Nanya Technology

Refresh delta (since 2026-07-06)

What changed:

1. Q2 2026 printed far above the July frame. Revenue was NT$82.55bn (+68.2% QoQ), gross margin 79.5%, operating margin 73.7%, net income NT$50.19bn and basic EPS NT$14.66 on 3,423m weighted shares. ASP rose >60% QoQ on flat bits. H1 EPS was already NT$23.38, more than half of the FY26 base of NT$38–42 that the July dossier carried.
2. Q3 revenue is already known, and it stepped up again. July was NT$43.87bn (+49.3% MoM, a contract reset), August NT$44.69bn and September NT$45.09bn. That makes Q3 NT$133.65bn, +61.9% QoQ and ~+612% YoY. Nanya said July's jump came from expired annual, semi-annual and quarterly contracts rolling into new ones.
3. The capex cycle was formally launched on 2026-08-05. The 2026 capexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. budget rose from NT$52.0bn to up to NT$69.7bn. Fab 5A phase-1 capex of up to NT$346.6bn for 2026–29 was approved, including EUV, for 35,900 wafer starts per month (WSPM). The full fab is 45K WSPM at ~US$16bn (NT$480bn in the Q2 press release). Wafer input starts 2H27, ramping to 30K WSPM in 2028 and 35.9K in 2029. The board also approved a US$2.1bn capital injection into Nanya Technology International to reduce FX exposure.
4. Nanya took its first concrete HBM-adjacent step. On 2026-10-01 the NSTC approved a 3D wafer packaging and test fab in Pingtung Science Park, using TSV and wafer stacking, "for high-performance and high-bandwidth memoryHigh-bandwidth memoryMemory stacked in layers beside a processor so data reaches it faster. The usual bottleneck in AI chips. solutions". Together with the customized AI / wafer-on-wafer (WoW) projects disclosed on the Q2 call, this weakens the July framing of "no HBM, sitting out the category".
5. CXMT listed and moved up a node. It IPO'd on the Shanghai STAR market on 2026-07-16 and put its G5 DRAM platform into mass production in September. It is reportedly weighing a sixth fab and targets ~17% of global DRAM supply by 2028. In the company's own words, the legacy-tier competitor is now a funded public company.
6. The stock moved +22% since the July print. On a fixed share count, earnings estimates roughly doubled. The close was NT$435.5 on 2026-07-10 and NT$530 on 2026-10-05. Over six months the range was NT$205–543, with 55 of 125 sessions (7 Apr → 5 Oct) moving more than ±5% close-to-close. The trailing P/E is ~20.6x, but only ~5.3–5.6x on annualized Q3E earnings (Lens 7).

Also new, smaller: a former senior engineer was indicted in August 2026 for filming process documents for use in China. Nanya is the victim. Management guided a ~40% ± dividend payout on the Q2 call. The 2026 bit-shipment target was raised from "teens %" to "high-teens %".

What held: the structural reading still stands. Nanya is a one-node-behind commodity-DRAM maker whose pricing power is borrowed from big-three HBM diversion. About 80% of its revenue is still DDR4/LPDDR4/DDR3, and DDR5 is ~10%. Its own deck says industry "greenfield capacity expansion [is] aligning with multi-year LTAs, beyond 2028", which supports our timing for the turn. The supply chain (Lens 2) is unchanged.

Correction to the July dossier: it carried a FY26 base of NT$38–42 and a sell-side mean target "below spot (~NT$325–397)". H1 alone delivered NT$23.38. The July base anchored on a stale Street number (NT$42.55) and should not be cited.

---

Phase A — Understand the business

Company Overview — targeted delta

The business model is unchanged: Nanya designs and fabs commodity DRAM and sells to module makers, OEMs and, increasingly, cloud, eSSD and networking customers. Revenue is bit shipments × ASP. Changes since July:

• Contract structure is shifting toward LTAs. On the Q2 call, management described several LTA types. Some are fixed volume and fixed price, renewed at the end of the commitment. Others commit volume but set price "according to the market trend", mostly quarterly. Short-term contracts reprice upward; long-term ones are "relatively stable". Asked whether it would add take-or-pay terms, price floors or prepayments to prevent LTA defaults in a downturn, management said "we will consider that". So there is no take-or-pay floor today, which is the important point for the bear case.
• End-market mix: AI infrastructure and servers made up >20% of H1-26 revenue.
• New capability: the Pingtung 3D wafer-packaging and test fab (TSV, wafer stacking), plus customized AI / WoW projects that "several of those projects [are] going on", with leading-foundry (TSMC) logic involvement acknowledged. Management declined to comment on a press report that Nanya LPDDR had been designed into NVIDIA's Vera Rubin supply chain (customer confidentiality).

