Macronix

Phase A — Understand the business

Lens 1 · Company Overview

Macronix International (founded 9 Dec 1989 by Dr. Miin Wu and ~40 engineers repatriated from Silicon Valley) is a fabless-turned-IDM specialist in non-volatile memory — it designs and manufactures its own chips in owned Taiwan fabs. It is not a commodity DRAM/NAND giant; it is the world's largest pure-play NOR flash maker and a dominant supplier of mask ROM and SLC NAND. Three product families:

• NOR flash (serial/parallel, Octa, wide-VCC) — code-storage memory for embedded systems. ~51% of 2023 revenue; ~58.5% of Q1'26 revenue.
• Mask ROM ("XtraROM") — customized read-only memory. This is the Nintendo business: Switch and 3DS Game Cards use Macronix XtraROM, not flash. ~34% of 2023 revenue.
• NAND flash + eMMC (SLC NAND, serial NAND, MLC eMMC) — ~15% of 2023 revenue but surged to ~30% of Q1'26 revenue as an SLC-NAND shortage detonated.

Business model: sell embedded code-and-data storage into long-lifecycle industrial, automotive, networking, IoT, consumer, computing and communications sockets, largely on a design-win-then-multi-year-supply basis (embedded memory is designed into a board and stays for the product's life — high switching cost, low churn). Customers span consumer electronics (Nintendo the marquee name), automotive tier-1s/OEMs, networking/industrial. Macronix reports as effectively one segment — it does not publish audited revenue splits by end-market or geography in its English disclosures. Sales regions: Taiwan (largest), China, Japan, US, Europe, Korea, Singapore.

Key contract dynamic that matters right now: Macronix historically priced quarterly; in Q1'26 it switched to monthly pricing to capture a shortage-driven spike — a structural tell that pricing power flipped hard in its favor.

Lens 2 · Supply Chain

Map, upstream → Macronix → end customer, named nodes:

Upstream (inputs):

• Silicon wafers (300mm/12-inch and legacy 200mm) — from Taiwan/Japan wafer suppliers (e.g. GlobalWafers, SUMCO, Shin-Etsu class suppliers) ``.
• Semiconductor equipment — deposition/etch/litho tools from ASML, Applied Materials, Lam Research, Tokyo Electron for the 12-inch fab expansion. Chokepoint flagged by management: new NAND equipment ordered for the 2026 capex has long lead times and won't arrive until H1 2027, capping near-term capacity.
• Manufacturing: Macronix is an IDM — it owns its fabs (Fab 2 legacy 200mm; Fab 5 300mm in Hsinchu, the growth engine, ~20k wafers/mo currently loaded of a ~25k near-term ceiling, roadmapped to 40k). This vertical integration is central to the moat (below).

Midstream (Macronix): wafer fab → in-house assembly/test (partly) + OSAT partners (ASE/Amkor-class) ``.

Downstream (end customers / demand pull):

• Nintendo (via Game Card XtraROM) — the single most identifiable customer, tied to console cycle/unit sales.
• Automotive tier-1s & OEMs — automotive is now ~24-25% of NOR revenue, up from single digits a decade ago; qualification (AEC-Q100, ISO 26262) is the barrier to entry.
• Networking / industrial / IoT OEMs — code-storage sockets.
• SLC-NAND buyers stranded by Samsung's exit — see Lens 3; Macronix says it is now "the only company globally capable of supplying this specification range at scale".

Single-source / chokepoint summary: the binding constraint is Macronix's own 12-inch capacity + delayed tool delivery to 2027, not an upstream input shortage. On the demand side, the risk is customer concentration in console cycles (Nintendo) and automotive design-win timing.

