (MCU/microcontroller, ~61.6% stake) plus logic/foundry/specialty business. This block drags consolidated gross margin and should be isolated in analysis.

The three businesses have different risk-return profiles, and stacked together they form Winbond's "barbell" structure. One end is high-elasticity niche DRAM: a profit amplifier in an upcycle (ASP +50% in a single quarter) and a profit grinder in a downturn, with the 2023 loss coming mainly from this end. The other end is the NOR Flash steady-state moat: a highly fragmented, low-unit-price market with long spec-qualification cycles where new entrants lack economics, and the world #1 23% share gives Winbond a through-cycle cash-flow floor whose earnings volatility is far below DRAM. SLC NAND sits in between, with the 24nm node's margin improvement a structural plus. The barbell implication: DRAM sets the stock's elasticity and volatility, NOR sets the floor and resilience of earnings, so linearly extrapolating the whole company from DRAM's peak elasticity overstates its downturn fragility, while valuing the whole company off NOR's stability understates its upcycle explosiveness; the micro-level reason a strong cyclical must be valued in segments.

2.2 Customer structure and pricing logic

Winbond's downstream is highly fragmented, covering many long-tail markets; consumer electronics, networking, industrial, automotive, servers, embedded; entirely unlike the majors' model of "a few mega-CSPs buying concentrated via LTA." This fragmentation is both a weakness and a strength: the weakness is the lack of a single large customer's scale premium and pricing voice; the strength is that demand remains relatively stable across any one CSP's capital spending cadence, and customer stickiness comes from long-life-cycle products, strict spec-qualification barriers, and supply-continuity requirements; a niche DRAM/NOR part qualified for automotive or industrial use faces high switching friction over a short-term price gap. Pricing is supported by qualification requirements and scarcity: in an upcycle, downstream buyers passively accept price increases because they cannot find alternative supply, ASP elasticity is extreme (DRAM ASP +50% in a single quarter), and because capacity is short-term rigid and manufacturing overhead already amortized, the revenue increment falls almost entirely to gross profit, producing the high operating leverage of "revenue gains drive gross margin +11pp in a single quarter." That leverage is double-edged: in a downturn ASP reverses just as violently, manufacturing overhead cannot be cut quickly, and gross margin falls at the same steep slope, which is why Winbond's gross margin collapsed toward ~20% or below in 2023.

2.3 Capacity and capital intensity

The company is headquartered in Taiwan, with main fabs in Taichung and Kaohsiung; the Kaohsiung fab expands from 15K to 24K wafers/month by end-2026, the main source of medium-term capacity elasticity. 2026 capital spending NT$40B, plus the additional NT$7.3B approved (CUBE and expansion), totals nearly NT$47.3B, typical heavy-asset operation. The in-house 16nm DRAM process is the foundation of its cost competitiveness in niche DRAM, but it also requires sustained R&D and equipment investment to keep up with process evolution. The heavy-capital spending structure has two implications: one, OCF is strongly supported at cyclical highs, but FCF is heavily suppressed by capital spending, and free-cash-flow release waits for the window after the expansion cycle ends while ASP is still high; two, the expansion itself reinvests "the price-increase dividend" into "future capacity," and if the capacity ramps 18-24 months out just as the cycle rolls over, that new capacity turns from a profit engine into a depreciation burden. Core memory warrants separate observation because the consolidated statement includes Nuvoton and minority interest, and the consolidated number understates the core's earnings quality.

3.1 Cycle positioning

At mid-2026, the memory industry sits at the absolute high of a structural AI capital spending-driven super-cycle; this round reflects a structural step-up in demand and a deliberate ceding on the supply side, setting it apart from an ordinary commodity-price rebound. Industry data show the magnitude: 1Q26 global DRAM industry revenue +81% QoQ; 2Q26 conventional DRAM contract prices +58~63% QoQ and NAND Flash contract prices +70~75%; full-year 2026 DRAM spot/contract prices have more than doubled versus early 2025. The supply-demand gap is at its widest since 2011 (DRAM/NAND/HBM ≈ 4.9%/4.2%/5.1%), DRAM inventory is as low as ~2-3 weeks, and supply growth (~16%) structurally lags demand growth (~35%). The root cause of the gap is that the three majors concentrate limited wafer capacity into AI/HBM/enterprise, actively squeezing conventional DRAM/NAND supply, driving across-the-board price increases and forcing CSPs into multi-quarter LTAs to secure supply. For a niche maker like Winbond, it sits in the "spillover-beneficiary" position of this gap; the conventional-capacity gap the majors vacated is precisely its home turf.

