Nanya Technology

Phase A — Understand the business

Lens 1 · Company Overview

Nanya is a stand-alone commodity-DRAM maker — it designs and fabricates DRAM chips (DDR4, DDR3, LPDDR4, and a growing DDR5 slice) and sells them as components to module makers, PC/consumer OEMs, industrial/automotive, networking, and increasingly cloud/server customers. It is the largest of Taiwan's memory makers by DRAM revenue and one of only a handful of pure DRAM players left globally outside the big three.

• Product mix (Q1 2026): ~10% of revenue from DDR5; the balance is legacy DDR4 / LPDDR4 / DDR3, plus early "customized AI UWIO" memory beginning to contribute. This mix is the whole story: Nanya is over-indexed to the legacy nodes that are now in acute shortage because the big three have diverted wafers to HBM.
• Business model: Capital-intensive, cyclical component manufacturing. Revenue = bit-shipments × ASP; both swing violently with the DRAM cycle. There is no take-or-pay recurring structure — pricing is spot/contract and moves monthly. That is why FY24 was a loss and Q1 2026 was a record within four quarters.
• Corporate structure: Member of Formosa Plastics Group; Nan Ya Plastics Corp holds board seats and provides "full support". The parent is a petrochemical conglomerate — an unusual, deep-pocketed backer for a memory maker, and a source of financing resilience through downturns.
• Historical lineage: Nanya's DRAM technology has for a decade been licensed from Micron (20nm, then 1x/1y under a Dec-2015 agreement). Micron acquired Nanya's former JV Inotera outright in 2016. So Nanya is technologically downstream of Micron's roadmap, now transitioning to self-developed 1B/1C nodes.

Key customers/suppliers/competitors: Customers = module makers + OEMs + (newly) US cloud giants and NAND/SSD vendors securing DRAM supply. Suppliers = the WFE toolmakers (ASML for future EUV, plus Applied/Lam/TEL) and the Formosa group for utilities/land. Competitors = Samsung, SK Hynix, Micron (the big three, ~90%+ of DRAM), plus Winbond and China's CXMT in the legacy/niche tier.

Lens 2 · Supply Chain

Upstream → Nanya → end customer, named at every link:

• Equipment (upstream): ASML (lithography; future EUV for 1C/1D), Applied Materials, Lam Research, Tokyo Electron (deposition/etch). FY26 capex ~NT$52bn, ~30% (~NT$15.6bn) to wafer-fab equipment. Chokepoint: EUV access — Nanya's new Taishan fab houses a dedicated EUV building; without EUV it cannot close the node gap to the big three.
• Wafers/materials: silicon wafers (GlobalWafers — also Taiwan/Formosa-adjacent — SUMCO), photoresist, specialty gases. Standard semiconductor-materials chain; no idiosyncratic single-source flagged.
• Nanya (the node): Fabs in Taiwan (existing 300mm capacity) + the new NT$300bn (~US$10.2bn) Taishan/Nanlin fab in New Taipei — 10nm-class, EUV-capable, ~45,000 wafers/month at full build, equipment move-in Q1 2027, commercial ramp thereafter. This fab is the single most important item on the supply map — it is both the growth engine and the multi-year capex sink.
• Downstream customers/partners: In April 2026 Nanya closed a NT$78.72bn private placement (10.19% collective stake) whose buyers are its supply-chain partners locking in DRAM: SanDisk (~4%), Kioxia (~2%, its first-ever memory-maker equity stake), Solidigm/SK Hynix (~2–3%), Cisco (~2–3%). Three are NAND/SSD players who need DRAM to ship enterprise SSDs; Cisco needs it for networking gear. This is the defining supply-chain fact of 2026 — customers took equity to guarantee allocation, which is the strongest possible signal of scarcity.
• End market: AI datacenters (server DRAM / RDIMM / LPDDR5 spillover), enterprise SSD, PC, industrial, automotive, networking.

