Macronix (2337.TW) — KGI Initiation, English Synthesis
Source: KGI Securities, "旺宏 (Macronix) — A Better Story Than DDR4," 18 pp, Mandarin Report date: 2026-03-12 Translation date: 2026-04-25 Translator: Claude (synthesis, not literal). All figures, tables, and numeric forecasts preserved. Original PDF archived at
raw/equity-research/macronix-kgi-2026-03-12.pdf. Canonical wiki page: 2337
Header card
| Coverage type | Initiation |
| Rating | Outperform (增加持股) |
| 12-month target price | NT$300 |
| Closing price (March 12, 2026) | NT$99.00 |
| Implied upside | +203% |
| Prior target | N/A |
| Sector / region | Memory / Taiwan |
Trading data
| Metric | Value |
|---|---|
| Market cap | NT$196.0 bn / US$6,169 mn |
| Shares outstanding | 1,980 mn |
| Foreign ownership | 433.7 mn shares |
| 3M average daily volume | 131.4 mn shares |
| 52-week range | NT$17.05 – NT$116.0 |
| 3M / 6M / 12M absolute price performance | +159.8% / +338.1% / +351% |
| 3M / 6M / 12M relative price performance | +140.7% / +306.3% / +300.3% |
EPS history (NT$, A = actual, F = forecast)
| Year | 1Q | 2Q | 3Q | 4Q |
|---|---|---|---|---|
| 2024 | (0.58)A | (0.15)A | (0.16)A | (0.84)A |
| 2025 | (0.47)A | (0.69)A | (0.47)A | (0.16)A |
| 2026 | 0.46F | 3.02F | 7.42F | 19.14F |
Headline thesis (KGI's three bullets)
- Supply-demand imbalance drives a sharp eMMC price surge.
- eMMC revenue is set for explosive growth.
- Macronix enters a long-term profit cycle.
KGI initiates with Outperform and NT$300 target based on 10× 2026 EPS, corresponding to the upper end of Macronix's 2017–2021 P/E band. The firm argues rising earnings visibility justifies a P/E framework.
Risk: Capacity-expansion equipment delivery delays.
1. The setup — eMMC structural shortage drives a long profit cycle
KGI's core claim: Macronix turns profitable in 2026 and enters a long, stable profit cycle, with valuation re-rating potential. Five reasons:
(1) MLC supply collapse
Major NAND foundries are progressively EOL'ing MLC, leaving a large supply gap. Global MLC bit supply falls from 91.77 bn Gb (2025) to 36.49 bn Gb (2028), with YoY growth of −51% / −30% / +17% for 2026 / 2027 / 2028.
(2) High switching cost + TLC tightness slows the migration; MLC gap widens fast
Theoretically, low-capacity eMMC can substitute MLC with TLC. In practice, three frictions block fast migration:
- Die-configuration efficiency differences,
- Controller and firmware redesign cost,
- Current TLC supply is also tight, so part of MLC demand cannot rapidly switch.
With supply exiting fast and demand only slowly drifting down, the MLC supply-demand gap widens through 2026–27, pulling eMMC into an upcycle.
(3) Macronix aggressively expands capacity to absorb eMMC demand; MLC/TLC wafer-in scales fast 2026–27
Macronix is positioned to be the only global supplier of low-capacity (≤32GB) eMMC after 2028. The company is rapidly adding new fab capacity:
- MLC/TLC wafer-in rises from ~1k wpm (2025) to ~8k wpm by 4Q26, then to ~17k by 3Q27.
- 3D TLC NAND is produced on 48L and 96L processes, with Bit per Wafer 3–9× the 19nm legacy node.
- Combined wafer-in expansion + Bit-per-Wafer uplift drive eMMC bit output +2,590% YoY in 2026 and +222% in 2027.
(4) eMMC revenue mix jumps from 3% (2025) to 75% (2027)
- KGI projects Macronix eMMC ASP/Gb of US$0.42 in 2026 (+505% YoY) and US$0.53 in 2027 (+24.8%).
- Bit shipment growth: +994.5% (2026) and +226.8% (2027).
- Combined effect: eMMC revenue scales NT$0.86 bn (2025) → NT$56.39 bn (2026) → NT$230.1 bn (2027) — a 265× expansion over three years.