Using the HBM competitive map as background positions Winbond more precisely. The highest-value HBM link is made by only three players (SK hynix ~62% share, Micron ~21%, Samsung ~17%) and uses annual pricing, so 2026 contract prices still partly lag the spot surge, and TrendForce expects HBM contract prices to multiply again in 2027. This means the thickest profit in memory is being captured by these three in HBM/enterprise, so they have every incentive to pull every available wafer out of conventional DRAM. Winbond is the mirror beneficiary of that withdrawal; it captures the price rise from supply vacated by HBM while sidestepping HBM's capital arms race and yield risk, earning "the price increase, under shortage, of the conventional demand the majors cannot be bothered with." This is a structurally steadier model with a lower ceiling.

3.2 Winbond's position in the value chain

Winbond sits in the "niche / mature-node" link of memory, specializing in filling the DDR4/DDR3/niche-DRAM, NOR, and SLC NAND gaps left after the three majors pivoted to HBM/enterprise. It operates outside mainstream HBM (only SK hynix, Micron, and Samsung do), so its AI exposure is indirect, mainly via two paths: one, edge custom high-bandwidth memory like CUBE; two, the niche DRAM still used in large volumes in AI servers (mature-node memory around management controllers, NICs, BMCs, and the like). This "indirect AI exposure" means Winbond can share in the AI-driven aggregate upcycle while staying outside a head-on contest with the three giants in HBM's capital and technology arms race; a presence with a different risk-return structure.

3.3 Peers and comparability

Nanya is the cleanest comp: a pure niche / mature-node DRAM player, supported by the same concentrated DDR4 supply structure as Winbond, with 1Q26 revenue US$1.55B, +60% QoQ, in the same direction as Winbond's DRAM +91.4%. The difference is that Winbond also carries NOR/SLC NAND and Nuvoton, a more diversified business; Nanya is a purer DDR4 scarcity-pricing beta. As the large-cap comparison, Micron provides a reverse reference; its 2026 earnings are closer to an absolute peak (FQ3 FY26 non-GAAP gross margin 84.9% at a historically exceptional level, the last cycle bottom FY23 gross margin −7.7%). Even the strongest industry player carries valuation risk on a low forward PE at the peak; Winbond adds a leg of "strength may extend into 2027."

3.4 Specific exposure to the China CXMT threat

DDR4/niche DRAM has historically been the link China (CXMT, ~7.7% DRAM share) breaks into first, so in theory CXMT is a direct competitive threat to Nanya and Winbond. The current picture is easing: CXMT is shifting capacity from DDR4 to DDR5/HBM to chase higher-value markets, which reduces niche DDR4 supply pressure. The key stock-specific risk is a forced reversal if technological or export-control obstacles in DDR5/HBM push CXMT to redeploy capacity into niche DDR4 at low prices. Such a move would weaken the concentrated DDR4 supply structure and Winbond's pricing power. Market chatter currently puts CXMT's DDR4 ASP 5-10% below the three majors, with overall behavior showing profit-oriented output discipline, while DDR4/niche is exactly the link China can most easily break into first. The tightness of US semiconductor export controls on China directly modulates this risk; tighter controls could push CXMT back into DDR4.

4.1 Quarterly trend (consolidated)

The key to reading this table is the transmission of operating leverage: consolidated gross margin +11pp QoQ, with operating margin 32.8% and EBITDA margin 42%, showing that part of the gross increment from price increases was diluted by expenses on the way down (the consolidated level carries Nuvoton's expense drag), yet it still retained a high profit-conversion rate. EPS NT$2.25 at an all-time high and beating consensus is the endpoint of all this. A reminder: this is the consolidated basis, net profit including minority interest, and the core memory business standalone (56.6% gross margin, 42.5% operating margin) is more profitable than the consolidated number.