Supply Chain — unchanged, see the previous dossier

One correction to the placement map, now from primary sources: the three named subscribers at NT$223.9/share were Sandisk 138.685m shares, Solidigm 71.393m, Kioxia 70.0m (280.08m of the 351.578m total). The remaining 71.5m shares (~2.07%) went to a fourth investor that the company's press releases do not name. The Q1 deck says "four major customers", and the July dossier's web sources named Cisco. Placement shares are locked for three years, to April 2029. At NT$530 the placement holders are up ~137%.

Competitive Advantages — targeted delta

The July call that the moat is "rented from the cycle" still holds, with two adjustments:

• Upward: Pingtung 3D packaging plus WoW custom memory is the first evidence that Nanya is building a non-commodity product line. It is early, there is no revenue guidance, and it depends on "which customer may be more successful". Treat it as an option, not a moat.
• Downward: CXMT's G5 platform entering mass production in September narrows the node gap from the other side. Asked about Chinese DRAM makers, management said the impact is "not directly to Nanya". That answer reads as a description of today's shortage, not of a cost position.
• Bargaining power is still inverted in Nanya's favour. Management said supply is "very difficult to support double booking". The DDR4/LPDDR4 tier is more constrained than DDR5 because "customer can get DDR5 supply from more supplier" (same source). That cuts both ways: Nanya's scarcity is specific to legacy nodes.

Segments — refreshed (now research-layer)

Nanya has one reportable business: DRAM was 99.97% of Q2-26 revenue. The notes disclose ship-to geography (primary destination, not end customer):

. USA grew from NT$0.17bn to NT$6.94bn YoY, the fastest of any destination. That is consistent with US cloud demand and the placement partners pulling supply. Hong Kong plus China together are ~47%, a real export-control and geopolitical exposure that ship-to data understates or misattributes.

Product mix (Q2-26): DDR4 + LPDDR4 ~70%, DDR3/LPDDR3 ~10%, DDR5/LPDDR5 ~10%, other ~10%. DDR5 is capped because "we don't have sufficient capacity to do more". The trend: all of the growth is still price. Bits were flat QoQ in Q2 and ASP rose >60%.

---

Phase B — Measure performance

Earnings Result — Q2 2026 (latest reported), with Q3 revenue already known

• Vs consensus: Q2 EPS of NT$14.66 compares with a reported forecast of NT$13.34, a beat of ~10%. The revenue line was effectively known in advance from monthly prints (April–June sum = NT$82.55bn).
• Driver: ASP +>60% QoQ with flat bits. FX was flat. Operating expenses rose NT$1.6bn QoQ on employee profit-sharing.
• Margins: cost of goods sold rose only 7% QoQ (NT$15.76bn → NT$16.93bn) against +68% revenue, which is how gross margin gained 11.6pp. The tax rate rose to 19.5% (NT$12.14bn on NT$62.33bn pre-tax) because the NT$3.33bn deferred-tax asset was used up (NT$0.10bn left at June-end).
• Cash and balance sheet (reviewed, 2026-06-30):
• Cash NT$217.6bn and gross debt NT$18.4bn (bonds NT$4.0bn + long-term loans NT$14.4bn), so net cash NT$199.2bn. The company deck states NT$198.4bn using unaudited cash of NT$216.8bn.
• Equity NT$322.7bn; book value per share NT$93.49 on 3,450.2m shares.
• H1 operating cash flow NT$87.41bn and H1 capex NT$6.86bn, so Q2 FCF was NT$51.8bn. The deck's unaudited figure was NT$50.97bn.
• Guidance and tone (2026-07-10): "Q3'26 operational results expect to further improve", sustainable "in next few quarters". On Q4, management said "we don't see major opportunity of major change from Q4 versus Q3".
• What Q3 already tells us: revenue of NT$133.65bn is +61.9% QoQ, against Q2's +68.2% on flat bits. With bits capacity-capped (2026 target "high-teens" YoY, largely met from inventory), most of the Q3 step is again ASP.
• Market reaction to the Q2 print: the stock closed limit-up +10.0% on 2026-07-09, the day before the print. It was flat on 07-10 and −2.9% on 07-13. The print was pre-traded off the June monthly revenue release.
• Unusual versus history: an 80% gross margin quarter has no precedent. The Q2-25 comparison quarter had a −20.6% gross margin.