(5) MLC supply contraction is irreversible; eMMC long-tail demand sustains Macronix
With major NAND OEMs exiting MLC and dismantling associated equipment, the low-capacity eMMC supply contraction is structurally irreversible. KGI estimates 2026/27/28 MLC supply gaps of 36% / 47% / 26%, deeper than the 20–30% DDR4 shortage of 2025–26. Even with Macronix's 2027 eMMC bit shipment up 36× from 2025, its share of total eMMC demand is still only ~25% — leaving substantial growth runway. Automotive, drone, IPC, and networking applications anchor the long-tail demand, so KGI argues Macronix can sustain stable profits "for the foreseeable years."
2. Low-capacity eMMC: only supplier post-2028
MLC supply structure shifts dramatically as major foundries EOL. NAND has matured rapidly. Cleanroom space is finite, and MLC has the lowest unit value of NAND types. To pursue scale and capital efficiency, major NAND OEMs are concentrating resources on TLC, QLC, and DRAM, and are progressively EOLing MLC.
Per KGI supplier checks, NAND OEM MLC EOL progression:
| OEM | Decision | Status / LTS schedule | KGI MLC wafer-in forecast |
|---|---|---|---|
| Samsung | Decided EOL late October 2025 | Commercial customer LTS = end of 1Q26; auto LTS = end of 3Q26. Retains minimal capacity for some auto customers | 100k → 33k → 0 (2025/26/27) |
| Micron | Notified auto customers + module makers of MLC exit | Currently very limited supply; mostly selling existing inventory | 12k → 5k → 0 (2025/26/27) |
| Kioxia | MLC supply has been declining for years | LTS preliminarily set at end of 2027 | 15k → 10k / 5k → 0 (2025/26/27/28) |
| SK Hynix | Strategy similar to Kioxia, actively exiting | — | 20k → 10k → 0 (2025/26/27) |
Against this, Macronix MLC/TLC wafer-in: 1k → 6k → 14k → 18k (2025/26/27/28) — the only supplier left at the end-state.
Figure 2: MLC NAND bit supply forecast (full table)
| MLC wafer-in (k pcs/month) | 2025 | 2026F | 2027F | 2028F |
|---|---|---|---|---|
| Samsung | 100 | 33 | 0 | 0 |
| SK Hynix | 20 | 10 | 0 | 0 |
| Kioxia | 15 | 10 | 5 | 0 |
| Micron | 12 | 5 | 0 | 0 |
| Macronix (MLC/TLC) | 1 | 6 | 14 | 18 |
| Total | 148 | 64 | 19 | 18 |
| Bit shipment (mn Gb) | 2025 | 2026F | 2027F | 2028F |
|---|---|---|---|---|
| Samsung | 62,208 | 20,529 | 0 | 0 |
| SK Hynix | 12,442 | 6,221 | 0 | 0 |
| Kioxia | 9,331 | 6,221 | 3,110 | 0 |
| Micron | 7,465 | 3,110 | 0 | 0 |
| Macronix (MLC/TLC) | 326 | 8,770 | 28,198 | 36,492 |
| MLC bit supply total (mn Gb) | 91,772 | 44,851 | 31,309 | 36,492 |
| YoY (%) | — | (51) | (30) | 17 |
KGI forecast: total MLC bit supply falls from 91.77 bn Gb (2025) to 36.49 bn Gb (2028).
3. Demand decomposition — TV and STB anchor low-capacity eMMC
MLC is used in eMMC, SSD, UFS, ROM, SD Card, UFD, and other NAND products. Today, MLC is mostly used for low-capacity eMMC (2–16 GB). KGI breaks down 2025 global eMMC TAM by end-application:
| Application | 2025 unit shipment | eMMC penetration | Capacity (mainstream) | 2025 ≤32GB eMMC bit demand | % of total ≤32GB demand |
|---|---|---|---|---|---|
| TV | 196 mn | 100% | 8/16 GB → 32/64 GB | 22.6 bn Gb | 28% |
| Set Top Box | 179 mn | ~90% (replacing HDD) | 8/16/32 GB | 19.6 bn Gb | 24% |
| Smartwatch | 172 mn | 75% (Apple Watch + some Samsung use other) | 8/16 GB | 14.4 bn Gb | 18% |
| Automotive | 84 mn vehicles | TCU + Infotainment + minor ADAS | TCU 8/16 GB; Infotainment 32/64 GB | 9.5 bn Gb | 12% |
| Server (BMC) | 12.24 mn BMC | 100% | 8 GB → 32/64 GB | 0.94 bn Gb | 1% |
| Mobile (smartphone/tablet) | 1,477 mn | 38% (UFS taking share) | 64/128 GB — not ≤32GB | n/a | n/a |
Total 2025 ≤32GB eMMC bit demand: 80.55 bn Gb, vs MLC bit supply 91.77 bn Gb — implying ~14% oversupply in 2025. This flips fast as MLC supply collapses 51% in 2026.
