2337.TW — Macronix International (旺宏)
Initial deep-dive synthesis, 2026-04-25. Built on the KGI initiation (2026-03-12) translated at macronix-kgi-2026-translation, the Collyer Bridge APAC wrap (17 April 2026, paywall-gated past the lede), and live yfinance data. Updated 2026-05-03 with multi-source synthesis (STF Q1-call digest + STF "Memory Trade After the Memory Trade" + Morgan Stanley "Old Memory: Ongoing Asymmetric Risks" Apr 19 2026).
SYNTHESIS UPDATE 2026-05-03 — Triangulating KGI vs STF vs MS vs Consensus
Demand-driver narrative correction
Dr. Wu (Macronix Chairman) on the Q1 2026 earnings call (per STF Research, Uncovering Alpha Beyond the Slides, Apr 27 2026):
"Please forget about AI. eMMC has essentially nothing to do with AI directly. [...] The true reason demand appears to have 'surged' — it has not actually surged. What happened is that the previous supply sources suddenly disappeared."
This is the central narrative correction. Earlier framing (and some bull cases) tied eMMC demand to AI infrastructure buildout. Macronix's Chairman explicitly disagrees. The eMMC surge is a supply-shortage story — Big Four exited a $1-2B niche they considered too small — not a demand-AI story. Customers are industrial/IoT/auto/comms/medical control systems.
Implication for durability: matters how long the supply-shortage holds. Winbond re-entry (per STF "Memory Trade") = 24-36 months from decision to volume = competition arrives 2028-2029.
FY26/FY27/FY28 forecast triangulation
| Source | Date | FY26 EPS | FY27 EPS | FY28 EPS | PT | View |
|---|---|---|---|---|---|---|
| KGI (Hanhsuan Shen) | 2026-03-12 | NT$30.04 | NT$107.25 | n/a | NT$300 | OW initiation, lone bull outlier |
| STF Research | 2026-04-21 | 30.04 (validates KGI) | 107.25 | n/a | implied ~NT$1,609 at 15× FY27 | Bull, builds KGI math from first principles |
| Morgan Stanley (Tiffany Yeh) | 2026-04-19 | NT$5.90 | NT$19.61 | n/a | NT$202 | OW but EPS only ~1/5 of KGI |
| Consensus | rolling | NT$4.59 | NT$8.03 | n/a | NT$176 | Sell-side average |
Current price 2026-04-27: ~NT$131-132. At consensus FY27 EPS NT$8.03, that's 16× forward. At KGI FY27 NT$107.25, that's 1.2× forward — STF's "placeholder, not a valuation."
KGI's three-multiplier bull math (STF first-principles validation)
STF's "Memory Trade After the Memory Trade" piece (Apr 21 2026) defends KGI's 265× revenue claim by decomposing into three independent multipliers:
| Multiplier | Mechanism | Magnitude |
|---|---|---|
| 1. Wafer starts | FY25 ~1k wpm → 4Q26 8k → 3Q27 17k. ASML Immersion DUV 18-month lead time gates 3Q27 step. | ×17 |
| 2. Bit per wafer | 19nm MLC = 28,416 Gb/wafer; 3D TLC 48L = 129,920 Gb (4.6×); 96L = 200,192 Gb (7.0×). Blended FY26-27 ≈ ×2.5-7. | ×2.5–7 |
| 3. ASP/Gb | FY25 US$0.07 → KGI FY26 US$0.42 (+505% YoY) → FY27 US$0.53 (+24.8% YoY). 1Q26 contracts already moved: 8GB +183%, 4GB +165%, 16GB +147%. | ×7.5 |
Combined: ×35 bit output × ×7.5 ASP = ×262 revenue. FY25 eMMC = TWD 867M. KGI FY27e eMMC = TWD 230B. The 265× headline is real if the multipliers compound on schedule.