A conflict to know about: one Taiwanese outlet reported "Q3 revenue NT$88.96bn, +7.76% QoQ". NT$88.96bn is July + September only (43,868 + 45,091 = NT$88,958m), with August omitted. The company's own monthly figures sum to 43,868 + 44,690 + 45,091 = NT$133,649m, and the primary sum is used throughout.

Earnings Calls (sentiment trend) — refreshed from 6 primary transcripts

Grounded in transcripts/2025-q1.md → transcripts/2026-q2.md. This replaces the July dossier's summary-derived arc.

• Q1-25 (Apr 2025), cautious: demand was propped by "China's stimulus policy" in mobile and consumer, and the language was short-term.
• Q3-25 (Oct 2025), inflection: a DDR4 shortage followed peers' "big surprise announcement on the reduction of their supply". "DDR4 shortage likely to prolong for some quarters." Capex would stay low until the new fab in 2027, and "2027 will be a critical year".
• Q4-25 (Jan 2026), tight everywhere: "Supply shortage may persist for multiple DRAM products", with "limited new supply capacity expected for 2026 and first half of 2027". Q1 price up "roughly 20% ± range".
• Q1-26 (Apr 2026), peak confidence: "demand remain strong, healthy, and sustainable", and "high gross margin is sustainable in the next few quarters". The language of LTA tiers appears: "long-term LTA, middle-term LTA, short-term LTA, and also monthly deal".
• Q2-26 (Jul 2026), confident but hedging:
• The operating margin is "already reasonably high", Q3 "will be even higher", and "Q4 wise, it will subject to market situation change".
• Management acknowledged cyclicality for the first time in this upcycle: "DRAM industry is cyclical, especially in those low-ticket items", noting the consumer "ecosystem is tougher now".
• It conceded LTAs lack downturn protection ("we will consider" take-or-pay or prepayment).
• It put a number on the dividend: "40% ± range".
• Arc: survival (2024) → inflection (late 2025) → euphoria (Q1-26) → confident but hedging (Q2-26). Two phrases are new: "LTA" as the industry norm, and "cyclical … low-ticket items". "Structural transformation" and "through end-2027", which the July dossier attributed to April coverage, do not recur in the Q2 call. The tell to watch on 12 October is whether Q4 language moves from "no major change" to "improve", or to "subject to market".

Comps — refreshed (subject row now primary-derived)

Yahoo TW shows Nanya's P/E at 20.41 against a 54.78 "peer average" (Taiwan semiconductor group, not memory peers). The point: on trailing earnings Nanya looks ordinary. On the earnings it is printing right now it is priced at ~5.3–5.6× EPS (85% vs 80% Q3 gross margin) and ~4× EV/EBIT. That is the market capitalizing a turn, the same way it priced Micron and Hynix in June. EV/Sales, dividend yield and 5-year average ROE for peers are n/a. The Comps sheet in model.xlsx is still blank.

Stock-Price Catalysts — refreshed; Q3 print 2026-10-12

The date: the company IR calendar lists the "Third Quarter 2026 Earnings Conference" on Monday 12 October 2026, 15:00–16:00 Taipei (07:00–08:00 UTC), registration via webex. Nanya usually publishes its press release and deck on the same day as the call, and the Q2 materials were dated 2026-07-10, the call date. The internal earnings calendar (research/calendars/earnings.json) has 2026-10-13. That is one day late and should be corrected.