Figure 3: Low-capacity (≤32 GB) eMMC bit demand forecast (consolidated)
| 2025 | 2026F | 2027F | 2028F | |
|---|---|---|---|---|
| Total ≤32 GB eMMC bit demand (mn Gb) | 80,552 | 70,293 | 58,595 | 49,015 |
| YoY (%) | — | (12.7) | (16.6) | (16.4) |
| TV | 22,600 | 15,600 | 10,337 | 7,548 |
| STB | 19,611 | 15,353 | 9,250 | 7,262 |
| Server (BMC) | 940 | 1,127 | 1,246 | 920 |
| Smartwatch | 14,423 | 16,991 | 19,005 | 18,533 |
| Automotive | 9,554 | 9,507 | 8,991 | 6,582 |
| Other (drone/IPC/networking) | 13,425 | 11,716 | 9,766 | 8,169 |
KGI's gap math: 2026/27/28 MLC supply gaps of 36% / 47% / 26%, deeper than the 20–30% DDR4 shortage of 2025–26.
4. Switching cost + TLC tightness sustain MLC demand; 2026–27 gap widens
If end users upgrade ≤32 GB eMMC capacity to 32–128 GB, they shift to TLC dies — theoretically substituting away from MLC. KGI argues this won't happen quickly:
- Die-configuration efficiency: Multiple small-capacity dies typically have higher IO speed than a single die of equivalent capacity, so MLC architecture retains efficiency advantage in some applications.
- Controller and firmware redesign: Controllers are designed around specific die IO speeds. Switching die types may require firmware rewrites or full controller redesign — expensive in mask costs and development time.
So switching cost is high; price and capacity are not the only considerations.
TLC market is also structurally short. Demand side: (1) Korean OEMs prioritize high-margin DRAM, structurally squeezing NAND; (2) Nvidia BlueField4 DPU will carry 16TB eSSD — substantially raising per-unit NAND content; (3) OEMs are developing high-IOPS NAND for AI inference, with mass production from 2027, further squeezing general-purpose NAND.
Supply side: (1) Pre-2028 wafer-in expansion is limited to Samsung P4, Kioxia/Sandisk K2, YMTC Fab2B/3; (2) Samsung P4 and YMTC Fab3 will prioritize DRAM, so usable NAND wafer flexibility is limited.
Against this backdrop, 512 Gb TLC die contract prices have risen since 4Q25, reaching ~US$20.5 in 1Q26, while 128Gb / 64Gb MLC die prices reach ~US$13 / US$11. With switching cost compounded by similarly-high TLC pricing, KGI argues that beyond some TV and STB downgrade-to-Downgrade-NAND or upgrade-to-TLC cases, automotive, drone, IPC, and networking demand will sustain MLC long-tail demand.
KGI projects 2026–27 TLC NAND prices continuing higher, with low-capacity eMMC demand only mildly declining.
5. eMMC pricing — the upcycle thesis
Under supply-shortage backdrop, eMMC contract prices have risen since 4Q25, with quarterly increases in the 10–20% range, then sharply expanding to 100–200% in 1Q26:
- 8GB +183%
- 4GB +165%
- 16GB +147%
Looking forward:
- 2Q26: each capacity +60–100% QoQ
- 3Q26: +30–50% QoQ
- 4Q26: +20–30% QoQ
Figure 4: MLC vs TLC die contract prices (US$)
| 3Q25 | 4Q25 | 1Q26 | |
|---|---|---|---|
| 512Gb TLC | 4.8 | 8.6 | 20.5 |
| 128Gb MLC | 3.4 | 5.5 | 13.0 |
| 64Gb MLC | 3.3 | 5.5 | 11.0 |
Figure 5: eMMC quarterly contract price increases (% QoQ, KGI estimate)
| 1Q26 | 2Q26F | 3Q26F | 4Q26F | |
|---|---|---|---|---|
| 64GB | 30 | 100 | 50 | 30 |
| 32GB | 35 | 110 | 50 | 30 |
| 16GB | 147 | 183 | 50 | 30 |
| 8GB | 183 | 110 | 50 | 30 |
| 4GB | 165 | 60 | 30 | 20 |
KGI note: Macronix's 1Q26 eMMC ASP may differ from these contract prices because of negotiated terms with major OEMs.