STF identifies 4 risks (vs KGI's single risk)
KGI flagged only equipment delivery delay. STF surfaces three more, all of which would push FY27 numbers right:
- 3D TLC yield risk — Macronix's 96L process needs 68-70% yield by mid-2026. Major NAND OEMs typically need 4-8 quarters to ramp 50% → 70%. If 96L comes in at 55% rather than 68% in 2H26, the 4Q26 eMMC revenue figure of TWD 39,714M does not arrive on schedule. No independent visibility into Macronix yield data.
- ASML equipment delivery — 3Q27 step (10k → 17k wspm) entirely contingent on Immersion DUV arriving on time. 18-month lead time = base-rate risk, not tail.
- Chinese substitution — TVs + STBs are ~52% of low-cap eMMC TAM by volume. GigaDevice has Chinese controller IP + OEM relationships. If GigaDevice + YMTC pair up, takes Chinese consumer share. Geopolitical de-risking partially offsets.
- Winbond re-entry — at 80%+ GM sustained for two years, Winbond has clear economic incentive to invest in TLC eMMC. 24-36 months from decision to volume = competition arrives 2028-2029. Caps how long elevated margins persist.
MS adds three nuances (Apr 19 2026 "Old Memory: Ongoing Asymmetric Risks")
Morgan Stanley is OW Macronix but more conservative than KGI:
- MS-Macronix PT NT$202 (vs KGI's NT$300, MS's bull case NT$326, bear case NT$119.5).
- Risk/reward skew 17.0 — extremely asymmetric upside.
- MS FY26 EPS NT$5.90 is 44.9% below consensus; FY27 EPS NT$19.61 is 23.7% below consensus.
- Why MS conservative: MS implicitly assumes some combination of slower wafer-in ramp, lower 96L yield, or ASP not fully holding through FY27.
- MS confirms MLC NAND/NOR pricing power accelerates from 2Q26 led by Macronix — directional agreement with KGI/STF on the structural shift.
- MS: AP Memory peer disclosure point — "the most important factor to focus on is prepayment from end-customers" — meaning prepayment-backed LTAs are spreading even into mid-cap Taiwan memory.
- Spec migration highlight from PSMC into 2027 — MS sees this as supportive backdrop.
Industry context — chicken game over (SemiConSam, Apr 26 2026)
Cross-reference: SemiConSam's "Why did the memory chicken game keep repeating—and who ultimately survived?" piece argues the industry-wide commercial model is structurally re-rating:
- Big 3 (Samsung/SK Hynix/Micron) won't restart consolidation due to antitrust + strategic government backing
- HBM "foundry-ization" via cHBM (custom HBM) drives LTA spread across DRAM
- "Both an escape from commoditization, and the foundry-ization of memory will happen. Within the next two years."
- "Shouldn't we stop valuing memory companies by P/B now?"
This applies to Macronix indirectly: if the entire memory complex is re-rating off cyclical multiples, Macronix at 1.2× FY27 KGI EPS is dramatically underpriced — even at MS's NT$19.61, current ~131 is 6.7× forward.
LTA / customer-side disclosure gap
Macronix has the same disclosure gap as Kioxia. Per STF Q1 call:
Question: From the perspectives of technical difficulty, time to volume production, and equipment availability, what are the barriers for new entrants looking to enter the 2D NAND space? Separately, are customers interested in signing LTAs (Long-Term Agreements)?
Dr. Wu: "Their situation is very strange." I would recommend asking the companies that want to enter, not me.
Macronix did not commit to whether they're signing customer LTAs. This is the same gap Kioxia has on disclosure (vs SK Hynix's 10-30% prepayments with MSFT/GOOG, SanDisk's 5 NBMs with $42B+ minimum revenue, Micron's HBM4 fully booked). If Macronix were signing LTAs, it would be a material durability-thesis upgrade. The fact that pricing is on monthly basis rather than fixed long-term is a tell — pricing power on the way up, but no contractual floor on the way down.