What to watch on 12 October (ranked):

1. Q3 gross margin vs ~85%. Revenue is known (NT$133.65bn), so the print is a cost and margin read. Q1→Q2 cost of goods sold rose only 7% on +68% revenue, so with bits flat, gross margin should exceed Q2's 79.5%. Taiwanese press reports analysts expect gross margin to hold around 80%. That is our Bear row (EPS ~NT$23.5, ~5.6× annualized). A print at 80% or below would mean costs or mix (DDR5, inventory depletion) are eating the price gains.
2. Q4 language. July said "no major change Q4 vs Q3". TrendForce (2026-09-30) projects conventional DRAM contract prices +10–15% QoQ in 4Q26, consumer DRAM "continue rising — albeit at a slower pace", and mobile growth moderating. If Nanya's Q4 guide upgrades, the turn is later than priced. If it repeats "subject to market", the stock likely trades the consumer weakness instead.
3. LTA terms. Did "we will consider" become take-or-pay, prepayments or price floors? That is the single biggest change that could lift the trough.
4. Capex cadence: H2 spend against the NT$69.7bn budget (H1 was NT$6.9bn), Fab 5A milestones (equipment move-in was guided for Q1-27, wafer input 2H27), Pingtung 3D-fab capex and customer, and the 2027 depreciation guide (asked in January, never quantified).
5. Dividend: confirmation of ~40% payout. On our FY26 base (~NT$250bn net income) that is ~NT$100bn, about NT$29/share, or a ~5.5% yield.
6. Bits and inventory: inventory fell from NT$27.3bn to NT$21.5bn in H1. "Most of the shipment is due to inventory consumption". A drained buffer caps Q4 and 2027 bits until Fab 5A.

Reaction pattern (last 6 months): 55 of 125 sessions (7 Apr → 5 Oct, close-to-close) moved more than ±5%, many at the ±10% Taiwan daily limit.

• Monthly revenue prints move the stock more than quarterly results. August 3–4 saw two consecutive +10% sessions into and on the July revenue release (+49% MoM). September 3 fell −7.8% the day after August revenue (+1.9% MoM).
• Quarterly prints are pre-traded: Q2 was +10% on the eve and flat on the day.
• Sharp drawdowns of −24.8% (Jul 27 close NT$436.0 → Jul 30 close NT$328.0) and −8.9% to −10% days in May and June. The specific cause of the late-July drawdown was not sourced. TWSE "disposition stock" (處置股) trading restrictions have repeatedly triggered limit-down cascades across Taiwan memory names in 2026.

What the market reacts to is unchanged from July: legacy-DRAM ASP, read monthly. Two new reactors are worth adding. First, the second derivative of monthly revenue: +1.9% and +0.9% MoM in August and September show the price step is flattening between contract resets. Second, regulatory trading curbs.

---

Phase C — Judge people & books

Management — targeted delta

• Same team: Dr. Pei-Ing Lee, President and Spokesman. Dr. Lin-Chin Su (EVP), Joseph Wu (SVP) and Philip Zhao (finance) appear on calls. Governance is still Formosa-group and parent-influenced. Nan Ya Plastics is disclosed as "the entity with significant influence over the Group".
• Capital allocation since July, all primary:
• It committed to Fab 5A phase 1 (up to NT$346.6bn, 2026–29) at what may be cycle peak. That is the July bear case's "worst capital-allocation risk", now board-approved.
• It raised 2026 capex 34% to NT$69.7bn, mostly for advance equipment payments.
• It is selling a sliver of PieceMakers (715k shares at ≥NT$545, gain ~NT$0.36bn, stake falls to 33.96%).
• It is moving US$2.1bn into Nanya Technology International to cut USD/TWD exposure.
• It signalled a ~40% ± payout ratio.
• Assessment: financing is disciplined. Phase 1 (NT$346.6bn ≈ 6.7 quarters of Q2 FCF of NT$51.8bn, or ~1.7× June net cash) is fundable from existing net cash plus roughly three quarters of current FCF, with no new equity. Whether the investment itself is wise is the open question. The board chose to lock in greenfield supply that lands in 2H27–2029, the window its own deck flags for industry greenfield capacity.
• Execution flags: the Ministry of Labor's February 2026 spring safety inspection found occupational-safety deficiencies at the expansion construction site, and Nanya committed to daily inspections. Separately, the trade-secret theft attempt was caught by Nanya's own IT security.