6. Macronix capacity ramp — wafer-in mix shift
Macronix runs two fabs: a 12-inch fab (Fab 5, capacity ~22 kwpm at full utilization) and an 8-inch fab (capacity 45 kwpm, all tools running).
2025 — 12-inch fab @ 100% utilization, wafer-in mix:
| Product | Wafer-in (k wpm) |
|---|---|
| NOR Flash | 9–10 |
| ROM | 4 |
| SLC NAND | 3 |
| MLC/TLC NAND | 1 |
| R&D | 4–5 |
2025 — 8-inch fab @ 84% utilization, wafer-in mix:
| Product | Wafer-in (k wpm) |
|---|---|
| NOR Flash | 19–20 |
| Wafer foundry (FBG) | 13–14 |
2026F — 12-inch fab capacity reallocation:
| Product | Change |
|---|---|
| NOR Flash | ~9k (held) |
| SLC NAND | ~3k (held) |
| ROM | down to ~3k (Japan-customer 128Gb development slow) |
| R&D | down to 3–4k |
| MLC/TLC NAND | 1Q26 = 2k → 2Q/3Q/4Q = 7k/7k/8k |
The MLC/TLC wafer-in expansion comes from: (1) ~3–4k from ROM and R&D reallocation; (2) ~3k new capacity added in 2Q26. The next ~7k step (3Q27 onward) needs new ASML Immersion DUV equipment with current 18-month lead times — so KGI projects MLC/TLC wafer-in steps from 10k → ~17k from 3Q27.
Overall: MLC/TLC wafer-in ramps from ~1 kwpm (2025) → 8k (4Q26) → 17k (3Q27), implying full-year average wafer-in growth of +500% (2026) and +125% (2027).
7. 3D NAND adoption drives Bit per Wafer explosion
Macronix currently mass-produces 2GB / 4GB / 8GB eMMC on 2D NAND 19nm, mostly for consumer (e.g. wearable) and networking applications. In 2026, the company will mass-produce 8GB / 16GB / 32GB eMMC for both commercial and industrial grades; pending validation, related products will start wafer-in in 2Q26–3Q26.
New eMMC products will use 3D TLC NAND (48L and 96L), with Bit per Wafer roughly 3–9× the 19nm node.
Figure 8: eMMC bit output key assumptions
| 2026F | 2027F | |
|---|---|---|
| 19nm 32Gb MLC | ||
| Gross die per wafer | 1100–1150 | 1100–1150 |
| Yield (%) | 80 | 86 |
| Bit per Wafer | 28,416 | 30,650 |
| 48L 128Gb TLC | ||
| Gross die per wafer | 1400–1500 | 1400–1500 |
| Yield (%) | 70 | 75 |
| Bit per Wafer | 129,920 | 139,200 |
| 96L 128Gb TLC | ||
| Gross die per wafer | 2200–2400 | 2200–2400 |
| Yield (%) | 68 | 76 |
| Bit per Wafer | 200,192 | 223,744 |
Combined wafer-in expansion + Bit-per-Wafer uplift: eMMC bit output 87.7 bn Gb (2026F) and 281.9 bn Gb (2027F) — exponential growth, +2,590% / +222% YoY.
8. eMMC revenue scaling
In 2026–27 eMMC structural-shortage backdrop:
- ASP/Gb: +505% YoY (2026), +24.8% YoY (2027), reaching US$0.42 / US$0.53.
- Bit shipment: +994.5% (2026), +226.8% (2027), reaching 44.5 bn Gb / 145.6 bn Gb.
- eMMC revenue: NT$0.86 bn (2025) → NT$56.39 bn (2026) → NT$230.1 bn (2027) — that is, +6,407% YoY in 2026, +308% in 2027.
Figure 1: Macronix product mix (% revenue)
| 2025A | 2026F | 2027F | |
|---|---|---|---|
| eMMC | 3 | 50 | 75 |
| SLC NAND | 13 | 6 | 4 |
| NOR Flash | 61 | 35 | 17 |
| ROM | 17 | 8 | 5 |
| FBG | 7 | 5 | 3 |
Mix shift: NOR Flash falls from 61% (2025) to 17% (2027) of revenue while eMMC rises from 3% to 75%.
9. SLC NAND and NOR Flash — supportive but secondary
SLC NAND — supply concentration
Kioxia and Micron continue contracting SLC NAND supply. The market today is mainly Winbond + Macronix. Supply-demand tightening drives the segment into shortage and rising prices. KGI projects 2026 Macronix SLC NAND prices rise quarter-by-quarter, 1Q26–4Q26 = +15% / +45% / +30% / +20% QoQ. Combined with price effects, 2026–27 SLC NAND revenue +87.6% / +73.4% YoY to NT$6.81 bn / NT$11.8 bn.