Net read for sizing (2026-05-03)
The thesis is directionally right across all 4 sources. The disagreement is execution risk on KGI's pace.
| Scenario | Where stock goes | Implied multiplier |
|---|---|---|
| KGI right (NT$300 PT) | +127% from NT$132 | KGI 10× FY26 EPS |
| STF stretch (15× FY27 KGI EPS) | +1,120% to NT$1,609 | re-rating to growth multiple |
| MS right (NT$202 PT) | +53% from NT$132 | MS 34× FY26 EPS |
| Consensus right (NT$176 PT) | +33% from NT$132 | consensus 38× FY26 EPS |
Entry discipline (unchanged): wait <NT$110, OR post-Q1 2026 earnings (~early August 2026) beat on either ASP pass-through or wafer-in ramp execution. Q1 2026 earnings is the first datapoint to test KGI's wafer-in ramp + ASP pass-through assumption — if Q1 2026 prints ASP/Gb in line with the 1Q26 contract +183% prints (8GB), the bull case is intact. If it prints below contract (negotiated terms with major OEMs), KGI's NT$300 starts to look more like MS's NT$202.
Sources for this synthesis: macronix-kgi-2026-translation, KB/raw/substack-archive/stf-research/2026-04-27-macronix-uncovering-alpha-beyond-slides.md (+ sibling .pdf), KB/raw/substack-archive/stf-research/2026-04-21-macronix-memory-trade-after-the-memory-trade.md (+ sibling .pdf), _Pitch Now/Memory/Old Memory- Ongoing Asymmetric Risks_MS_26.04.19.pdf, KB/raw/substack-archive/semiconsam/2026-04-26-why-did-the-memory-chicken-game-keep.md, KB/raw/substack-archive/viks-newsletter/2026-04-26-twic-credo-eats-dust-cpu-and-nand.md.
Supplementary STF Q1 call data points (added 2026-05-12 from full PDF re-read)
The May 3 synthesis above captured the narrative quotes from the STF post. A re-read of the full STF Q1 call digest surfaces six discrete operating data points worth pinning down separately — these are facts, not interpretation, and they materially tighten the model:
- Capex allocation explicit — NT$22B entirely to eMMC + NAND. Dr. Wu: "What you can infer is that the NT$22 billion investment should be entirely directed toward eMMC and NAND capacity additions." No NOR, no SSD, no foundry split. Confirms the KGI bull math premise that 100% of new capex flows into the eMMC/NAND ramp.
- 12-inch fab practical ceiling ~25K wspm (not 30K). Dr. Wu: "As for reaching 30,000 wspm, the 12-inch fab probably cannot get there. [...] The practical ceiling is approximately 25,000 wspm." 3D NAND is too process-step-intensive. Caps the KGI 4Q26 8K → 3Q27 17K → terminal ceiling — Macronix cannot get to 30K at the 12" fab without additional capex. KGI's 17K 3Q27 number is consistent with this 25K ceiling; getting beyond requires the 192L+/300L step.
- MLC density-range monopoly — Macronix the only supplier at 4GB-8GB-16GB-32GB. Dr. Wu: "In the 4GB-8GB-16GB-32GB density range, there are basically no large companies providing supply anymore. [...] In that specific density range, Macronix should be the only supplier that can support these customers." Samsung shifted up to 64/128GB MLC. Pins the monopoly framing to specific density bands — sub-32GB eMMC is the moat zone.
- All new equipment fully completed H2 2027. Dr. Wu: "All new equipment is primarily 3D-based. I believe all of it should be fully completed by H2 2027." Concrete date for the capacity-ramp-completion milestone — relevant to the KGI 3Q27 17K wspm assumption and the STF/MS divergence on FY27 wafer-in ramp pace.
- NOR:NAND ratio 1:0.5 today → 1:1 post-additions. Dr. Wu: "NOR versus NAND (including eMMC NAND and raw NAND) is currently approximately 1 to 0.5. After capacity additions, the goal is for the ratio to reach approximately 1 to 1." Implies NAND output doubles relative to NOR through the capex cycle — a clean way to size the eMMC bit-output assumption against KGI's wafer-starts × bit-per-wafer × ASP framework.