Lens 3 · Competitive Advantages (moats)

Blunt assessment: Nanya's durable moat is thin; its 2026 windfall is a cyclical advantage, not a structural one.

• What is NOT a moat: Technology. Nanya runs a generation behind the big three — 1B (2nd-gen 10nm-class) in full production while Samsung/SK Hynix/Micron are on 1c/1γ and pushing 1d, and its nodes are licensed-derived, benchmarked closer to CXMT's G3 (~1x). It has effectively no HBM — it sits out the highest-value, fastest-growing DRAM category entirely.
• What IS working right now (cyclical, borrowed moat): Nanya is over-weight the legacy nodes (DDR4/DDR3/LPDDR4) that the big three are exiting to free wafers for HBM (HBM consumes ~3× the wafers per bit of DDR5). That crowd-out created a legacy-DRAM shortage where Nanya is one of the few remaining volume suppliers — hence 68% gross margin on non-HBM DRAM in Q1 2026. This is a moat rented from the cycle, not owned.
• Real, narrower moats: (1) Formosa Plastics backing — a survival moat: it let Nanya endure eight straight loss-making quarters (2023–24) without distress. (2) Trusted non-China, non-big-three supply — for Western customers de-risking away from CXMT, Nanya is a politically safe second source, which is exactly why SanDisk/Kioxia/Cisco bought in.
• Bargaining power: Temporarily inverted in Nanya's favor — customers are pre-paying via equity. In a normal cycle, a sub-scale commodity maker has weak pricing power against both toolmakers (oligopoly suppliers) and large buyers. Assume today's power reverts.

Lens 4 · Segments

Nanya does not report rich product/geography segment splits publicly, and segments.csv is empty, so this is coarse ``:

• By product: DDR4/DDR3/LPDDR4 legacy ≈ ~90% of revenue; DDR5 ≈ ~10%; early "AI UWIO" custom memory a small emerging slice. The strategic tension: DDR5 is the future mix but Nanya is deliberately holding DDR4 capacity because the DDR4 shortage is more profitable — the GM explicitly said DRAM margins across all segments currently top HBM economics for them. "DDR4 shortage limits DDR5 shift" is the segment headline.
• By geography: Taiwan-based manufacturing; global sales (Asia OEM/module base + growing US cloud). No published geographic revenue split available — n/a.
• Trend & cause: Legacy mix is accelerating in revenue terms purely on ASP (+>70% QoQ, +>200% YoY in Q1 2026) despite declining bit shipments (down mid-single-digit % QoQ). Price, not volume, is doing all the work — the signature of a supply-constrained cyclical top-of-cycle, not demand-led secular growth.

---

Phase B — Measure performance

Lens 5 · Earnings Result (Q1 2026 — the latest print)

The most extreme quarter in Nanya's recent history.

• Revenue: NT$49,087m, +63.1% QoQ, +582.9% YoY. H1-2026 cumulative revenue NT$131.64bn (~US$4.1bn), +643% YoY.
• Gross margin: 67.9% (+18.9pp QoQ). Operating income NT$30,111m, operating margin 61.3% (+22.2pp QoQ). Net income NT$26,058m, net margin 53.1%. EPS NT$8.41 — roughly 4× the entire FY25 net income (NT$6.60bn) in a single quarter.
• Driver: ASP +>70% QoQ (+>200% YoY) with bit shipments −mid-single-digit % QoQ. Pure price. The DRAM shortage — big-three HBM crowd-out + CXMT DDR4 retrenchment + Micron/PSMC legacy exits — is the entire cause.
• Balance sheet (Q1 2026): Cash NT$86.0bn, debt NT$17.8bn → net cash NT$68.2bn before the April placement; +NT$78.72bn placement lifts gross cash to ~NT$165bn, net cash ~NT$147bn. FCF NT$28.5bn in the quarter; capex only NT$2.8bn in Q1 (back-end-loaded to the new fab). This is now a fortress balance sheet — the cyclical windfall is being banked, not blown.
• Guidance/tone: Management guided Q2 ASPs up double-digit-to-"tens of percent" QoQ, "high gross margin sustainable for the next few quarters," tight supply "potentially through end-2027". Tone: unusually confident for this management, and explicitly framing the DRAM industry as in "structural transformation."
• Market reaction: Stock ran from a NT$40.20 52-week low to a NT$505 high — a >10× move — with Morgan Stanley's bull case flagging NT$1,160 if DDR4 stays strong and 1a/1b ramps faster than expected. The tape has priced in a lot.
• Unusual vs. history: Everything. Nanya lost money for eight consecutive quarters into 2024; a 68% gross-margin quarter is without precedent in its recent record and is definitionally not a run-rate.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on the research-layer shelf; sentiment reconstructed from call summaries ``.