Unpacking the gross-margin bridge is clearer: the ~3.2pp gap between core memory 56.6% and consolidated 53.4% comes almost entirely from the drag of Nuvoton and the logic/foundry business; and the core's quarter-over-quarter gross-margin jump is driven mainly by DRAM ASP (+50%+) and Flash ASP (+30%+) both leaping, with volume secondary. This "price-driven" margin improvement is the purest form of operating leverage; because niche memory's capacity is rigid on a quarterly scale and manufacturing overhead and depreciation are already amortized over existing output, each notch of ASP increase falls almost dollar-for-dollar to gross profit, then amplifies to operating profit after relatively stable operating expenses. When ASP reverses, the same leverage bites back at the same slope and the operating-margin decline outpaces revenue, as it did in the 2023 fall from a high into losses.

4.2 Monthly revenue (consolidated)

The shape of monthly revenue tells the momentum story better than a single quarter: the YoY rate jumps from +88% in February to +182% in April-May, the slope accelerating rather than decelerating; the consolidated revenue of just April-May (NT$39.2B) already exceeds all of Q1 (NT$38.25B), and adding June, the Q2 sequential step-up over Q1 is nearly certain. This is the most direct evidence extending the "inflection" from the results-disclosure date to the present, and why the short-term fundamental direction is undisputed; the dispute is only over how long this high level can last.

4.3 Consensus benchmarking

Comparing current hard data with consensus reveals the direction of estimate revisions and the implied assumptions:

• 2026 revenue estimate lifted from NT$174.9B to NT$198.1B, reflecting systematic sell-side upgrades after the Q1 results.
• 2026 EPS estimate lifted from the earlier/conservative NT$7.52 to the recent NT$10.98; given Q1 already at NT$2.25 and Q2 stepping up, the full-year EPS could realistically land at NT$11-15, meaning consensus may still be too low with room for further upgrades.
• 2027 consensus EPS ~NT$20.9 (next-financial-year basis), implying 2027 earnings double again versus 2026; the market has priced "niche-DDR4 scarcity extending into 2027" directly into earnings estimates, exactly the evidence that the bull camp's core assumption is being priced.
• Target-price mean ~NT$192 (range NT$121-255); another basis gives NT$145.7 (range NT$100-200). The dispersion is unusually wide, with the upper and lower bounds more than 2x apart, itself a quantified expression of high disagreement over "how long the cycle can last."

4.4 Through-cycle normalized floor (empirical)

For strong cyclicals, treating peak earnings as the norm creates substantial valuation risk. Winbond offers a clean empirical floor: in the last downturn (2022-2023 memory oversupply), the company fell into losses or near break-even, with gross margin collapsing from cyclical highs toward ~20% or below. This accounting pattern already occurred last cycle. The inferred through-cycle midpoint uses a gross margin of 30-35% and normalized EPS of ~NT$3-6. Compared with the current 53.4% gross margin and ~NT$9-13 annualized EPS, the normalized range shows that earnings are near a cyclical high. Peak earnings and normalized earnings therefore require separate scenario weights before applying a forward PE.

4.5 Capital and cash (basis note)

2026 capital spending is NT$40B plus an additional NT$7.3B for CUBE and expansion, while the Kaohsiung fab expands from 15K→24K wafers. At cyclical highs, OCF is strongly supported by earnings while FCF is suppressed by capital spending. Quarterly OCF, capital-spending, FCF and net-debt figures provide the cash-conversion reference.

Peer multiples provide a reference, three-scenario probability weighting forms the primary valuation center, and DCF defines sensitivity bounds (see §10). Peak and normalized earnings are valued separately to reflect the scenario dispersion of a strong cyclical.

5.1 Peer table

The peer-multiple method has inherently limited reliability at the top of a strong cycle, because the denominator (forward EPS) is at peak, and the "low multiple" on the numerator side is the top-of-cycle norm rather than a cheap signal. Winbond's forward PE of ~17x for 2026 and ~10x for 2027 is structurally identical to Micron's pricing on a low forward PE at the peak; its reference value lies in cross-sectional comparison, instead of an absolute valuation reference.