Lens 3 · Competitive Advantages (moats)

Four durable moats, in order of strength:

1. IP estate + patent-plaintiff posture (strongest). Macronix holds one of the largest US patent portfolios of any Taiwanese company and has repeatedly sued and won/settled against larger rivals: Spansion/Cypress (ITC 337-TA-909 + District Court + German suits, settled 2015 — the settlement even cleared the Spansion/Cypress merger), Toshiba (settled 2018), and others. Being the aggressor in NVM patent litigation is a real, cash-generative moat and a barrier to challengers.
2. Embedded design-win switching costs. NOR/ROM go into a board and stay for the product's 5-15 year life (industrial, automotive, networking). Requalifying a second source is expensive and slow — this is why the business is stickier and higher-margin than commodity DRAM/NAND.
3. SLC-NAND scarcity / last-man-standing supply. Samsung and other majors exited legacy SLC NAND to chase HBM/high-density 3D NAND, leaving a supply vacuum Macronix can uniquely fill at scale — a temporary but potent pricing moat while it lasts.
4. Vertical IDM integration + proprietary process (3D NOR). Macronix is developing in-house 3D NOR (density scaling up to ~8x planar), sampling slated for late 2026 — a potential process-technology edge in embedded code storage.

Bargaining power: Currently high over customers (shortage + monthly pricing), but this is cyclical, not structural — it evaporates when supply normalizes. Over suppliers (equipment makers) it has little power — it waits in the ASML/AMAT/Lam queue like everyone. Verdict on moat: genuine in NOR/ROM (IP + switching costs), rented in NAND (scarcity).

Lens 4 · Segments

Macronix does not publish audited segment P&L; the product mix below is management-disclosed revenue share, not ``:

Within NOR, automotive = ~24-25% of NOR revenue, up from single digits ~10 years ago — a deliberate, value-accretive mix shift.

Geography: reported single-segment; principal sales Taiwan > China > Japan/US/Europe/Korea/Singapore. No audited geographic split available — n/a at the percentage level.

Read: the segment story is a mix-shift toward NAND (cyclical spike) and automotive NOR (structural). Management's stated goal is to move the NOR:NAND revenue ratio from ~1:0.5 toward 1:1.

---

Phase B — Measure performance

Lens 5 · Earnings Result (latest print: Q1 FY2026, reported ~2026-04-27)

The headline: Macronix snapped a TEN-QUARTER losing streak. After nine straight quarterly losses through Q4'25, Q1'26 swung decisively to profit.

⚠️

What drove it: ~all of the beat is NAND flash + eMMC (+90% QoQ / +382% YoY), whose revenue share jumped to ~30% from ~11% a year earlier, on the SLC-NAND supply vacuum + monthly price hikes. NOR pricing also firmed. Secondary help: reversal of prior inventory-valuation write-downs and rising fab utilization (~20k of 25k wafers/mo loaded).

Balance-sheet flags:

• Cash & equivalents ~NT$16.0B; total debt ~NT$22.5B → modest net debt ~NT$6.5B ``. Not a fortress balance sheet, but manageable given the earnings inflection.
• Inventory ~NT$13.4B (2023 year-end reference) — was the source of prior write-downs; now being reversed as prices rise (a tailwind that flatters current GM and will not repeat).
• Total assets ~NT$82.2B.

Guidance/tone: Chairman Miin Wu — "Each quarter will be better than the last"; capacity to "hit full load in H2 2026"; prices "continue rising through H2" but "increases cannot be unlimited". Tone flipped from years of caution to confident.

Market reaction: the stock is up ~576% over the trailing year — the turnaround was heavily anticipated and extended by the sector-wide memory mania. That is the crux: the operational good news is real but largely priced.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts/ on disk; sentiment reconstructed from reported call commentary:

• Q4'24 (Feb 2025): Chairman doubts 2025 demand recovery but "better than 2024"; focus on inventory reduction, 3D NOR contribution targeted for 2026. Tone: defensive/cautious.
• Q2'25 (Jul 2025): hit by forex losses; "stays on course for 2025 profitability" (which did not materialize — full-year 2025 was a NT$3.3B loss). Tone: hopeful-but-missing.
• Q3'25 (Oct 2025): ninth consecutive losing quarter; stock fell ~10% on the print. Tone: capitulation/trough.
• Q1'26 (Apr 2026): profit swing, "each quarter better than the last," monthly pricing, NAND shortage "supply-demand gap very large." Tone: bullish/inflection.