• Q4 2024 → Q1 2025: Defensive. Eight straight loss quarters; management blamed "Chinese oversupply" (CXMT) for the prolonged downturn and predicted a 1H-2025 rebound. Recurring phrase: survival / discipline / cost.
• Q4 2025: Turn confirmed. Record Q4 EPS NT$3.58, flagging "tight DDR5/4/3 supply" and Q1 price hikes ">20%". Tone flips to tight supply / pricing power.
• Q1 2026: Euphoric-but-measured. "Cannot 100% satisfy market demand this year," scarcity into 2027, "structural transformation". What they stopped saying: the CXMT-oversupply lament of 2024–25. What they started saying: structural/multi-year, and "DRAM margins top HBM" — a tell that they are content to not chase HBM.
• Sentiment arc: survival (2024) → inflection (late 2025) → confidence (2026). The risk in a call arc like this is that peak confidence tends to coincide with peak cycle.

Lens 7 · Comps

DRAM peers. Multiples are `` with source/date or n/a. Do not read across too hard — Nanya's trailing P/E is distorted by a loss-to-boom base, and the big three carry HBM mix that Nanya lacks.

EV/Sales, EV/EBIT, dividend yield, 5-yr avg ROE: n/a for Nanya and peers (a loss-year history makes 5-yr ROE meaningless anyway). Read-through: memory is cheap on trailing earnings everywhere because the market prices in cyclicality; Nanya at ~10.6× 2026E is not obviously cheap versus Micron (~8×) or SK Hynix (~6× fwd) once you adjust that Nanya's earnings are more cyclical and less HBM-durable.

Lens 8 · Stock-Price Catalysts

What actually moves 2408.TW (>5% moves), ``:

• The DRAM price cycle is the master catalyst. The 40→505 run in ~12 months maps 1:1 onto DDR4/DDR5 contract-price inflection and monthly-revenue prints (Jan-2026 revenue +6× YoY was a step-change).
• Monthly revenue disclosures (Taiwan requirement) — Nanya reports monthly; each record print is a discrete catalyst.
• The April private placement (SanDisk/Kioxia/SK Hynix/Cisco buying in) — validated scarcity and re-rated the name.
• Sell-side targets — Morgan Stanley's NT$1,160 bull scenario and "memory panic-buying" call drove momentum.
• Return of the dividend — ex-dividend NT$1.347 after a two-year hiatus drove a ~7% surge.
• CXMT / big-three capacity headlines move it the other way: CXMT DDR4 expansion, Micron/SK Hynix returning to legacy, and CXMT node-gap-narrowing stories are the recurring bear catalysts.
• Pattern revealed: the market treats Nanya as a high-beta call option on legacy-DRAM ASP. It reacts to price/supply data far more than to anything company-specific — because the company is the cycle.