5.2 Three scenarios (probabilities sum to 100%)

The design logic of the three scenarios: the bull bets on "durability" (DDR4 scarcity holds past 2027, CUBE delivers on the option), pairing peak 2027 EPS with a medium multiple; the base bets on "high but eventually normalizing," pairing through-cycle EPS with a higher multiple (a durability premium); the bear bets on "the top is near," pairing normalized EPS with a normal multiple plus a one-time de-rating. Probability-weighted midpoint: 0.30×330 + 0.45×185 + 0.25×90 ≈ NT$205 (≈US$6.4). Note the shape of the distribution; the bull and bear per-share-price references are nearly 3.7x apart (330 vs 90), a fat left tail, which means that although the weighted midpoint is close to the current price, the uncertainty around it is enormous, and the "information content" of the expected value is diluted by the wide distribution.

5.3 Cross-validation and method divergence

The peak method (NT$210-294) and the normalized method (NT$72-108) diverge by more than 3x, far beyond 40%, indicating wide scenario dispersion. This divergence reflects in valuation the fundamental question "is this in fact the top of the cycle": believe in durability and you approach the peak method, believe in mean reversion and you approach the normalized method. This report sets the fair-value midpoint at the blend between the probability weighting (NT$205) and the conservative normalization, landing at NT$185-205, with Neutral at NT$190. The current price NT$206.5 implies about −8% versus the fair-value midpoint NT$190, and about 0% versus the weighted NT$205. Overall judgment: the fundamentals are real but already fully reflected in the market price, while sentiment/liquidity has pushed valuation into overshoot territory, so bullish momentum and valuation overshoot offset each other; hence Neutral. The peak/normalized divergence >3x supports Medium confidence.

Scenario probabilities shift with the operating indicators. Niche DDR4 8Gb contract prices continuing QoQ higher in 2026Q4 and 2027Q1, a production-grade lead-customer design-win for CUBE, and Nanya/Winbond further extending LTAs would raise the bull-case weight. Niche DDR4 contract prices turning down QoQ for the first time, CXMT re-expanding DDR4 and dumping at low prices, downstream DDR4→DDR5 migration outpacing supply exit, and a widening Nuvoton loss would raise the bear-case weight. Until these indicators print new reads, the 30/45/25 allocation reflects the current balance of "fundamentals rising with limited distance to the top and a fat left tail." The current price already sits near the probability-weighted midpoint, while the downside left tail (bear-case value NT$90) is a halving-scale move once triggered; this risk-return structure supports Neutral.

Management's guidance on the 2026Q1 earnings call was positive and specific. CEO Chen Pei-ming was explicit that Q2 revenue and profit will grow significantly and that Q2 price increases will be at least equal to Q1, and laid out full-year 2026 capital spending of NT$40B plus the additional NT$7.3B approved (CUBE), reinvesting the current price-increase dividend into future capacity and new business in line with a continued-upcycle view. The chairman publicly warned in 2026-05 of a widening memory shortage and was pessimistic about whether relief would come before 2H 2027 (Nikkei-affiliated reporting, 2026-05-30), reinforcing that view. The QoQ direction of niche DDR4 contract prices will test the durability of this outlook. On capital allocation, the company maintains ~61.6% control of Nuvoton and pushes the Kaohsiung 15K→24K expansion via heavy capital spending.

Company guidance and consensus define the next 4-6 quarters. The short term (Q2-Q3 2026) has extremely high visibility: the company guides Q2 revenue and profit to grow significantly with Q2 price increases at least equal to Q1, monthly-revenue YoY stays three digits, and earnings keep setting records. The main uncertainty begins in late-2026 and extends into 2027; consensus has lifted 2026 revenue to NT$198.1B and 2027 EPS to ~NT$20.9, which amounts to pricing in strength extending into 2027 and earnings doubling again. The bull case assumes that the concentrated DDR4 supply structure lets Winbond's high-margin plateau peak later than the majors, extending the high-earnings platform into 2027 or beyond. The bear case assumes that capital spending ramps 18-24 months out while downstream demand migrates to DDR5, pushing the cycle to peak or roll over in late-2026 or 2027. The QoQ direction of niche DDR4 contract prices separates the cases; a first decline in 2026Q4 or 2027Q1 would falsify the bull case's durability leg. CUBE's mass-production milestone provides another outlook variable and would lift the bull scenario probability if delivered.