NOR Flash — supply tight, AI Server demand kicker
Supply: thanks to logic-foundry margin advantages, China's three NOR foundries (XMC, Hua Hong, SMIC) have been adjusting capacity. SMIC has exited NOR foundry; Hua Hong has stopped 12-inch NOR foundry, retaining only 8-inch. Only XMC still provides NOR foundry services to China, so overall NOR supply tightens.
Demand: AI Server NOR Flash demand is rising rapidly, with single-system content multiplying vs PC. Combined with supply constraints, NOR Flash market faces undersupply and rising prices. KGI projects 2026 Macronix NOR Flash prices rise quarter-by-quarter, 1Q26–4Q26 = +30% / +35% / +20% / +10% QoQ.
Capacity: 8-inch NOR wafer-in expanding to ~32 kwpm in 1Q26, +62% YoY. NOR bit shipments +13.4% (2026) / +5.3% (2027). Revenue: 2026/27 NOR Flash revenue +120.2% / +34.8% YoY to NT$38.85 bn / NT$52.38 bn.
Figure 11: Macronix NOR Flash and SLC NAND quarterly contract price increases (% QoQ)
| 1Q26 | 2Q26F | 3Q26F | 4Q26F | |
|---|---|---|---|---|
| NOR Flash | 30 | 35 | 20 | 10 |
| SLC NAND Flash | 15 | 45 | 30 | 20 |
10. Initiation rating, target, and EPS
KGI initiates Outperform with NT$300 target = 10× 2026F EPS, corresponding to the upper end of the 2017–2021 P/E band.
EPS bridge
2026F EPS = NT$30.04 (+vs NT$1.77 loss in 2025). Drivers:
- 1H26: NOR Flash ASP/Gb +30%/+35% QoQ in 1Q/2Q26; SLC NAND ASP/Gb +15%/+45% QoQ.
- 2H26: MLC/TLC new wafer-in starts to convert to bit shipments — 3Q26 EPS jumps to NT$7.42, 4Q26 to NT$19.14. eMMC bit shipment +130% QoQ in 3Q26, +180% QoQ in 4Q26. eMMC ASP/Gb +40%/+20% QoQ in 3Q26/4Q26.
2026F: Revenue NT$112.4 bn (+289.1% YoY), GM 68.1%, EPS NT$30.04.
2027F EPS = NT$107.25 (+257% YoY). Drivers:
- eMMC ASP/Gb continues rising into 1H27.
- 3Q27 ~7k new wafer-in, transitioning to bit shipments by 4Q27 — 4Q27 bit shipment +60% QoQ.
2027F: Revenue NT$305.1 bn (+171.5% YoY), GM 82.3%, EPS NT$107.25.
Long-term thesis
KGI's central claim: as major NAND OEMs exit MLC and dismantle equipment, low-capacity eMMC supply contraction is structurally irreversible. KGI estimates 2026/27/28 MLC supply gaps of 36% / 47% / 26%. Even with Macronix's 2027 eMMC bit shipment 36× higher than 2025, its share of total eMMC demand is still only ~25%, so further bit-output growth is possible. With Automotive (typically 8+ year supply cycle), Drone, IPC, and Networking anchoring long-tail demand, KGI argues low-capacity eMMC long-tail demand will sustain Macronix's stable profits "for the foreseeable years."
TAM / market-share argument
On a bit-supply basis: 2026 DDR4 TAM ≈ 1.72× 2027 low-capacity eMMC TAM. In DDR4 (excl. Samsung/SK Hynix/Micron), Nanya ≈ 25% share, Winbond ≈ 13%; Nanya market cap ≈ NT$740 bn. KGI projects Macronix's 2027 low-capacity eMMC market share at ~51%. Comparatively, KGI's NT$300 target implies market cap ~NT$580 bn — argued as reasonable given TAM scale and share advantage.
