Nanya Technology (2408.TW): Commodity DRAM Upside Is Reflected in Valuation

The stance of this report can be summarized in one line: the momentum is real, the price is stretched, and discipline argues for a bearish lean. The bull option that structural scarcity grants Nanya is genuine; the earnings inflection has been validated by hard 2026Q1 numbers, and the company's better-than-Q1 guide will most likely deliver. But at NT$449 the price has already priced in 2026–2027 peak earnings almost in full, and on the assumption that they are sustainable. The market applies near-peak forward EPS to back out a forward P/E of 7.75x, manufacturing the appearance of "cheap" while masking the hard constraint that Nanya's through-cycle earnings center is very low (mid-cycle EPS NT$3–8, trough losses). Cyclical-stock valuation discipline requires separate probability weights for peak and normalized earnings, with annualized single-quarter peak EPS treated as one scenario input. On that basis we land on neutral-bearish, fair-value center NT$385, implied about −14%, confidence medium. This conclusion acknowledges Nanya's improving fundamentals and prices the risk embedded in cycle-top earnings.

Nanya is the highest-operating-leverage beneficiary of this commodity-DRAM up-cycle and, at the same time, the name most directly exposed to CXMT's DDR4 competition. Earnings momentum is real, while valuation is stretched.

Price action and operating results point to distinct conclusions. Nanya is up about 17x from its 2025-01 low, with the trailing 12 months YoY +745%, 6 months +137%, 3 months +105%; it touched NT$505 in the week of 2026-06-22 before pulling back, then fell 5.87% on 6/26; the chart already shows a parabolic shape. The dominant operating driver is the structural climb in commodity-DRAM contract prices, with the inflection already validated by the 2026Q1 results, while memory-theme sentiment added a liquidity premium as memory ETFs rose 14.2% on the week and a 2x leveraged product approached listing. These factors explain why the current price already exceeds most earnings-based sell-side targets. Earnings momentum is real, while the final leg of the rally carries a clear sentiment and liquidity component. The operating data confirm momentum, while the price data show an extended valuation.

The three majors began shifting DDR4/LPDDR4 capacity toward HBM/DDR5/enterprise from 2025Q2. This reduced commodity-DRAM supply, with DDR4 capacity by end-2026 potentially falling to roughly a quarter of early-2025 levels and the shortage spilling into DDR3/DDR2. The resulting scarcity supports pricing power for the specialist commodity-DRAM suppliers Nanya and Winbond. Nanya's DRAM-only mix provides the highest operating leverage to this demand path.

Thesis one: the scarcity benefit in commodity DRAM is Nanya's strongest stock-specific alpha, and pricing reflects scarcity rather than cost

From 2025Q2 the three majors successively trimmed DDR4/LPDDR4 supply and shifted capacity toward HBM/DDR5; TrendForce estimates a 19–20% DDR4 supply gap in 2H26 (as cited in Morgan Stanley's memory-industry report). Nanya's 2026Q1 gross margin of 67.9%, against a full-year 2024 loss and only 22.5% for full-year 2025, jumped directly on operating leverage; depreciation and wafer-fab costs are relatively rigid, so a doubling of ASP flows almost entirely to gross profit. The shortage has spilled into DDR3/DDR2: TrendForce estimates DDR2 contract prices rose +55–60% QoQ in 2026Q2 and a further +35–40% in Q3. The company guides Q2 gross margin to approach 70%, with double-digit Q2 contract-price increases already locked. The earnings inflection is validated, with price beta adding to the move.

Sequential weakness in DDR4/DDR3 contract prices, or an announced refill of commodity capacity by the three majors, would narrow the scarcity benefit.

Thesis two: four customer-grade strategic shareholders bought in via private placement and signed long-term agreements, wiring Nanya into the AI supply chain

In 2026-04 the company completed a NT$78.7B private placement (about 351M new shares at NT$223.9/share), with subscribers SanDisk (about 138.7M shares / NT$31.05B ≈ US$1.0B, about a 3.9% stake), Kioxia (about 70M shares / NT$15.6B), Solidigm (wholly owned by SK Hynix) and Cisco (each about 71.4–71.5M shares); all four are Nanya customers, together holding about 10% of post-raise equity, and they signed multi-year supply agreements (LTAs) totaling about US$2.5B. Customer equity and long-term purchase commitments support demand visibility, reduce volume risk in a downcycle and explain the confidence behind the sharp capex expansion. These commitments add demand visibility.

LTAs are mostly floating-price (mark-to-market), so the dollar value shrinks as prices fall; if customers redirect orders toward CXMT or their own capacity, the long-term-contract protection is discounted.