Below, catalysts and risks are paired as "event; observable signal," each bound to a trackable indicator to avoid vague statements.

8.1 Catalysts (each with an observable signal)

8.2 Downside risk matrix

The first four risks concern industry fundamentals and have observable operating signals. The final item reflects valuation and sentiment. A nearly 10x advance in a year, the June parabola and the 2x-leveraged memory ETF listing indicate elevated sentiment. Niche DDR4 contract prices remain the primary variable for the investment outlook.

Three lines warrant attention on the regulatory and geopolitical front. The first is US semiconductor export controls on China, which act on Winbond indirectly through CXMT's competitive cadence: tighter controls slow CXMT's DDR5/HBM upgrade, in theory favoring Winbond's scarcity pricing in DDR4, but could also force CXMT back into expanding niche DDR4 and pose a dumping risk; a two-way key. The second is the Taiwan Strait: as a Taiwanese manufacturer, Winbond's capacity and supply chain are highly concentrated in Taiwan, and the Strait situation is its systemic tail risk that cannot be hedged operationally. The third is the immediate energy and shipping disruption; the 2026-06-27~28 additional US strikes on Iran and Strait of Hormuz tension bring oil and shipping tail risk, a marginal effect on input costs and logistics for Taiwanese manufacturing. There is also a common macro headwind: Fed rate expectations have flipped from "rate cuts this year" toward "possible hikes" (CPI 4.2%, market pricing ~25% probability of a 7/29 hike), compressing market-wide duration valuations; memory has held up relatively on robust current fundamentals, but this means that once fundamental momentum weakens, the valuation side will face both duration and cycle de-rating pressure simultaneously.

For strong cyclicals, DCF serves only as cross-validation because it cannot support a standalone conclusion, because its output is determined almost entirely by two assumptions; "normalized earnings level" and "terminal multiple"; which are exactly the most uncertain parts for a cyclical. The core assumption ranges this report sets: through-cycle revenue NT$140-180B, normalized EBIT margin 25-30%, WACC 10-12%, terminal growth 2-3%. Across these assumptions, the valuation result is extremely sensitive to parameters; taking the optimistic end (normalized revenue NT$180B, EBIT margin 30%, WACC 10%) lands in the bull range; taking the conservative end (NT$140B, 25%, 12%) lands in the bear range. This extreme sensitivity is exactly the mathematical root of the >3x divergence between the peak method (NT$210-294) and the normalized method (NT$72-108). So this report uses DCF to frame the plausibility bounds of the three scenarios and to remind readers that any "precise" DCF target price on a cyclical is false precision.

1. Company 2026Q1 earnings call — 2026-05-05 — for: Q1 consolidated/memory-segment figures, capex, Q2 guidance, CUBE, expansion, share.
2. TrendForce "1Q26 DRAM industry revenue +81% QoQ" — 2026-06-01 — for: industry price increases, Winbond DRAM US$568M +91.4%, Nanya US$1.55B +60%.
3. TrendForce LTA feature (Nanya / Winbond lock-volume float-price) — 2026-01-21 — for: LTA lock-volume float-price structure, 2-year terms stretching to 2030, server-DRAM gap ~15%.
4. Company monthly revenue announcements — for: April NT$19.245B +182%, February NT$11.97B +88%, May NT$20.0B +182%.
5. Sell-side consensus aggregation — for: target-price mean NT$192 (121–255), consensus EPS.
6. Sell-side consensus — for: 2026 EPS NT$10.98, 2027 EPS ~NT$20.9, growth expectations.
7. Public market data — for: market cap NT$396.9B@2026-04-24 (share-count back-solve).
8. Winbond company IR / product pages on CUBE and industry media — for: CUBE/3DCaaS technical positioning.
9. Nuvoton company IR — for: Winbond ~61.6% stake, Nuvoton estimated 2026 loss.
10. Bank of Taiwan / public exchange rate USD/TWD — 2026-06-26 — for: USD/TWD 31.86.
11. TWSE quotes (2344 weekly/daily, 2408 weekly) — 2026-06-26/28 — for: price, multi-window returns, volatility, comps.
12. Nikkei-affiliated reporting and public earnings-call records — for: chairman's memory-shortage continuation warning, industry reads.