---

Phase C — Judge people & books

Lens 9 · Management

• Key figure: Dr. Pei-Ing Lee — President since Oct-2015, 30+ years in DRAM; now the public face/GM articulating the "structural transformation" thesis. Board includes Nan Ya Plastics representatives (Chia-Chau Wu, Pei-Ing Lee) — parent-controlled governance.
• Track record: Steered Nanya through the brutal 2023–24 downturn (eight consecutive loss quarters, cumulative losses ~NT$12bn+ across FY23–24) without balance-sheet distress, then captured the 2026 upswing — respectable survival-and-capture execution. But note: they did not build an HBM position or close the node gap during the last cycle, which is the strategic knock.
• Skin in the game / ownership: Effectively a Formosa Plastics Group affiliate — the parent's stake dominates the register. Insider-transaction data: n/a. This is professional management operating a strategic asset for a conglomerate parent, not founder-owner-operators.
• Capital-allocation history: Conservative and cyclical-appropriate. Cut/suspended the dividend through the downturn, reinstated it in 2026 (NT$1.347), and is banking the windfall as net cash (~NT$147bn) rather than levering up while funding the NT$300bn Taishan fab partly via government grants and the equity placement (avoiding heavy debt). That is disciplined. The open question is whether pouring NT$300bn into another commodity-DRAM fab at cycle-peak is value-accretive or classic top-of-cycle capex.
• Red flags: None acute. The structural concern is strategic dependence — a decade of Micron-licensed technology and a perpetual one-generation lag. Governance is parent-dominated (minority-holder alignment is a standing question for any Formosa affiliate), but no related-party abuse is flagged in public sources.
• Archetype: Professional managers / conglomerate stewards, not founders. Implication: expect cycle discipline and survival, not moonshot node leadership.

Lens 10 · Forensic Red Flags

Web-only; every figure /.

• Earnings quality: Q1 2026 earnings look high quality on the surface — FCF NT$28.5bn roughly tracks net income NT$26.1bn, so profits are cash, not accrual. No obvious cash-vs-earnings divergence.
• The real "flag" is durability, not fraud: 68% gross margin is a cycle artifact. Modeling it as a run-rate is the single most dangerous analytical error on this name — the same business printed negative operating margin in 2024.
• Inventory/receivables: In a shortage with bit-shipments falling and prices rising, inventory risk is currently low. But this is a company with a large inventory-write-down exposure on the downcycle (memory is the classic obsolescence/write-down business). No current red flag; a structural one for the eventual turn.
• Dilution: The NT$78.72bn private placement increases share count — existing holders were diluted (the ~3.10bn implied pre-placement share base rises). Consensus 2026E EPS of NT$42.55 must be read post-dilution. Flag for the projection.
• SBC / non-GAAP games: Nanya reports on Taiwan-IFRS; no evidence of aggressive non-GAAP adjustment surfaced. n/a on SBC specifics.

Regulatory findings (required): Per regulatory/regulatory-findings.md (generated 2026-07-06): Nanya has no CIK and files no SEC documents → no SEC Litigation Releases or AAERs are searchable (total_sec_findings: 0). Non-SEC web search ("Nanya Technology" (FTC OR DOJ OR FDA OR CFPB OR consent decree OR settlement OR fine OR penalty) enforcement) surfaced no material enforcement actions. There is a historical industry note: DRAM makers globally (including Taiwan producers) were subject to price-fixing scrutiny in the mid-2000s, but nothing material or recent attaches to Nanya in current public sources. Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (n/a, no CIK), web search, and public sources as of 2026-07-06. Item 3 Legal Proceedings is not applicable (no 10-K filer).

---

Phase D — Project & stress-test

Lens 11 · Forward Projection

Nanya's fiscal year = calendar year. Forecasting DRAM EPS three years out is inherently low-confidence because it is a price-cycle guess, not an operating-leverage model. Every line (anchor) or (mine).

Anchors: FY25 actual EPS NT$2.13 (net income NT$6.60bn). Q1 2026 EPS NT$8.41. Street 2026E: revenue ~NT$226.5bn, EPS ~NT$42.55. Price ~NT$449 → ~10.6× 2026E.