Forensic Red Flags — refreshed (now on reviewed statements)

• Earnings quality is high, but operating cash flow is temporarily flattered by unpaid tax. H1 operating cash flow of NT$87.4bn exceeds net income of NT$76.25bn. However, income taxes paid were only NT$0.67bn against NT$17.8bn of tax expense, and the current tax liability is NT$14.35bn. Taiwan's annual tax settlement falls in May, so ~NT$15–20bn+ of 2026 cash tax will hit 2027 operating cash flow. On a normalized basis, H1 FCF is ~NT$63–65bn, not ~NT$80bn.
• Receivables are growing with revenue, not ahead of it. AR was NT$40.2bn at June-end against NT$16.5bn at December. Days sales outstanding are ~44 (Q2) vs ~46 (Q1) and ~51 (Q4-25). Collection is improving, so this is not a flag.
• Inventory is falling into rising prices, from NT$27.3bn to NT$21.5bn in H1. Write-downs reversed (NT$11m). The write-down risk on the downturn remains structural.
• Employee profit-sharing is accrued in COGS and opex: other payables rose from NT$4.3bn to NT$14.0bn. Asked by BofA whether bonuses were "already expensed", Pei-Ing Lee answered "Some of the expense is included" (Q2, "also first quarter too") and "We'll continue to reserve cash each quarter for employee bonus". Some of the year's bonus cost is therefore in reported EPS, but not necessarily all of it. Treat a year-end true-up as a modest downside risk to Q4 margins.
• Dilution: 351.578m placement shares were issued 2026-04-08 at NT$223.9, for NT$78.72bn. Employee options add ~121m to the diluted count (diluted Q2 EPS NT$14.12 vs basic NT$14.66). Share-based payment expense in H1 was NT$175m, which is immaterial.
• Deck vs reviewed statements: the unaudited July deck had Q2 operating cash flow at NT$55.01bn. The reviewed statements show NT$55.86bn (a derived figure, H1 less Q1). The difference is small, but quote the reviewed number.
• Related parties:
• Nan Ya Plastics leases land and buildings to Nanya (10-year leases from 2020).
• Formosa Advanced Technologies is a 32% associate handling assembly and test.
• Sales to the European subsidiary were NT$7.94bn (6.0%), and are intra-group.
• No abuse is flagged.
• Covenants: a NT$25bn Bank of Taiwan syndicated facility carries a current ratio ≥100%, liabilities to tangible net worth ≤200% and tangible net worth ≥NT$10bn. All are complied with and far from binding.

Regulatory findings (required):

• SEC: Nanya has no CIK, so regulatory/regulatory-findings.md (regenerated 2026-10-05) shows 0 SEC Litigation Releases and 0 AAERs.
• Company disclosure (TIFRS Note 9(b), Contingent liabilities): "In 2010, the Company was charged by Brazil's Ministry of Justice as being involved in the International Monopolies, which influences Brazil's DRAM market … The lawsuit was in a court hearing". This is the long-running DRAM price-fixing matter. It is still open, but no provision is disclosed.
• Web: no FTC/DOJ/other enforcement actions against Nanya surfaced. Two items are relevant:
• A Taiwan prosecution of an ex-engineer under the Trade Secrets Act, where Nanya is the victim.
• Ministry of Labor safety-inspection deficiencies at the expansion site, February 2026 [company PR].
• Conclusion: no material adverse regulatory finding against the company. The Brazil antitrust contingency is disclosed and unresolved. Verified via SEC EFTS (n/a, no CIK), TIFRS Note 9 and a web search as of 2026-10-05.

---

Phase D — Project & stress-test

Forward Projection — re-forecast

Workbook status: model.xlsx was rebuilt on 2026-10-05 from the new our figures. It has 8 quarters, and Opening balance sheet sourced: yes. However, our model reports "No computed values": the workbook has never been opened and saved, so there are no cached outputs to cite. The seeded drivers in our figures need Connor's view before any of them is meaningful:

• share price = 0 (spot is NT$530; note the model is in USD)
• tax rate 15% (the actual Q2 rate is 19.5%)
• payout 0% (management says ~40%)
• US-style CAPM (risk-free 4.2%, ERP 5%) on a TWD business

The projection below is therefore prose, as the skill requires when the workbook is uncomputed.