Thesis three: process catch-up (1B / 10nm-class) and the DDR5 ramp determine whether Nanya can push from "commodity" into the mainstream increment

The 1B process already accounts for about a quarter of monthly wafer starts and can contribute about 40% of die output; 16Gb DDR5 and 8Gb/4Gb DDR4 are in volume production and shipping. 2026 capex was raised sharply to NT$52.0B (about 2.7x), funding the 1B process, the new Taishan 5A fab, cleanrooms and AI-memory R&D; new capacity tools in early next year, with a plan to nearly double capacity over 2–3 years, and the new Taishan fab tooling in early 2027 and ramping in 2H27. Nanya still trails the three majors by one to two generations, while HBM exposure is zero; DDR5 and 1B are its pass mark for maintaining competitive relevance. Earnings depend on capacity and process delivery, while the expansion raises late-cycle risk.

A shortfall in 1B yields or DDR5 customer qualification, or a new-fab depreciation and capacity burden during a price downturn, would weaken this thesis.

Thesis four: applying near-peak 2026E EPS to a forward 7.75x looks cheap, but it masks extremely low normalized earnings and trough losses

A forward P/E of 7.75x implies the market is using forward EPS of about NT$58; FactSet's 2026E EPS consensus is NT$40.46 and the consensus target is only NT$305. Against the normalized floor: 2025Q1 net loss of NT$1.941B (EPS −0.63), 2025Q2 net loss of NT$4.11B, a full-year 2024 loss, and FY2025 full-year EPS of just NT$2.13. The empirical mid-cycle EPS is NT$3–8, with the trough negative. The current price implies the market is extrapolating 2026–2027 peak earnings as a sustainable baseline; cyclical-stock discipline requires weighting peak versus normalized separately, and a low forward P/E read in isolation is the most common pricing trap at a cycle top. Peak earnings and normalized earnings remain distinct valuation cases.

LTAs and customer equity that materially reduce Nanya's earnings volatility, such as gross margin holding at 40%+ even in a downcycle, would require a higher normalized floor and strengthen the "cheap" case.

Nanya is a pure-play DRAM producer with direct sensitivity to contract prices; relative to Winbond (memory + logic + consolidated Nuvoton), its business mix is cleaner, which gives it higher beta leverage in an up-cycle and a more extreme sensitivity to a single variable (ASP).

Product lines: pure DRAM, weighted to commodity / mature nodes

The mainstay is commodity / consumer / industrial DRAM; DDR4 (8Gb/4Gb the volume shippers), DDR3, DDR2; plus 16Gb DDR5 now in volume production; the product mix is limited to DRAM, with NAND, NOR, HBM and HDD exposure at zero. On a bit basis, DDR4/DDR3 remain the revenue base, with DDR5 the increment. HBM exposure is zero, so earnings capture AI-memory economics indirectly through the commodity gap left by the three majors' capacity reallocation.

The direction in which the product mix evolves determines Nanya's medium-term fate. The current revenue base is still in mature nodes such as DDR4/DDR3; the source of this cycle's pricing dividend, but also exactly the node CXMT competes in head-on. DDR5 and the 1B process are Nanya's only route to migrate up the value chain; 16Gb DDR5 is already in volume production and shipping, while the 1B process is about a quarter of monthly wafer starts and contributes about 40% of die. If the DDR5 mix can keep rising over the next 2–3 years as the new Taishan fab ramps, Nanya's earnings quality improves from "pure commodity leverage" toward "mainstream DRAM participant," and the normalized floor rises with it; if the migration falters, Nanya stays locked into the mature node most easily breached by China. This is why thesis three (process catch-up) is a "pass mark" rather than a "bonus"; it bears on whether Nanya avoids being structurally marginalized.

The dispersion of downstream applications is one of Nanya's more resilient features. Commodity DRAM serves module houses, consumer electronics, industrial control, networking, automotive and some enterprise; a broad range of customer applications keeps single-customer dependence relatively dispersed, while the four post-placement customer-shareholders raise concentration moderately. This diversification reduces the concentration risk associated with one or two mega-customers. Direct exposure to high-growth, high-price applications such as AI servers and HBM remains limited, with the ceiling on growth set by total mature-node demand.

Customer structure and concentration

Downstream coverage spans module houses, consumer electronics, industrial control, networking, automotive and some enterprise. Post-placement, the four strategic customers (SanDisk / Kioxia / Solidigm / Cisco) are both shareholders and long-term-contract buyers, raising customer concentration via the LTAs. Geographically it is weighted to Asia-Pacific (including the mainland China module ecosystem) plus the domestic Taiwan operations.

Monetization and gross-margin logic: a high-operating-leverage commodity-memory model

A textbook capital-intensive model; depreciation and wafer-fab costs are relatively rigid, so ASP is the dominant earnings variable. When ASP doubles, gross margin jumps from a loss zone to 60–70% (2025Q1 loss → 2026Q1 67.9%), and reverses just as violently on the way down. This non-linear, two-way amplification is the core of understanding Nanya's valuation: explosive earnings power on the way up, and reliable trough losses on the way down.