- LEO satellite / Starlink: Macronix the only Western-qualified supplier for satcom NOR. Dr. Wu: "Macronix's products, as of now, are the only products verified by Western markets (US and Europe) to be fully compliant with satellite communication requirements." Two reasons: extreme-temperature tolerance (-50°C+) and radiation hardness. Customers came proactively. Order volumes not disclosed. This is a net-new high-spec NOR revenue stream not modeled in any of KGI/STF/MS — it's a small-but-high-margin optionality kicker on top of the eMMC thesis. Worth tracking as a Starlink/LEO constellation buildout proxy.
Net effect on existing synthesis tables: None of the four FY26/27 EPS forecasts (KGI / STF / MS / Consensus) need revision from these points. They tighten the operating assumptions underlying the KGI math (25K wspm ceiling caps the wafer-starts multiplier; NT$22B confirms the capex direction; H2 2027 dates the wspm ramp completion). The Starlink line is an additive optionality not in any model. Entry discipline unchanged.
Identity
| Field | Value |
|---|---|
| Name | Macronix International Co., Ltd. (旺宏電子) |
| Ticker | 2337.TW (TWSE) |
| Sector | Memory — NAND Flash, NOR Flash, ROM, eMMC, Foundry |
| HQ | Hsinchu, Taiwan |
| Founded | 1989 |
| Manufacturing | One 12-inch fab (Fab 5, ~22 kwpm) + one 8-inch fab (45 kwpm) |
| Shares outstanding | 1.86 bn |
| Market cap (2026-04-24) | NT$245 bn / ~US$7.6 bn |
| Spot price (2026-04-24) | NT$132.0 |
| 52-week range | NT$18.15 – NT$170 |
| 12-month price action | +551% over the last 12 months; ran from NT$18 to NT$170 then back to NT$132 |
The thesis in plain language
Major NAND foundries (Samsung, SK Hynix, Kioxia, Micron) are end-of-life'ing their MLC product lines to free up capacity for higher-margin HBM and high-density TLC/QLC NAND. The end-applications that need low-capacity (≤32GB) eMMC — TVs, set-top boxes, smartwatches, automotive telematics, drones, IPC, networking gear — cannot easily switch to TLC because (a) MLC dies have IO advantages at low capacities, (b) controllers and firmware would need full redesigns, and (c) TLC is itself supply-tight. As the Big Four exit, Macronix becomes effectively the only sub-32GB eMMC supplier post-2028.
That's the bull case. The thesis was first surfaced in Pink's network by Collyer Bridge (illyquid) in their 17 April 2026 APAC wrap — framed there as Macronix "emerging as a highly profitable monopoly." KGI Securities then formalized the numbers in their 12 March 2026 initiation: Outperform, NT$300 12-month target (203% upside from the NT$99 cover price).
The full report breakdown lives in macronix-kgi-2026-translation. This page synthesizes and stress-tests the thesis.
Business mix
2025A revenue mix (from KGI Fig 22):
- NOR Flash: 61% — the legacy core; multiple-decade business
- ROM: 17% — Japan-customer-heavy, mature
- SLC NAND: 13% — duopoly with Winbond after Kioxia/Micron pull back
- FBG (foundry): 7%
- eMMC: 3% — historically tiny, the new story
2027F revenue mix (KGI projection):
- eMMC: 75%
- NOR Flash: 17%
- SLC NAND: 4%
- ROM: 5%
- FBG: 3%
If KGI's numbers play out, this is a business-model transformation in two years, not just a cycle upturn. eMMC goes from rounding error to three-quarters of revenue. The whole investment debate is whether 2026/27 actually deliver that mix.
Why the mix shift is asymmetric on margins
- 2025 GM: 17.8%
- 2026F GM: 68.1% (KGI)
- 2027F GM: 82.3% (KGI)
GM jumps 50 ppts in one year because eMMC ASP/Gb is forecast +505% (1Q26 contract prices already up 100–200% QoQ across capacity tiers — this part is observed, not modeled) while bit cost falls thanks to 3D TLC NAND (48L/96L) Bit-per-Wafer being 3–9× the legacy 19nm 2D process. Short of a demand collapse, the margin print should be extreme — though 82% GM in a memory company is a number normally reserved for software businesses, and warrants healthy skepticism.