Sentiment arc: cautious → hopeful → capitulation → decisive optimism. The phrase they stopped saying is "inventory correction / demand uncertainty"; the phrase they started saying is "shortage / monthly price hikes / full load." Classic memory-cycle turn language. The risk in that language is that it always sounds identical at the top as at the bottom of the recovery.

Lens 7 · Comps

Taiwan/global memory peers. Multiples are `` with source/date or n/a. No multiple is fabricated.

Comps read: the entire Taiwan/China memory complex is trading on "how long can the boom last," not on trailing earnings — Nanya at NT$1.39T and GigaDevice at 118x P/E are melt-up valuations, and GigaDevice itself warned. Macronix's forward P/E ~6.6 looks cheap only on the assumption that peak-cycle earnings are the new run-rate; on 2026E NT$16.3B net profit its implied fwd P/E is ~17x ``, which is not cheap for a cyclical at what may be a cycle peak. NOR-specific peers (Winbond, GigaDevice, Infineon) are the honest comp set; the DRAM names (Nanya, Micron) are a different market.

Lens 8 · Stock-Price Catalysts (last ~5 years, >5% moves)

The name is a textbook memory-cycle beta — it moves on the commodity cycle and, lately, on AI-memory sentiment:

• 2021-22 boom → bust: revenue peaked NT$50.6B (2021), collapsed to NT$27.6B (2023) as NAND/NOR oversupply crushed spot prices (512Gb TLC hit record lows).
• 2022-2025 grind: ten straight losing quarters; stock fell to a 52-week low of NT$18. Each losing print (e.g. Q3'25, −10%) was a down-catalyst.
• Late-2025 → 2026 melt-up: the AI-driven memory shortage + Samsung's SLC-NAND exit + monthly pricing drove a ~576% 1-yr rally to a 52-week high NT$192; Q1'26 profit swing confirmed the turn.
• 2026-07-01/02 rout (fresh): Meta announced it would sell idle AI compute, triggering a global chip-stock shakeout — SOX −10%, SK Hynix −14.6%, Kioxia −13.5%; Macronix −7.7% on July 1. Current price ~NT$144.5 (2026-07-06), well off the NT$192 high.

Pattern the market actually reacts to: (1) memory spot/contract pricing direction, (2) quarterly profit/loss vs the streak, (3) AI-capex sentiment as a macro overlay. It does NOT reward the boring embedded-NOR annuity — it's traded as a leveraged memory-cycle call. Beta is enormous (18→192→144 in ~12 months).

---

Phase C — Judge people & books

Lens 9 · Management

• Dr. Miin Wu — Founder, Chairman (since 2005), CEO. Stanford MS (Materials Science); prior senior roles at Intel, VLSI Technology, Rockwell, Siliconix in Silicon Valley; left a secure US career to found Macronix in 1989, repatriating ~40 engineers to build Taiwan's first indigenous NVM supplier. A founder-operator archetype, ~78 years old, ~35 years at the helm.
• Track record: built the world's #1 pure-play NOR maker and the ROM franchise that won Nintendo; navigated multiple brutal memory cycles without going under. But also presided over the 2021-peak-to-2025-trough destruction — ten losing quarters, cumulative ~NT$8B+ of net losses 2023-25. Cyclical, but the trough was deep.
• Skin in the game: insiders (Wu + affiliates) hold low-to-mid teens % of shares; foreign + domestic institutions a large double-digit block; retail the rest. Meaningful founder alignment, not control-level.
• Capital allocation: paid a NT$1.8/sh dividend in 2024 (on 2023 earnings); likely suspended/cut through the loss years (n/a for 2025 dividend). Now re-levering into capacity: NT$22B 2026 capex to expand 12-inch NAND/NOR — a pro-cyclical bet placed at the up-cycle. History: heavy fab investment (Fab 5 to 40k wafers), IP monetization via litigation. ROE swings wildly with the cycle (strongly positive 2021-22, negative 2023-25) — 5-yr average ROE is not cleanly sourceable → n/a.
• Red flags: key-man/succession risk (a ~78-yr-old founder-CEO with no publicly-named successor is the single biggest governance risk); pro-cyclical capex timing; single-segment reporting reduces transparency. No evidence of related-party self-dealing or promotional accounting surfaced.
• Archetype implication: a technically-brilliant, patriotic founder-technologist — excellent at surviving and out-engineering, historically less disciplined on cycle timing (invests into strength, gets caught by the bust).