The math of operating leverage can be reverse-engineered from the gross-margin path. Full-year 2025 gross margin of 22.5% jumped to 67.9% in 2026Q1, a single-quarter sequential gain of +18.9pp, against revenue QoQ of +63.1%; gross profit grew far faster than revenue. That is the textbook signature of fixed costs (depreciation + labor + part of manufacturing overhead) being rapidly diluted while incremental revenue drops to gross profit at close to the ASP increase rate. The 2026Q1 net margin of 53.1% means more than half of every NT$ of revenue converts to after-tax profit, an extreme reading that appears in commodity memory during an ASP surge and has historically proved temporary. When ASP retreats, fixed costs remain sticky and gross margin falls faster than revenue; this is the cause of the back-to-back net losses in 1H25 (2025Q1 −NT$1.941B, 2025Q2 −NT$4.11B) and the reason valuation must fold trough losses into the normalized floor.

The business-model difference versus Winbond gives the two distinct leverage and risk profiles. Winbond's mix includes NOR Flash and SLC NAND that provide relatively smooth qualification-type commodity revenue, while consolidated Nuvoton drags overall gross margin (2026Q1 53.4%), partly hedging its cyclical swings; Nanya is a cleaner pure-DRAM single bet, with higher up-cycle leverage and less downside protection. The investment implication: Nanya suits a high-beta expression of the commodity-DRAM cycle and carries limited appeal as a through-cycle core holding.

Capital intensity and segment basis

2026 capex of NT$52.0B (about a 2.7x increase), together with the NT$78.7B placement, supports the 1B process and the new Taishan fab; this is heavy investment at a cycle peak, with capacity nearly doubling in 2–3 years. The business is concentrated in DRAM, while Formosa Plastics Group cross-holdings provide parent-group backing and financial support.

At mid-2026 memory sits at the absolute peak of a structural super-cycle driven by AI capex. Nanya's position is fully staked on the scarcity of commodity DRAM, and its largest tail uncertainty is CXMT.

Industry cycle position

TrendForce data show 1Q26 global DRAM industry revenue up +81% QoQ; 2Q26 conventional DRAM contract prices up +58–63% QoQ and NAND Flash contract prices +70–75%. The supply-demand gaps (DRAM/NAND/HBM about 4.9% / 4.2% / 5.1%) are the widest since 2011, DRAM inventory is around 2–3 weeks, and supply growth (about 16%) structurally lags demand (about 35%). The three majors are shifting capacity toward AI/HBM/enterprise, squeezing conventional DRAM/NAND supply; prices are rising across categories, with CSPs locking volumes via multi-quarter long-term agreements. Lenovo calls high memory prices the new normal lasting through 2030.

The nature of this cycle needs to be characterized clearly: it is a structural super-cycle driven by AI capex. Ordinary commodity inventory rebounds arise from restocking and small swings in end demand, with limited amplitude and short duration. This cycle centers on HBM resetting the entire supply-side capacity structure, with supply growth of about 16% structurally lagging demand of about 35%, the gap at its widest since 2011, and DRAM inventory compressed to an extremely low 2–3 weeks. The imbalance is a multi-year phenomenon created by the dislocation between the capex cycle and the AI demand curve. The implication for Nanya: this cycle's magnitude and durability far exceed an ordinary cycle, and 2026Q1 single-quarter EPS exceeding prior full years is the embodiment of that. A super-cycle also creates a forceful reversion when supply ramps en masse in 2027–2028 and the gap converges, making the price retreat structural.

Two facts must be acknowledged at once: the fundamental boom is real and rare (which supports the explosion in Nanya's earnings), and sector sentiment has entered an overheated zone (memory-theme ETFs up 14.2% last week, a 2x leveraged product about to list). Their coexistence makes current pricing especially difficult to judge. Strong earnings can obscure the sentiment component in price, while overheated sentiment can obscure the strength of reported earnings. Viewing them in layers: the earnings layer gives Nanya a fundamentally supported high-valuation floor, and the sentiment layer adds a liquidity premium on top of that floor, with the portion of NT$449 that exceeds mainstream sell-side targets contributed mainly by the latter.

Position in the value chain: a commodity / mature-node specialist

Nanya sits below the three majors' (Samsung / SK Hynix / Micron) mainstream and advanced nodes and alongside CXMT's catch-up. Its HBM exposure is zero, placing its participation in the commodity spillover beneath the highest-priced node of the AI-memory value chain. Only three players can make HBM (SK Hynix about 62%, Micron about 21%, Samsung about 17%); HBM uses annual pricing and contract prices could multiply several-fold in 2027, with the resulting economics accruing to those suppliers. Nanya profits from the commodity gap left behind by the majors' capacity reallocation. This cycle's pricing power comes from a structural supply-side exit: mainstream makers are moving DDR4/LPDDR4 capacity toward HBM/DDR5, and DDR4 capacity by end-2026 may fall to roughly a quarter of early-2025 levels.