Demand stack — why MLC eMMC has long-tail demand
Total 2025 ≤32GB eMMC bit demand (KGI Fig 3): 80.55 bn Gb. Breakdown:
| End market | Bit demand | % of total | KGI's read-through |
|---|---|---|---|
| TV | 22.6 bn Gb | 28% | Largest, but capacity creep (8/16GB → 32/64GB) means upgrade risk to TLC |
| Set-Top Box | 19.6 bn Gb | 24% | HDD-replacement story, mostly 8/16/32GB — sticky low-capacity |
| Smartwatch | 14.4 bn Gb | 18% | 75% eMMC penetration; rest is custom (Apple Watch / some Samsung) |
| Other (drone/IPC/networking) | 13.4 bn Gb | 17% | Long-tail industrial, very sticky |
| Automotive | 9.5 bn Gb | 12% | 8+ year supply cycle requirements — switching cost is contractually enforced |
| Server (BMC) | 0.9 bn Gb | 1% | Boot ROM for server BMC chips — ubiquitous, low priority |
| Mobile | 0 | 0% | Smartphones/tablets are 64GB+ now, fully on UFS/TLC |
Critical structural point: automotive customers contractually require 8+ years of supply continuity. Even if Samsung wanted to keep some auto MLC, those programs have decade-long lifecycles, so once a Tier-1 specifies Macronix into a 2027 vehicle program, that's locked-in revenue through 2035-ish. This is the "long tail" CB and KGI both reference.
Supply collapse — the mechanical case
Total MLC bit supply, KGI projection:
| Year | 2025 | 2026F | 2027F | 2028F |
|---|---|---|---|---|
| MLC bit supply (bn Gb) | 91.8 | 44.9 | 31.3 | 36.5 |
| YoY | — | -51% | -30% | +17% |
| Macronix share of supply | 0.4% | 19.6% | 90.1% | 100% |
Samsung, SK Hynix, and Micron go to zero by 2027. Kioxia hangs on through 2027 then exits. Macronix's wafer-in ramp from 1 kwpm (2025) to 17 kwpm (3Q27) — combined with Bit-per-Wafer multiples from the 3D process migration — is what fills the gap.
The 2028 +17% YoY supply growth is entirely Macronix. Every other player is dark by then.
What the market is already pricing in
This is where the thesis gets uncomfortable for an entry today.
| Metric | Value | Read |
|---|---|---|
| Spot price | NT$132 | +33% above KGI's NT$99 cover price |
| 12M absolute return | +551% | Already a 7-bagger off the 2025 lows |
| Forward P/E (yfinance) | 4.7× | Extremely cheap if EPS holds |
| Forward EPS (yfinance) | NT$28.04 | Vs KGI 2026F NT$30.04 — the street has moved up to KGI |
| 7-analyst target mean | NT$176 | +33% upside from spot |
| Target median | NT$170 | +29% upside |
| Target high | NT$300 | KGI is the high |
| Target low | NT$67 | Implies -49% downside |
The forward EPS number is the key tell. When KGI initiated on 2026-03-12, consensus 2026 EPS was NT$4.59 (per their own Fig 16). Six weeks later, consensus forward EPS sits at NT$28 — i.e., the street has revised up by 6× to roughly KGI's view. The sell-side bull case is no longer contrarian; it's the new consensus.
What's still contrarian: KGI's 2027 EPS of NT$107 vs consensus NT$8.03. That number — implying a 257% YoY EPS growth on top of the 2026 explosion — is the real bull case. If you believe KGI's 2027, the stock at 4.7× FY26 forward and ~1.2× FY27 forward is enormously cheap.
Translating KGI's NT$300 target
- 10× 2026F EPS NT$30.04 = NT$300 (KGI's framing)
- 2.8× 2027F EPS NT$107.25 — i.e., target is below fair value if 2027 plays out
- Implies market cap NT$580 bn (~US$18 bn) at 1.86 bn shares
For comparison, Nanya Technology (DDR4 ~25% share) trades at ~NT$740 bn market cap.
Catalysts (what moves the story up)
- 1Q26 / 2Q26 earnings — KGI projects 1Q26 EPS NT$0.46 (vs 4Q25 -NT$0.16) and 2Q26 NT$3.02. First print is the live test of the eMMC ramp narrative.