Anchors:

• FY25 EPS NT$2.13 and H1-26 EPS NT$23.38.
• Q3-26 revenue NT$133.65bn, known.
• 3,450.2m shares outstanding, plus ~130m potential dilution.

Q3-26E. Revenue is NT$133.65bn. Cost of goods sold is NT$17.4–26.7bn (87% to 80% gross margin): Q2 was NT$16.93bn on flat bits, and Q1→Q2 COGS rose only 7%. Gross margin is the swing assumption. The Street's ~80% is our Bear row. Our 85% base rests on COGS rising only modestly with flat bits. Opex is ~NT$7.5bn (profit-sharing scales with profit). Non-operating income is +NT$1.5bn and tax is 19.5%.

FY2026E.

• 9M net income ≈ NT$76.25bn (H1) + NT$86.6bn (Q3E) = NT$163bn.
• Q4E: revenue flat to +10% QoQ. This sits between management's "no major change" and TrendForce's +10–15% for conventional DRAM, discounted for Nanya's LTA mix. Net income NT$87–96bn.
• FY net income ~NT$250–259bn. Weighted shares ~3,355m (3,098.6m for ~97 days, 3,450.2m after).
• EPS ≈ NT$74–77. Base NT$75. That is ~1.8× the July dossier's NT$42.55 Street anchor.

FY2027E: the real fork.

• Base, NT$95–115 (point ~NT$105). Pricing plateaus at Q4-26 levels through 1H27, with LTAs dominant. Bits grow only low-teens because Fab 5A only begins wafer input in 2H27. Gross margin is ~78–82%. Revenue is ~NT$520–600bn, net income ~NT$340–390bn on ~3,460m shares.
• Bull, ~NT$130. Contract prices keep climbing into 2027, consistent with "PC OEMs … expect DRAM supply will remain tight in 2027".
• Bear, ~NT$55–65. Consumer demand destruction (management itself flagged the "low-ticket items" ecosystem), plus CXMT G5 volume, rolls prices down 25–35% from 2H27.

FY2028E: wide, NT$20–80.

• Fab 5A ramps to 30K WSPM. Phase-1 capex of NT$346.6bn over ~5-year equipment lives implies ~NT$50–70bn a year of depreciation by 2029, against ~NT$11.8bn annualized today.
• Industry greenfield lands "beyond 2028" by Nanya's own description. In a down-cycle year a doubled depreciation base can wipe out most of the profit, as in 2023–24, when Nanya lost money with D&A of ~NT$15–16bn a year.

Balance-sheet consequence: even the bear case never draws on borrowings. End-Q3 net cash should be ~NT$240–250bn. Fab 5A phase 1 is fully self-funded at anything near current prices. The risk is earnings and returns on the new fab, not solvency.

(No our model create — unattended run, per brief.)

Bull vs Bear — re-run

Bull case. Nanya is the purest listed exposure to a shortage that has already beaten every forecast, including ours.

• It is printing ~NT$100bn of quarterly operating income on a NT$1.8tn market cap, with ~NT$200bn+ net cash and a ~40% payout coming (~5.5% yield on FY26E).
• Supply is genuinely capped. Nanya cannot add meaningful bits until 2H27, and the big three are prioritising HBM and RDIMM.
• The customer base has de-risked. US-destination revenue is up 41× YoY, and strategic holders are locked in until 2029.
• Pingtung 3D packaging and WoW give a credible path from commodity to customized AI memory. A Vera Rubin LPDDR design win, if real, would re-rate the "no HBM" discount.
• At ~5.3× Q3E annualized earnings (~5.6× if gross margin is the Street's ~80%), the market is already pricing a sharp fall. If FY27 merely plateaus (NT$~105), the stock is ~5× forward with cash rising fast.

Bear case. Three permanent-impairment routes:

1. The cycle turns into Nanya's own capex. Fab 5A's NT$346.6bn phase 1 lands in 2H27–2029, exactly when Nanya says industry greenfield arrives. Memory has always over-built into strength, and this time Nanya is one of the builders.
2. CXMT is now a funded public company with G5 in volume, a sixth fab under consideration and a ~17% share target for 2028. It aims at the same DDR4/DDR5/LPDDR tier.
3. Demand destruction at the low end. Management conceded the "low-ticket items" ecosystem "is tougher now". TrendForce flags consumer-side pressure and smartphone weakness. Nanya's ~80% legacy mix is concentrated exactly where buyers are weakest.