Ranking Nanya along the memory value chain clarifies its position and risk. The highest-priced, highest-alpha node is HBM (only three can make it, annual pricing, several-fold in 2027), where Nanya's exposure is zero. The middle is mainstream DDR5 / enterprise DRAM and enterprise SSD, which Nanya has only just entered with 16Gb DDR5 and where its share is still small. The bottom is commodity / mature-node DDR4/DDR3/DDR2, Nanya's core market and exactly the node CXMT breaches first. This ranking shows that Nanya captured high pricing from supply withdrawal at the lowest node of the value chain, creating extremely high leverage with a shallow moat. Its difference from Micron is one of kind: Micron's 84.9% gross margin is underpinned by the structural moat of HBM, while Nanya's 67.9% rests on the majors' temporary withdrawal from DDR4. Micron therefore ranks higher on earnings quality and sustainability in the same up-cycle.

The remaining suppliers: Nanya and Winbond

Nanya and Winbond (2344.TW, commodity DRAM + NOR + SLC NAND) are the remaining major suppliers in 2027 DDR4 supply. Both are in the DRAM up-cycle and share the supply-demand gap, but Nanya is the purer commodity-leverage name. Winbond's 2026Q1 gross margin of 53.4% and EPS NT$2.25 are below Nanya, with the difference partly from business mix and the drag of consolidated Nuvoton.

HBM capacity crowding out: the supply-side root of Nanya's boom

The majors' capacity reallocation is the primary driver of this commodity-DRAM upcycle. The HBM content of a single AI chip rises from 96/192GB to 216/288GB, with HBM3E about two-thirds of 2026 HBM shipments and HBM4 ramping; HBM consumes far more capacity per wafer than conventional DRAM, so to meet the HBM allocations for platforms such as NVIDIA Rubin, the majors must hand mature-node (DDR4/DDR3) lines over to advanced nodes and HBM. This supply-side reallocation moves Nanya, a mature-node specialist with zero HBM exposure, into the position of residual-supply price setter. Nanya's earnings surge is the spillover of the HBM super-cycle, with its economics accruing through the mature-node vacuum left by other suppliers. The dividend's longevity therefore depends entirely on when the three majors refill commodity capacity and when CXMT fills in.

CXMT / China threat: the most important stock-specific risk

CXMT's global DRAM share is about 7.7%, focused on DDR4/DDR5 catch-up, precisely Nanya's frontline battlefield. Market chatter puts CXMT DRAM ASP 5–10% below the three majors (one account as low as $150 versus a global $300–400). CXMT has turned profitable for the first time and its IPO is imminent, heavily discussed on social channels in June; on top of that, 6/27 reports describe Apple lobbying Washington to allow purchases of CXMT memory. Commodity DRAM is the node China breaches first, and a large CXMT DDR4 ramp under limited restrictions would present a direct volume-and-price threat to Nanya. The prevailing industry view expects commercially rational pricing, while CXMT remains the largest tail uncertainty in Nanya's valuation. Tighter export controls would favor Nanya's competitive landscape by capping CXMT's expansion pace and extending Nanya's scarcity benefit.

The CXMT threat reaches Nanya through two channels. The first is supply volume: a ramp in CXMT DDR4 monthly capacity directly refills the gap the majors vacated, diluting the residual-supply pricing power Nanya now enjoys. The second affects contract pricing: if CXMT wins orders at ASPs 5–10% lower, it pulls the entire commodity DDR4 contract-price center down and compresses Nanya's high margins. The trigger timing of both channels depends heavily on policy. If large customers such as Apple are permitted to buy CXMT memory, demand is diverted to Chinese capacity and Nanya faces simultaneous volume and price pressure. This is why "CXMT monthly capacity/ASP and policy progress on large-customer order shifts" is listed as the highest-weight downside observation signal. The key current judgment is that CXMT's motive looks more like commercially rational pricing during a boom, which gives Nanya a time window of uncertain length.

The earnings inflection has been validated, and 2026Q1 single-quarter EPS of NT$8.41 already exceeds prior full years, underscoring the extreme ASP leverage and the scale of downside under mean reversion.

4.1 Quarterly trend

Against consensus: 2026Q1 EPS of NT$8.41 and the full-year NT$40.46 linear run-rate of about NT$10/quarter cover different periods and require separate comparison. Q1 delivered about 20%+ of the full-year consensus, while the higher Q2 guide preserves the implied second-half step-up. The earnings inflection has been validated, with earnings as the primary driver and valuation beta as a secondary contributor. Reported results and higher next-quarter guidance provide fundamental support for the price advance. The duration of these single-quarter earnings remains the key valuation variable.