- MLC/TLC wafer-in ramp delivery — KGI's bridge needs 1Q26=2k → 4Q26=8k. Watch for Macronix capex commentary on equipment delivery.
- eMMC contract price action 2Q–4Q26 — KGI projects +60–100% in 2Q26, then +30–50% in 3Q26. Independent contract-price scrapes will validate or break this.
- Big Four MLC EOL announcements — Samsung's October 2025 EOL is the anchor. SK Hynix and Micron formalizing similar timelines tightens the supply story.
- 3D TLC NAND yield curves on 48L/96L — Macronix has not previously run these nodes at scale. Yield surprises (positive or negative) materially shift the BPS multiplier.
Risks (what breaks it)
KGI's stated risk
- Capacity-expansion equipment delivery delays. ASML Immersion DUV is on 18-month lead times; the 3Q27 step from 10k → 17k wafer-in MLC/TLC depends on tools that need to be ordered now.
Wiki reviewer additions
-
Consensus has caught up. The asymmetry that existed when KGI initiated at NT$99 is gone at NT$132 with forward EPS now at NT$28. The trade was "everyone underestimating Macronix"; now everyone agrees with KGI's 2026 number. Position-sizing logic that worked at NT$50 doesn't work at NT$132.
-
2027 EPS NT$107 is heroic. It assumes (a) eMMC ASP/Gb +24.8% on top of 2026's +505% — i.e., no demand pushback after one of the steepest price moves on record; (b) further wafer-in expansion to 17k by 3Q27 with no equipment slip; (c) 3D TLC NAND yields ramp to 75–76% on a node Macronix has never volume-produced. Any one of these slipping by a quarter drops 2027 EPS materially.
-
Pre-buy-in consensus is suspect. When the only new dollar of demand is "narrative-driven retail + sell-side momentum following price," 6× consensus EPS revisions in six weeks is not a confidence signal — it's a sentiment signal. Reflexivity cuts both ways.
-
Macronix lost money 2023–25. Cumulative ~NT$8.2 bn in net losses across the three years (NT$1.7 bn / NT$3.2 bn / NT$3.3 bn). Operating cash flow turned positive in 2025 (NT$4.8 bn) on inventory drawdown, but free cash flow has been negative or marginal for years. The market is being asked to underwrite a 56× net income turn in one year, on the back of a balance sheet that took three years of losses. Net debt at 4Q25 was NT$9.3 bn; KGI projects net cash by end-2026 if the eMMC ramp delivers.
-
TLC substitution risk past 2027. KGI's whole argument that MLC isn't substitutable rests on three things: die IO efficiency, controller redesign cost, and TLC supply tightness. If TLC supply normalizes in 2028+ (e.g. YMTC's three new fabs that CB flagged in their 17 April wrap each producing 100k wpm, with DRAM optionality), the controller-redesign capex becomes a one-time cost the customer base will eat to escape Macronix's monopoly pricing. The monopoly is durable only as long as TLC is also tight.
-
YMTC is the wildcard. Three new Chinese NAND fabs (per CB's reporting) come online over the 2026–28 window, each producing 100k wpm, with Wuhan's third fab >50% domestic-equipped. Even if YMTC focuses on TLC and DRAM, the bit-supply additions to global NAND tighten the case for sub-32GB MLC monopoly pricing past 2028. KGI's model effectively assumes YMTC doesn't matter — that's a heroic assumption for any 2028+ projection.
-
Sell-side initiation pattern. KGI is the high target on the street (NT$300 vs median NT$170). Sell-side initiations with 200%+ upside on a stock that's already 7-bagged in 12 months are pattern-match for momentum-following research, not contrarian work. The actual contrarian call would have been at NT$18–25 last summer.