• FY2026E (base): ~NT$40–43 EPS. Ties to Q1 NT$8.41 annualized-and-rising (Q2 guided up) → the Street's NT$42.55 is credible if ASPs hold through H2 and the placement dilution is absorbed. `` — I would sit slightly below NT$42.55 to respect dilution and any H2 price plateau, call it NT$38–42.
• FY2027E (base): NT$25–40, extreme uncertainty. Bull: management's "tight through end-2027" holds, new-fab bits and 1C add volume → EPS above 2026. Bear: DRAM cycles peak and roll; if ASPs mean-revert even partway, a 68%-GM business can halve earnings fast. Base assumes a high plateau, then softening → ~NT$25–35 ``.
• FY2028E (base): NT$5–25, wide. By 2028 the new Taishan fab is ramping (depreciation drag) and CXMT capacity is materially larger; the odds of the cycle having turned down are high. A commodity-DRAM maker's FY28 EPS three years out is a coin-flip on where the cycle sits — I decline to pretend precision: NT$5–25 ``, with genuine downside to loss if the cycle craters as in 2024.

Base / Bull / Bear (FY2027, the pivot year):

• Bull: ASPs hold, 1C ramps, new-fab bits accretive → EPS ~NT$45+, stock re-rates toward the MS NT$1,160 scenario.
• Base: high plateau then fade → EPS ~NT$30, stock range-bound to lower as the market discounts the coming turn.
• Bear: cycle rolls in 2027, big-three + CXMT reflood legacy → ASPs −40–50%, EPS collapses toward NT$5 or a loss on new-fab depreciation.

(No forecast.ts create in --watchlist mode, per skill.)

Lens 12 · Bull vs Bear

Bull case. Nanya is the cleanest, cheapest listed proxy for the legacy-DRAM shortage — a genuine multi-year supply/demand dislocation where the big three have structurally abandoned commodity DRAM for HBM, CXMT can't yet fill the gap on advanced nodes, and Western customers actively want a non-China second source (proven by SanDisk/Kioxia/SK Hynix/Cisco taking equity). The balance sheet is now net-cash ~NT$147bn, the dividend is back, and the new NT$300bn EUV fab gives a real volume-growth runway into 2027–28. If the "structural transformation" framing is right and tightness runs through end-2027, current ~10.6× 2026E earnings understate the earnings power — the MS NT$1,160 bull scenario is the embodiment. Bear case. Three ways this permanently disappoints: (1) It's a cycle, and cycles turn — 68% gross margin is a top-of-cycle artifact; the same fabs printed operating losses in 2024, and memory always over-builds into strength. (2) The node gap is structural — a generation behind the big three, no HBM, licensed-derived technology; when the shortage ends, Nanya reverts to being a sub-scale price-taker. (3) CXMT — China's champion did US$8bn revenue in 2025 (+130%), is closing the node gap from ~5yr to ~3yr "faster than expected," has a >15% price advantage and state subsidies; its capacity ramp is precisely aimed at the legacy DDR4/DDR5 tier that is Nanya's entire franchise. Pre-mortem (18 months out, thesis broke): DRAM ASPs peaked in mid/late-2026; by end-2027 big-three HBM yields improved and freed capacity back to DDR5, CXMT flooded DDR4, contract prices fell 40%+, and Nanya's new-fab depreciation turned a boom into a margin squeeze — the stock round-tripped much of the 10× move. Are multiples too high? On 2026E earnings, no (~10.6×). On normalized mid-cycle earnings, almost certainly yes — the entire question is what you capitalize. The market is capitalizing near-peak earnings at a cyclical multiple; that only works if the peak persists. Contrarian view (what the market refuses to see): The bull consensus treats "AI structurally lifts all memory forever" as a floor. The overlooked truth is that Nanya's windfall is specifically a big-three-HBM-diversion artifact — it is short their capacity discipline, not long a durable moat of its own. The moment HBM economics normalize or CXMT scales, Nanya's borrowed pricing power is handed back. This is a trade on a dislocation, dressed up as an investment in a structural winner.