On the quarterly sequence, Nanya swung from a 2025Q2 single-quarter net loss of NT$4.11B to a 2026Q1 single-quarter net profit of NT$26.058B in four quarters, an earnings swing exceeding NT$30B/quarter; this is an extreme sample of commodity-memory cyclical leverage. 2025Q4 already delivered EPS NT$3.58 and net profit NT$11.083B, and 2026Q1 more than doubled again to EPS NT$8.41, accelerating sequentially. Comparing 2026Q1 single-quarter annualized (NT$8.41×4 ≈ NT$33.6) with the FactSet full-year consensus of NT$40.46 shows the market already expects Q2–Q4 to keep climbing, with full-year EPS landing above NT$40; this is exactly the denominator behind the forward P/E of only 7.75x.

4.2 Annual trend

High-frequency monthly revenue tracking (Taiwan-listed monthly disclosure): 2026-02 NT$15.607B (YoY +586.71%); 2026-05 NT$27.670B (YoY +730.14%, MoM +8.55%, record high); cumulative Jan–May 2026 NT$102.248B (YoY +649.62%). May revenue growth slowed from prior months but the pricing thread held (broker comment), an early signal of momentum decelerating at the margin, and one to keep watching. The marginal slope of monthly revenue is a more sensitive leading variable than YoY: the YoY comparison against a high prior-year level persists through full-year 2026, but the MoM sequential slope directly reflects the current-period mix of contract price and shipment volume. May MoM of +8.55% is still positive growth and a record high, so the pricing thread is intact, but the sequential rate has eased from the highs of prior months, and together with spot DDR4 at only +0.28% week-on-week it forms a "prices still rising, slope flattening" combination consistent with the mid-to-late phase of a cycle.

4.3 Cyclical evidence (the normalized floor)

This cycle trough: 2025Q1 net loss of NT$1.941B (EPS −0.63), 2025Q2 net loss of NT$4.11B (widening sequentially), a full-year 2024 loss. Nanya's through-cycle earnings center is very low and the trough is reliably loss-making. Historical-peak reference: in the prior boom year (around 2018) annual EPS was on the order of NT$7–8; the current 2026Q1 single-quarter EPS of NT$8.41 already exceeds prior full years, underscoring how extreme this ASP leverage is while implying large downside on mean reversion. The empirical normalized mid-cycle EPS range is NT$3–8 (including trough losses), and this is the normalized-floor reference for valuation.

4.4 Capital structure and funding sources

The funding side saw two structural changes in 2026. The first is the NT$78.7B placement completed in 2026-04, issuing about 351M new shares at NT$223.9/share, directly replenishing equity capital and funding the 1B process and new-Taishan-fab expansion; the second is the rapid improvement in internal cash flow after FY2025 returned to profit, with the 2026Q1 single-quarter net profit of NT$26.058B itself a strong source of cash. Together they support the 2026 capex increase to NT$52.0B (about 2.7x). The capital structure cuts two ways: equity-funded expansion keeps financial risk relatively controlled, while the placement brings about +11% equity dilution, with the fully diluted share count reaching about 3.45B and market cap about NT$1.55T, so per-share earnings reflect the dilution. The company's 2026Q1 consolidated statements provide the balance-sheet reference for these figures.

From a valuation standpoint, dilution is an adjustment that must be folded in. Using the ~3.10B shares currently outstanding gives a market cap of NT$1.39T; using the fully diluted ~3.45B shares lifts it to about NT$1.55T (≈ US$49B). When comparing P/E and per-share value, the selected share-count basis shifts the conclusion by about 11%. This report's fair-value center and scenario per-share values already implicitly account for the downward pull of dilution on EPS, and the bear scenario specifically lists "placement dilution" as one of the factors depressing per-share value.

The peak method (forward 8x) points to "cheap," the normalized method (mid-cycle EPS NT$5 × 12–15x ≈ NT$60–75) points to "expensive by 5–6x," and the two diverge by more than 3x; the wide spread supports Medium confidence.

5.1 Peer comparables

Using Micron as an anchor quantifies Nanya's earnings-quality discount. Micron's FQ3 FY26 non-GAAP gross margin of 84.9% and EPS $25.11, with HBM3E and HBM4 sold out through CY2027 and multiple take-or-pay LTAs in hand, give its high margin the backing of an HBM structural moat. Nanya's 2026Q1 gross margin of 67.9% is about 17 percentage points below Micron's, and the gap maps exactly to Micron's HBM exposure and process lead versus Nanya's zero HBM exposure and process lag. Both move in the same direction with the same source of leverage in the up-cycle, but Micron's peak earnings rest on greater durability, while Nanya's depend more on the relatively fragile condition of tight commodity supply. Nanya's lower forward P/E therefore reflects weaker earnings quality and sustainability, the key distinction in a peer comparison that can otherwise be distorted by low-P/E optics.