Financials snapshot (yfinance + KGI projection)
Historical (NT$ bn except EPS)
| 2022A | 2023A | 2024A | 2025A | |
|---|---|---|---|---|
| Revenue | 43.5 | 27.6 | 25.9 | 28.9 |
| Gross profit | 19.2 | 6.8 | 6.1 | 5.1 |
| GM (%) | 44.2 | 24.5 | 23.6 | 17.8 |
| Operating profit | 9.4 | -2.4 | -3.9 | -3.7 |
| Net income | 9.0 | -1.7 | -3.2 | -3.3 |
| EPS (NT$) | n/a | -0.92 | -1.73 | -1.77 |
| EBITDA | 15.0 | 2.6 | 1.7 | 1.9 |
| FCF | 1.7 | -8.2 | -5.6 | 3.0 |
| Net debt | -3.3 | 8.3 | 14.6 | 10.0 |
Macronix did make money in 2022 (NT$9.0 bn net income, 44% GM) — that's the prior-cycle peak. The current thesis is for 2026/27 to massively exceed that peak.
KGI forward (NT$ bn except EPS)
| 2026F | 2027F | |
|---|---|---|
| Revenue | 112.4 | 305.1 |
| GM (%) | 68.1 | 82.3 |
| Operating profit | 65.5 | 233.9 |
| Net income | 56.1 | 200.4 |
| EPS (NT$) | 30.04 | 107.25 |
| EBITDA Margin | 62.9% | 79.1% |
| FCF | 17.6 | 164.4 |
| Net cash | yes | yes |
| Pre-tax ROCE | 64% | 164% |
A 164% ROCE in 2027F is the kind of number that is either truly extraordinary or wrong by a multiple. The market will figure that out within a couple of quarters.
Decision log
2026-04-25 — Initial wiki page created from KGI initiation translation + CB reference + yfinance snapshot. No position. Thesis is intellectually compelling but timing is hard:
- Entry at NT$132 already prices in the bull EPS revision (forward P/E 4.7× FY26 with consensus already near KGI's number).
- Asymmetry has compressed massively since CB flagged it on 17 April at the prior price level.
- The "real" call is now on KGI's 2027 EPS NT$107, which the street has not yet underwritten — but underwriting that call requires belief in (a) eMMC ASP/Gb still +25% off a +505% base, (b) wafer-in ramp on schedule with no ASML DUV slip, (c) 3D TLC NAND yields ramping on a node Macronix has never produced at scale.
- Watch list, no position. Re-examine after 1Q26 earnings (date TBD) — that print is the first real test of the eMMC ramp narrative. If 1Q26 EPS prints meaningfully above KGI's NT$0.46 with a credible 2Q26 guide, thesis hardens. If it disappoints, the consensus revision unwinds fast and the stock retraces.
- Position-sizing context: this is a single-name, single-narrative, sell-side-driven bet on a Taiwanese small-cap memory company with a recent loss history. If a position is taken, size for total loss: 1% on the bull conviction, 2% on the home-run conviction.
Cross-references
- macronix-kgi-2026-translation — full English synthesis of KGI's 12 March 2026 initiation report (18 pp, all figures and tables preserved)
- data-center-memory-types — memory hierarchy primer; this thesis lives in the NAND/eMMC sub-segment that primer covers in passing
- source-illyquid — Collyer Bridge (illyquid) flagged Macronix in their 17 April 2026 APAC wrap as the original surfacing into Pink's research network
- watchlist — 2337 row, source-attributed to illyquid/CB
- inv-q — both the original CB article and this synthesis logged
Source provenance for this page
- Equity research: KGI Securities Investment Advisory, "Macronix — A Better Story Than DDR4," 18 pp, 2026-03-12 (Mandarin original). Translation in macronix-kgi-2026-translation; PDF archived at
raw/equity-research/macronix-kgi-2026-03-12.pdf. - Independent commentary: Collyer Bridge / illyquid, "APAC Wrap: 17 April 2026," substack, paywalled past the lede. Free preview captured 2026-04-25; key Macronix paragraph: "In brief, the Macronix (2337.TW) thesis is that it is emerging as a highly profitable monopoly in the eMMC memory market as the 'Big Four' memory giants abandon this market and reallocate their capacity toward high-margin HBM and high-end NAND flash."
- Live market data: yfinance, pulled 2026-04-25 (price, financials, analyst targets, earnings history).