Lens 10 · Forensic Red Flags

Web-only; no filings to forensically parse — treat this lens as lower-confidence than for an SEC filer.

• Inventory-valuation reversals flatter current margins. The Q1'26 GM of 40.8% is boosted by reversing prior inventory write-downs taken in the down-cycle. This is legitimate accounting but non-recurring — normalized GM is lower. Watch for GM to give back gains once reversals exhaust.
• Forex exposure. Q2'25 earnings were hit by FX losses; TWD moved ~10% vs USD over the trailing year. Reported NT$ earnings carry meaningful translation/transaction FX noise quarter to quarter.
• Cash-flow vs earnings: cannot reconcile precisely without statements (n/a for FCF/quarter). Net debt ~NT$6.5B `` with a NT$22B capex program means 2026 free cash flow will likely be pressured even as net income recovers — a divergence to watch.
• Single-segment reporting limits the ability to see which product line carries margin — a transparency gap, not a fraud flag.

Regulatory findings (from regulatory/regulatory-findings.md, generated 2026-07-06):

• SEC (EDGAR EFTS — LR + AAER): zero findings. Macronix has no CIK and is not an SEC filer, so no EDGAR enforcement search is possible.
• Non-SEC (web search "Macronix" (FTC OR DOJ OR FDA OR CFPB OR settlement OR fine OR penalty) enforcement): the material legal history that surfaces is Macronix as patent PLAINTIFF (Spansion/Cypress, Toshiba, others — all settled, several in Macronix's favor), not as a regulatory defendant. No FTC/DOJ/FDA/CFPB consent decree, fine, or penalty against Macronix surfaced.
• Item 3 Legal Proceedings: n/a — no 10-K exists (foreign filer).
• Conclusion: No material regulatory or enforcement findings against Macronix — verified via SEC EDGAR EFTS (LR, AAER: 0 hits), web enforcement search (0 material hits), as of 2026-07-06. Its litigation footprint is offensive IP enforcement, which is a moat, not a liability.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (FY2026 / FY2027 / FY2028 EPS)

Built bottom-up from the Q1'26 actuals + management guidance. All outputs `` with arithmetic shown; no forecast.ts logged (watchlist/unattended run).

Anchors: Q1'26 net profit NT$1.78B, EPS NT$0.90; ~1,865-1,930M shares; management "each quarter better than the last," full utilization H2'26; NT$22B capex; consensus 2026E net profit NT$16.3B, 2026E EPS ~NT$13.29.

Base call (for the record, not logged): FY2026 EPS ≈ NT$13 (consensus-aligned), with high variance and 2027-28 skewed to the downside as the cycle matures and Macronix's own new capacity + industry additions arrive into a cooling market. The three-year path is not a smooth compounder — it is a peak (2026) → plateau (2027) → rollover (2028) shape unless the AI-memory super-cycle thesis proves structural rather than cyclical.