5.2 Three scenarios (probabilities sum to 100%)

5.3 Peak vs normalized cross-check

Why the low forward P/E is a trap can be shown by examining price and EPS separately. Forward P/E = price ÷ forward EPS. 7.75x is low because the denominator uses near-peak 2026E EPS (the NT$40–58 range), while through-cycle normalized EPS provides the relevant comparison. Using normalized mid-cycle EPS (NT$3–8), the same price implies a P/E of 50–150x. The appearance of "cheap" therefore comes from peak EPS and offers limited evidence of genuine undervaluation. The correct approach for a cyclical stock is to anchor the floor with normalized EPS at a reasonable multiple (such as NT$5 × 12–15x ≈ NT$60–75), then fold the sustainability of the peak in via scenario probability. This is exactly why the fair-value center of NT$385 sits well below the current NT$449 and well below the bull value of NT$700 derived by extrapolating peak EPS.

The gap between the peak method (NT$700+) and the normalized method (NT$60–75) exceeds 3x, and a spread above 3x supports medium confidence. The divergence shows that Nanya's value depends heavily on the duration of peak earnings, and the 2026Q4–2027Q1 contract-price data will reveal the durability of that assumption. The valuation conclusion is a probability-weighted distribution, with the fair-value center of NT$385 as its weighted center.

5.4 Valuation conclusion

• Current price NT$449 (2026-06-26). Fair-value center NT$385 (range NT$305–700, wide dispersion).
• Implied up/down about −14% (the price sits above MS NT$380 and FactSet NT$305, below only the most aggressive NT$805).
• Confidence medium (peak/normalized divergence >3x, wide scenario dispersion, CXMT tail uncertainty).
• Stance neutral-bearish (acknowledging the bull option from structural scarcity, the price has already run ahead of mainstream targets).

The company is led by President Pei-Ing Lee. On the 2026Q1 earnings call and at the 5/21 AGM, management gave clear guidance: 2026Q2 better than Q1, gross margin approaching 70%, double-digit Q2 contract-price increases already locked; management expects DRAM supply tightness to extend into 2027; and estimated 2026–2029 global DRAM bit-demand CAGR at about 16–20% (pulled by AI PCs, smartphones, robotics and servers).

Capital allocation

2026 capex of NT$52.0B (about 2.7x), the NT$78.7B placement bringing in four customer-shareholders and US$2.5B of LTAs, expansion nearly doubling capacity over 2–3 years. The Formosa Plastics Group cross-holding structure provides parent-group backing and financial support. After FY2025 returned to profit the company has dividend capacity; the specific payout is subject to AGM resolution (the confirmed dividend amount remains pending).

Strategic shareholder structure (post-placement)

The signal value of customer-shareholders buying in is that it pre-sells Nanya's output and underwrites it from the demand side, structurally lowering downcycle tail risk. But the LTAs are mostly floating-price, so the dollar value shrinks as prices fall, and the degree of de-cyclicalization protection can only be truly tested in the next downcycle.

On governance and backing structure, Nanya's particularity is that it has two layers of support at once. The first is the Formosa Plastics Group's cross-holdings and parent-group financial support, historically the key to Nanya surviving losses through prior downcycles; Nanya has repeatedly lost money through cycles yet always survived, directly tied to the group's capital tolerance. The second is this cycle's new layer of four international customer-shareholders (SanDisk/Kioxia/Solidigm/Cisco), at once buyers and shareholders, with interests bound to Nanya's capacity utilization. The two layers together give Nanya higher "survival certainty" than a typical independent commodity maker. Sustainable high margins remain governed by the ASP cycle, which continues to set the normalized earnings floor.

The key observation point on capital-allocation discipline is this: expanding aggressively at 2.7x capex at the absolute peak of the boom is a pro-cyclical decision. Pro-cyclical expansion amplifies growth if the bull extends and amplifies depreciation and impairment pressure if the cycle turns. Management justifies the expansion with judgments that "DRAM supply tightness extends into 2027" and "2026–2029 bit-demand CAGR of 16–20%," which are themselves bull assumptions. If they deliver, the expansion is correct; if 2027 tops out early, the capacity formed by this capex becomes a burden on utilization and depreciation during the price downturn. The rightness or wrongness of capital allocation ultimately returns to that same falsifiable contract-price metric.

Company guidance is clear and skews positive: Q2 better than Q1, gross margin approaching 70%, and management expects supply tightness to extend into 2027. The market accepts the current momentum; the debate concerns its durability into 2027–2028 and the timing of CXMT's effect on the commodity node.