Lens 13 · Devil's Advocate (short-seller)

• What structurally breaks the money machine: ASP reversion. 100% of the earnings surge is price. When (not if) DRAM prices normalize, a one-generation-behind commodity maker with no HBM is back to sub-10% margins or losses — as in 2024.
• Revenue concentration: Concentrated in legacy nodes (DDR4/DDR3/LPDDR4), the exact tier that (a) the big three are exiting temporarily and can re-enter, and (b) CXMT is expanding into permanently. The concentration is in the most contestable part of the market.
• Why the moat is weaker than bulls think: It has no proprietary node lead, no HBM, and licensed-derived technology; its "moat" is the big three's voluntary absence, which is reversible.
• Most dangerous competitor bulls underestimate: CXMT. US$8bn 2025 revenue, tripling wafer capacity, node gap closing "far faster than expected," >15% price edge + subsidies. It is aimed straight at Nanya's franchise, and China's memory push is a state priority.
• Worst capital-allocation risk: Committing NT$300bn to a new commodity-DRAM fab at cycle-peak. Classic memory over-investment; if it ramps into a glut, the depreciation converts a cyclical downturn into a structural loss.
• What must hold for today's price: That DRAM tightness persists through ~2027 and Nanya's node/dilution don't erode the per-share windfall. If 2027 growth disappoints 20–30% (i.e., ASPs merely plateau rather than keep climbing), the earnings and the multiple compress together — the classic cyclical double-whammy.
• Single scenario that permanently impairs: CXMT reaching cost-competitive DDR5 at scale and the big three returning to legacy simultaneously → structural oversupply that leaves Nanya as the marginal, sub-scale, node-lagging producer nobody needs. Plausibility: moderate and rising over a 2–3-year horizon.

Lens 14 · Management Questions (ordered by information value)

1. What is your explicit ASP/bit-cost assumption behind the "high gross margin sustainable for several quarters" guidance, and at what DDR4/DDR5 price level does gross margin fall below 30%?
2. When the big three redeploy freed capacity back to DDR5 as HBM yields improve, what is your defensible cost position on DDR5 versus them — and versus CXMT?
3. Is the NT$300bn Taishan fab decision robust to a DRAM down-cycle in 2027–28? What DRAM price sustains its returns, and what is the depreciation trajectory once it ramps?
4. You said DRAM margins currently top HBM for you — is opting out of HBM a permanent strategic choice, or a timing one? What is your path (if any) into HBM/advanced packaging?
5. How do you assess CXMT's DDR4/DDR5 cost curve versus yours today and in 24 months, given their >15% price advantage and state subsidies?
6. What share of the April private-placement partners (SanDisk/Kioxia/SK Hynix/Cisco) is tied to binding multi-year volume/price agreements versus pure equity — i.e., how much revenue is actually locked?
7. On the 1C node — what is the realistic qualification-to-volume timeline, and what happens to your competitive position if 1C slips past end-2026?
8. What is your normalized mid-cycle gross-margin and ROIC assumption — the number you underwrite the fab and dividend against?
9. Your technology has been Micron-licensed for a decade; what is the current IP/licensing status and how self-sufficient is the 1C/1D roadmap?
10. How do you intend to deploy the ~NT$147bn net cash — buybacks, dividends, further capacity, or M&A — and what return threshold governs it?
11. What is your EUV adoption plan and ASML tool-allocation timeline for 1C/1D, and is EUV access a gating risk?
12. What visibility do you have into inventory across your customer channel, and how do you avoid being the last to see a demand air-pocket?
13. How exposed is your order book to US-China export controls or Taiwan-specific trade actions on memory?
14. What is the governance framework ensuring capital-allocation decisions serve all shareholders, not just Formosa Plastics Group's strategic interests?
15. What single indicator do you watch to know the cycle is turning, and what pre-committed action follows?

---