- Historicals: Macronix annual financials 2022–2025 from yfinance / KGI Fig 14.
Briefings
- 2026-04-26 · Memory Sector Primer: HBM, NAND, and the 2026 Up-Cycle · vault
Source updates (auto-maintained)
Drop/Books (Feb 25, 25) - SKOREA_20240820_2100
Morgan Stanley's August 2024 "Cycle Playbook" lists Macronix explicitly under "Least Favored / Cyclical rotation" alongside Micron and Phison, flagging it as a late-cycle sell amid rising inventory and decelerating demand.
Relevant to your thesis: Contextualizes MS's conservative NT$202 PT (vs KGI's NT$300) — MS was already bearish on Macronix pre-eMMC thesis, framing it as commodity cyclical exposure rather than the structural supply-gap monopoly KGI/STF identified in early 2026.
Source: dropfile://Books/Macro/SKOREA_20240820_2100.pdf
Drop/2. Frontend (Apr 21, 26) - Old Memory- Ongoing Asymmetric Risks_MS_26.04.19
Morgan Stanley reiterates OW on Macronix (PT NT$202, bull case NT$326, bear case NT$119.5), flagging MLC NAND/NOR pricing power accelerating from 2Q26, with both up 100% possible in 2Q, maintained at high double digits into 2H; FY26e EPS NT$5.90, FY27e NT$19.61.
Relevant to your thesis: Directly corroborates the supply-shortage pricing thesis while anchoring the conservative EPS scenario already tabled in the KGI/STF/MS triangulation — risk/reward skew of 17.0 confirms asymmetric upside framing.
Source: dropfile://2. Frontend/Memory/Old Memory- Ongoing Asymmetric Risks_MS_26.04.19.pdf
Drop/2. Frontend (May 3, 26) - Macronix_ The Memory Trade After the Memory Trade
STF Research argues Macronix is the only vertically integrated MLC/TLC eMMC supplier remaining after Samsung/SK Hynix/Kioxia exit, with KGI's 265× revenue projection decomposed into three compounding multipliers (×17 wafer starts, ×2.5–7 bit output, ×7.5 ASP/Gb) that STF validates from semiconductor first principles.
Relevant to your thesis: This is the primary source already embedded in the 2026-05-03 synthesis — the three-multiplier table, 47% supply deficit framing, and Winbond re-entry timeline all originate here.
Source: dropfile://2. Frontend/Memory/Macronix/Macronix_ The Memory Trade After the Memory Trade.pdf
Drop/2. Frontend (Apr 28, 26) - Macronix- The Memory Trade After the Memory Trade
STF Research validates KGI's 265× eMMC revenue projection as mathematically sound from first principles, decomposing it into three independent multipliers: ×17 wafer starts, ×2.5–7 bit-per-wafer (3D TLC migration), and ×7.5 ASP/Gb — yielding FY27e eMMC revenue of ~TWD 230B from a FY25 base of TWD 867M.
Relevant to your thesis: This is the primary source already synthesized in the 2026-05-03 update; the Drop copy is a duplicate — no new information, no revision needed.
Source: dropfile://2. Frontend/Memory/Macronix/Macronix- The Memory Trade After the Memory Trade.pdf
Drop/2. Frontend (Mar 17, 26) - Macronix KGI
KGI initiated 2337.TW with an Outperform rating and NT$300 target (March 12, 2026), projecting 265× eMMC revenue growth by 2027 driven by MLC supply collapse, Macronix's 1k→17k wspm wafer ramp, 3D TLC bit-per-wafer uplift, and ASP/Gb rising 505%/25% in FY26/27 to US$0.42/0.53.
Relevant to your thesis: This is the KGI initiation report underlying all downstream analysis in the wiki — the NT$30.04/NT$107.25 FY26/27 EPS forecasts, the three-multiplier bull math, and the supply-EOL timeline (Samsung LTS 1Q26, Kioxia 2027, Micron 2027) are the primary source documents already synthesized above.
Source: dropfile://2. Frontend/Memory/Macronix/Macronix KGI.pdf