Near-term guidance

• 2026Q2: revenue better than Q1, gross margin approaching 70%, double-digit Q2 contract-price increases already locked.
• Supply-demand outlook: DRAM supply tightness extends into 2027; TrendForce estimates a 19–20% DDR4 supply gap in 2H26, DDR2 contract prices +55–60% in Q2 and +35–40% in Q3.
• Demand structure: 2026–2029 global DRAM bit-demand CAGR of about 16–20%, pulled by AI PCs, smartphones, robotics and servers.
• Capacity cadence: 1B process about a quarter of wafer starts, contributing about 40% of die; the new Taishan 5A fab tools in early 2027 and ramps in 2H27; capacity plan nearly doubles over 2–3 years.

Outlook frame: turning "momentum extension" into a falsifiable question

The bull scenario holds if DDR4 scarcity extends into 2027–2028 and CXMT ramps later than expected; the bear scenario is triggered by 2027 contract prices turning down and commodity supply refilling. Spot is already showing buyer resistance to high quotes (DDR4 +0.28% week-on-week, TrendForce 6/24), and the May revenue deceleration is an early signal of momentum slowing at the margin. Nanya's own capacity nearly doubling in 2–3 years coincides with the supply-response window; a double-edged sword in the back half of the cycle: expanding at the peak goes with the flow, but after prices top, the new depreciation and capacity can amplify the downturn in reverse.

The supply response is the most important endogenous constraint of a strong cycle. The current boom comes from a passive contraction on the supply side (the majors' reallocation + limited commodity-maker capacity), and the price signal incentivizes everyone to expand: Nanya's 2026 capex up to 2.7x, the new Taishan fab ramping in 2H27, capacity planned to double over 2–3 years; CXMT expanding in step and raising IPO funds; the three majors potentially refilling some mature-node capacity once HBM allocations are met. Semiconductor capacity typically has an 18–24 month lag from capex outlay to actual ramp, which means the large-scale 2026 expansion releases en masse from 2H27 into 2028, overlapping exactly with the point at which demand growth may ease; the classic trigger structure for a price top. Nanya's own capacity-doubling plan is a growth engine in the bull and a double burden of depreciation and utilization when the cycle turns.

The demand-side support is relatively structural: the company estimates a 2026–2029 global DRAM bit-demand CAGR of about 16–20%, pulled by AI PCs, smartphones, robotics and servers. Even so, whether a 16–20% demand CAGR can keep sustaining the gap against the supply growth that ramps en masse in 2027–2028 is the core variable deciding whether the boom extends or tops. This is why the falsifiable metric is locked to the TrendForce quarterly commodity DDR4 contract-price QoQ; the most direct, highest-frequency price expression of the supply-demand gap.

Catalysts (with observation signals)

Downside risk matrix (ranked by weight)

Nanya has risen about 17x from its 2025-01 low and pulled back after recently touching NT$505. Earnings improvement drove the move, while liquidity and sentiment added a material premium. Memory-theme ETFs rose 14.2% on the week, and a 2x leveraged product approached listing, signaling elevated sector sentiment.

The interrelation of the four risks is worth noting. CXMT supply refill (highest weight) and the cycle-top ASP turn (high weight) are two ends of the same cause-and-effect relationship; the CXMT ramp is the cause, the ASP turn the effect, and the two most likely occur together or in sequence, forming the core trigger of the bear scenario. Placement dilution (medium weight) is a known discount item with a quantified impact (share count +11%). Price above targets (medium weight) is a valuation-layer risk that amplifies the downside elasticity of any fundamental deterioration; a price already ahead of mainstream targets produces a larger pullback when contract-price data weaken. Separating common-cause risks from independent ones, and known effects from forward uncertainties, limits double-counting and supports the relatively high 30% bear-scenario probability.

The first intersection of catalysts and risks on the timeline is the 7/10 earnings call. It is at once the nearest positive catalyst (Q2 actuals, Q3 ASP guidance) and the nearest falsification window (whether gross margin tops 70%, whether Q3 contract-price guidance is still strong). After the call, the decisive checks fall on the 2026Q3 and 2026Q4–2027Q1 TrendForce commodity DDR4 contract prices: the data points that truly resolve the binary judgment of "momentum extension vs cycle top."

The US 100% tariff threat on memory was specifically addressed on the company's earnings call; Nanya's direct US export exposure is limited, with advanced-logic and HBM exposure at zero, but global memory pricing and end demand would be indirectly affected by tariff disruption. The degree to which CXMT is subject to export controls is, conversely, the key variable shaping Nanya's competitive landscape: tighter controls would cap CXMT's expansion pace and extend Nanya's commodity-scarcity benefit, while looser controls (such as the lobbying to allow Apple to buy CXMT memory, if it materializes) would accelerate the breaching of the commodity node by China.