The forecast that actually matters (would-be Brier line): "Macronix FY2026 net-GAAP EPS ≥ NT$13" — p ≈ 0.55 `` — resolves 2027-01 on the FY2026 print. Cheap to hit if H2'26 stays tight; misses if pricing peaks now.

Lens 12 · Bull vs Bear

Bull case. Macronix is a fixed company: a decade-deep NOR/ROM franchise with genuine switching-cost and IP moats, now handed a windfall NAND up-cycle because the majors (Samsung) abandoned SLC NAND — leaving Macronix as the scaled last-man-standing supplier with monthly pricing power and a fully-loading 12-inch fab. Automotive NOR (~25% of NOR, ISO 26262-qualified) is a structural, sticky, high-value growth leg independent of the commodity swing. 3D NOR (8x density, sampling late-2026) could extend the embedded moat. On 2026E NT$16.3B net profit the stock is optically ~17x forward `` with a founder who owns low-to-mid-teens %. If the AI-memory shortage is a multi-year regime (Micron/SK Hynix capex all pointing that way), 2026 earnings are a floor, not a peak.

Bear case (2-3 permanent-impairment risks).

1. This is a rented up-cycle, not a re-rating. ~All the Q1'26 beat is NAND scarcity + inventory-reversal accounting — both non-recurring. When Macronix's own NT$22B capacity + industry supply arrives in H1 2027, the same pricing power that made the quarter reverses. Memory has done this every cycle; management's own new fab is the mechanism of the next bust.
2. The stock already 6x'd. A ~576% year means the good news is priced; the 2026-07-01/02 rout (−7.7% on Meta selling compute, SOX −10%) shows how fast sentiment unwinds. You are buying a leveraged memory-cycle call after the move, at a cyclical peak, with GigaDevice warning on its own valuation.
3. Key-man risk. A ~78-year-old founder-CEO with no named successor — a disorderly transition would hit the IP-litigation posture and strategic direction that constitute the moat.

Pre-mortem (18 months out, thesis broke): it's early 2028. NAND/NOR pricing peaked in mid-2026; the H1'27 capacity wave — Macronix's and everyone else's — hit a market that AI-memory demand no longer needed at the same intensity after the Meta-style compute-liquidation spread. Inventory-reversal tailwinds exhausted, GM fell back to the low-20s, EPS rolled from NT$13 toward breakeven, and the stock round-tripped from NT$144 toward the double digits. The turnaround was real but cyclical, and the crowd bought the annuity story at peak-cycle multiples.

Are multiples too high? On trailing earnings, yes (losses/negative). On 2026E, ~17x forward `` is a fair-to-full price for a cyclical at a probable peak — cheap only if you believe the super-cycle is structural. The market is paying for durability the business has never demonstrated across a cycle.

Contrarian view (what the market refuses to see): the crowd is trading Macronix as an AI-memory beta (Nanya/Micron correlation), but the durable value is the unglamorous, un-sexy embedded NOR/ROM + automotive annuity that the market ignores at the bottom and over-extrapolates at the top. The real long-term question isn't "how high does NAND pricing go" — it's "does 3D NOR + automotive design-wins build a through-cycle earnings floor high enough to justify a non-cyclical multiple." Nobody is underwriting that; they're underwriting the shortage.

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

Dismantling the bull case:

• Structural break in how it makes money: the profit is a commodity-price spike Macronix does not control — it prices monthly because the market lets it today. The moment supply catches demand (its own H1'27 tools + rivals), pricing power vanishes and the fixed-cost 12-inch fab becomes a margin liability at low utilization. This has happened to Macronix every single cycle (2021 peak → 2023-25 ten-quarter loss streak).
• Revenue concentration: NAND (30% of Q1'26) rests on Samsung's temporary absence from SLC NAND — a decision Samsung (or SK Hynix, or a Chinese entrant like GigaDevice/YMTC) can reverse. ROM (~34%) rests heavily on the Nintendo console cycle — a hardware refresh miss or a shift away from cartridge media impairs a third of the business. Automotive NOR is the only leg with structural stickiness.
• Moat weaker than bulls think: GigaDevice is gaining NOR share in price-competitive segments and warned it's overvalued — meaning the challenger is both cheaper and self-aware; Infineon owns the top of automotive NOR (SEMPER, ASIL-D). Macronix's IP moat is real but doesn't stop commodity price competition at the low end.
• Most dangerous underestimated competitor: Chinese NOR/NAND (GigaDevice, YMTC-adjacent) — subsidized, share-hungry, and structurally deflationary for exactly Macronix's price-sensitive segments.
• Worst capital allocation: NT$22B pro-cyclical capex placed at the up-cycle — the textbook memory-maker mistake that builds the next glut. Tools arriving H1'27 into a possibly-cooling market is the single most dangerous decision on the table.
• Assumptions the price requires: that 2026 peak-cycle EPS (~NT$13) is durable, that the shortage extends through 2027, and that new supply doesn't compress margins — a stack of cyclical bets dressed as a growth story.
• −20-30% growth-disappointment scenario: if 2026 revenue lands 25% below the bull path (pricing peaks early), net income could more than halve given operating leverage on a fixed fab base — EPS from ~NT$13 toward ~NT$5-6 ``, and a ~17x-forward stock re-rates toward single-digit trailing losses again. The valuation is exquisitely sensitive to the cycle turning.
• Single permanent-impairment scenario & plausibility: a structural Chinese SLC-NAND/NOR flood + Samsung re-entry simultaneously — plausible over 24-36 months (medium probability) — would permanently reset Macronix's pricing floor and strand the new capacity.

Lens 14 · Management Questions (ordered by information value)

1. What share of Q1'26 gross margin came from inventory-valuation reversals, and what is normalized GM at current pricing excluding reversals?
2. When your NT$22B of new 12-inch capacity comes online in H1 2027, what is your base-case assumption for NAND/NOR contract pricing at that point — and what happens to margins if pricing has normalized?
3. Samsung and the majors exited SLC NAND. What is your assumption for how long that vacuum persists, and what is your plan when they (or a Chinese entrant) re-enter?
4. Succession: Dr. Wu is the founder-CEO after 35 years — who is the named successor, and how is the IP-litigation and strategic capability institutionalized beyond one person?
5. What percentage of revenue is Nintendo/console ROM, and how exposed is that to a console-generation transition or a shift away from cartridge media?
6. Walk through the 3D NOR economics — sampling late-2026, but when does it move the P&L, at what margin, and what is the addressable embedded market it unlocks?
7. Your automotive NOR is ~25% of NOR and structurally sticky. What is the design-win pipeline and expected automotive revenue CAGR through 2028, independent of the commodity cycle?
8. What is your through-cycle mid-point gross margin target — the GM you underwrite across a full cycle, not at the peak or trough?
9. How do you defend the low-end price-competitive NOR segment against subsidized Chinese competition (GigaDevice) without destroying mix?
10. What is your capital-allocation framework — how do you decide between capex, dividend restoration, and buybacks, and why is now (up-cycle) the right time for NT$22B capex rather than balance-sheet repair?
11. What is expected 2026 free cash flow after the NT$22B capex, and does the ~NT$6.5B net-debt position constrain the expansion?
12. How much of the current order book is firm/take-or-pay vs. spot-priced, and how quickly can customers walk if pricing softens?
13. What is your FX-hedging policy after the Q2'25 forex losses — how much reported earnings volatility is translation vs. transaction?
14. Where does 3D NAND (192-layer done, 300+ needing "government support") realistically go, and are you structurally sub-scale vs. Micron/SK Hynix/Kioxia in high-density NAND?
15. If the memory cycle rolls over in 2027, what is the cost structure you can flex — what's the utilization break-even on the 12-inch fab?

---