Pre-mortem (it is April 2028 and the thesis broke). DDR4/LPDDR4 contract prices peaked in 1H27. Consumer OEMs cut builds and substituted DDR5 from more suppliers. CXMT's G5 plus Korean RDIMM capacity normalized supply in 2H27. Nanya's FY27 came in near NT$60 instead of NT$105, and Fab 5A depreciation began weighing on FY28. LTAs without take-or-pay were renegotiated downward. The stock round-tripped to around 2.5–3× book (~NT$300).

Are multiples too high? On spot earnings, no. On mid-cycle earnings they are high. If FY24-type losses alternate with peaks, mid-cycle EPS might be NT$20–30. At NT$530 that is ~18–26×, plus net cash of ~NT$70/share on the end-Q3 estimate (June 30 actual: NT$57.7/share). The market is pricing roughly three to four years of near-peak cash generation, then normalization. That is a reasonable center, which is why the bet is about duration, not direction.

Contrarian view. The consensus debate is "how long does the shortage last". The overlooked variable is Nanya's own capital allocation. A company that sat out the last capex cycle and earned 80% margins because it was capacity-starved has just committed ~US$11bn to phase 1 of a greenfield fab. The pattern a skeptic should watch for is the one where Nanya's own supply arrives into its own downturn, and that risk will not show up in any monthly revenue print until 2028.

Devil's Advocate (short-seller) — re-run

• What breaks the money machine: price. Nearly all of the YoY gain is ASP. Bits rose only "high-twenties %" YoY in Q2 while revenue rose 684%. Gross margin is resilient: with costs flat, a 30% ASP reversal from peak takes it from ~85% to ~79%, and a 50% reversal to ~70%. Earnings are not resilient. A 50% ASP reversal more than halves EPS, because revenue halves while opex and the coming Fab 5A depreciation do not.
• The LTAs are softer than they sound. Management admitted they lack take-or-pay, floors or prepayments ("we will consider that"). Price-floating committed-volume LTAs are a shortage artifact and give no downside protection.
• Concentration: ~47% of revenue ships to Hong Kong and China, and ~80% is legacy DDR3/DDR4/LPDDR4. Those are the two most contestable exposures, one geopolitical and one competitive.
• The dangerous competitor is still CXMT, now with public-market capital and a G5 node in mass production. Management's "not directly to Nanya" is a statement about 2026 allocation, not about 2028 cost curves.
• Worst capital-allocation move: approving the phase-1 Fab 5A budget at what may be peak margins, plus a 34% in-year capex raise for advance payments. That is classic late-cycle behaviour, even if it is cash-funded.
• What must hold for NT$530: roughly 3 years of ≥NT$70 EPS to justify ~7× FY26 on mid-cycle math. If 2027 disappoints by 20–30% (EPS ~NT$75–85 rather than ~105), forward P/E drifts to ~6–7×. In memory, that has historically been where peak-cycle names fall to, not where they find support.
• Single permanently impairing scenario: big-three RDIMM/DDR5 normalization, CXMT G5 scale and Fab 5A's first wafers all arriving in 2H27–2028, in a demand air-pocket. Plausibility: moderate, and higher than it was in July, because the Nanya-side supply leg is now board-approved.

Management Questions — carried, see the previous dossier, with three added at the top by information value

0a. What share of 2026 and 2027 bit volume is under LTAs, split between fixed-price and market-indexed, and will any new 2027 LTAs carry take-or-pay, prepayment or price-floor terms? 0b. With phase 1 of Fab 5A committed at up to NT$346.6bn, what DRAM price (or gross margin) does the investment case require in 2028–29, and what is the 2027–29 depreciation schedule? 0c. What is the Pingtung 3D packaging fab's capex, timing and anchor customer, and does it make Nanya an HBM or custom-stack supplier or only a WoW subcontractor?

(Original questions 1–15 stand. Q4, on whether opting out of HBM is permanent, is partly answered by Pingtung.)

---