Geopolitics is a two-way variable for Nanya, and the direction must be distinguished. On one hand, export controls targeting China's semiconductors objectively suppress CXMT's equipment access and expansion pace, indirectly protecting Nanya's commodity pricing power; the tighter the controls, the longer Nanya's survivor window. On the other hand, tariffs and trade friction raise end-electronics costs and suppress downstream DRAM demand (Apple has already raised Mac/iPad prices, console memory costs are up), and this negative demand-side transmission hurts all DRAM makers. The net effect depends on the policy mix: tight controls on CXMT combined with resilient global demand would favor Nanya, while broader CXMT purchasing access combined with high tariffs and weaker demand would create the most adverse combination. Progress on Apple's lobbying to buy CXMT memory is the single event most worth tracking on this policy question.

The rate dimension has an amplifying effect on a high-valuation leveraged name like Nanya. Fed rate expectations have flipped from "cut this year" toward "possible hike" (the market prices about a 25% chance of a hike on 7/29), lifting both the risk-free rate and the discount rate in DCF; for a name whose value is heavily concentrated in peak cash flows over the next few years, every uptick in the discount rate more markedly depresses its present value. The memory sector has relatively withstood the duration pressure on the strength of current robust fundamentals, but that rests on earnings continuing to beat; once the contract-price data weaken, the valuation suffers a double hit of "earnings downgrade" and "discount-rate rise" simultaneously. This macro overlay is one reason confidence is held at medium rather than raised.

Macro and geopolitical overlay

• Rates: Fed rate expectations have flipped from "cut this year" toward "possible hike" (the market prices about a 25% chance of a hike on 7/29), compressing duration valuations market-wide; memory relatively withstands it on fundamentals, but high-valuation leveraged names are more sensitive to duration shifts.
• Geopolitics: the US 6/27–28 additional strikes on Iran and Strait of Hormuz tension constitute a tail risk for oil/shipping, an uncertain disturbance to costs and downstream demand.
• Overheating signal: memory-theme ETFs up 14.2% last week and a 2x leveraged product about to list mark sector sentiment entering an overheated zone, and warrant weight in risk management.

Commodity DRAM cash flows are extremely unstable, so DCF is used only for sensitivity; the three-scenario analysis provides the standalone conclusion. Its sole value is to quantify "how sensitive valuation is to the through-cycle-norm assumption."

Key assumptions

• 2026–2027 sustains the high (annual FCF in the tens of billions of NT$), normalizing from 2028.
• Terminal growth 2%, WACC 10–11%.
• Intrinsic per-share value is extremely sensitive to the "normalized gross margin" assumption.

The fundamental limit of DCF on a commodity-memory stock is that the model's most sensitive inputs (normalized gross margin, the terminal-phase ASP center) are exactly the least predictable variables. DRAM prices are set by the marginal balance of global supply and demand, and through-cycle swings can run from a loss zone (negative gross margin) to this cycle's 67.9%; any terminal assumption carries enormous subjectivity. Feeding a variable that resists reliable estimation into a DCF highly sensitive to terminal value yields a precisely wrong number. DCF therefore serves only as a sensitivity tool: a 15-percentage-point error in normalized gross margin can halve the valuation. The three-scenario probability weighting provides the primary pricing basis through discrete, testable assumptions about the duration of DDR4 scarcity and the timing of CXMT's ramp.

The triangulation conclusion is that the three methods roughly converge at the center and diverge sharply at the tails. The peer-multiple method (forward 7.75x looks cheap, needs an earnings-quality discount), the three-scenario probability weighting (NT$367), and the sell-side consensus targets (FactSet NT$305, MS NT$380) are roughly consistent in the NT$305–385 center band; the divergence concentrates at the bull tail; between the aggressive foreign-broker NT$805 and the normalized floor of NT$60–75 is a gap of more than 10x. The right response to this "center converges, tails diverge" structure is to accept the center as the baseline price and manage tail risk with scenario probability, rather than picking a side at the bull or bear extreme. The fair-value center of NT$385 is where this thinking lands.

Basis and confidence note: the core revenue/gross margin/EPS/price/target/placement/capex figures in this report are all backed by primary company disclosure or cross-verified against mainstream financial data; full quarterly cash-flow detail, net cash and the confirmed dividend amount were not obtained item-by-item from primary sources, as flagged in §04. With the peak and normalized methods diverging by more than 3x, wide scenario dispersion and CXMT tail uncertainty, confidence is set to medium overall. The currency basis is primarily NT$, with USD conversion at USD/TWD ≈ 31.5 (approximate, not a locked rate). The Morgan Stanley and FactSet targets are third-party sell-side views, shown for reference and not the conclusion of this report.