11. Key financial summary
Income Statement (NT$ mn)
| Dec-23A | Dec-24A | Dec-25A | Dec-26F | Dec-27F | |
|---|---|---|---|---|---|
| Revenue | 27,624 | 25,883 | 28,880 | 112,386 | 305,111 |
| Gross profit | 6,761 | 6,099 | 5,131 | 76,575 | 251,095 |
| Operating profit | (2,407) | (3,924) | (3,698) | 65,530 | 233,947 |
| Net profit (after tax) | (1,700) | (3,214) | (3,308) | 56,137 | 200,415 |
| EPS (NT$) | (0.92) | (1.73) | (1.77) | 30.04 | 107.25 |
| EBITDA Margin | 6.6% | 3.6% | 4.8% | 62.9% | 79.1% |
| GM (%) | 24.5 | 23.6 | 17.8 | 68.1 | 82.3 |
| OP margin (%) | (8.7) | (15.2) | (12.8) | 58.3 | 76.7 |
Balance Sheet & Returns
| Dec-23A | Dec-24A | Dec-25A | Dec-26F | Dec-27F | |
|---|---|---|---|---|---|
| Total assets (NT$ mn) | 77,811 | 77,876 | 78,500 | 135,800 | 320,355 |
| Cash + ST investments | 11,906 | 11,668 | 14,920 | 34,957 | 184,420 |
| Inventory | 13,369 | 13,406 | 9,813 | 25,883 | 32,909 |
| Total debt | 20,231 | 23,576 | 20,289 | 19,996 | 19,702 |
| Equity | 48,326 | 43,980 | 45,808 | 102,803 | 286,257 |
| Net debt position | NT$7,558 mn | NT$13,876 mn | NT$9,317 mn | Net cash | Net cash |
| Pre-tax ROCE (%) | (4.1) | (6.5) | (7.0) | 64.4 | 164.3 |
Cash Flow
| Dec-23A | Dec-24A | Dec-25A | Dec-26F | Dec-27F | |
|---|---|---|---|---|---|
| Operating CF | (526) | (175) | 4,838 | 39,609 | 184,354 |
| Capex (investing) | (7,592) | (5,173) | (1,724) | (21,670) | (18,507) |
| Free cash flow | (4,947) | (4,598) | 2,730 | 17,607 | 164,441 |
Dividends & Per-share
- BPS: NT$24.05 → NT$23.72 → NT$24.51 → NT$55.01 → NT$153.18
- DPS forecast: NT$0.50 (2025) → NT$9.01 (2026F) → NT$32.17 (2027F)
12. KGI vs consensus — the bull-case discount
Figure 16: 2026/2027 KGI vs market consensus
2026F
| Metric | KGI | YoY | Consensus | KGI - cons (%) |
|---|---|---|---|---|
| Revenue (NT$ mn) | 112,386 | +289.1% | 49,932 | +125.1% |
| Gross profit | 76,575 | — | 19,416 | +294.4% |
| Operating profit | 65,530 | — | 10,049 | +552.1% |
| Pre-tax profit | 66,047 | — | 10,231 | +545.6% |
| Net profit | 56,137 | — | 9,233 | +508.0% |
| EPS (NT$) | 30.04 | — | 4.59 | +554.9% |
| GM (%) | 68.1 | +50.4 ppts | 38.9 | +29.3 ppts |
| OP margin (%) | 58.3 | +71.1 ppts | 20.1 | +38.2 ppts |
| Net margin (%) | 49.9 | +61.4 ppts | 18.5 | +31.5 ppts |
2027F
| Metric | KGI | YoY | Consensus | KGI - cons (%) |
|---|---|---|---|---|
| Revenue (NT$ mn) | 305,111 | +171.5% | 67,243 | +353.7% |
| Net profit | 200,415 | +257.0% | 17,790 | +1,026.6% |
| EPS (NT$) | 107.25 | — | 8.03 | +1,235.6% |
Observation: KGI's 2026 EPS is 6.5× consensus; 2027 EPS is 13.4× consensus. KGI is the outlier, not the consensus. The bull case rests entirely on (a) MLC supply gap unfolding as described, (b) eMMC ASP/Gb sustaining +505% / +24.8%, and (c) Macronix executing the wafer-in ramp on schedule.
13. 1Q26 / 2Q26 — KGI vs near-term consensus
Figure 15: 1Q26F / 2Q26F vs consensus
1Q26F
| KGI | QoQ (%) | YoY (%) | Consensus | Diff (%) | |
|---|---|---|---|---|---|
| Revenue | 9,694 | +25.4 | +58.0 | 7,729 | +25.4 |
| Gross profit | 3,232 | +73.0 | +197.4 | 2,448 | +32.0 |
| Net profit | 856 | — | — | 530 | +61.5 |
| EPS (NT$) | 0.46 | — | — | 0.27 | +67.7 |
2Q26F
| KGI | QoQ (%) | YoY (%) | Consensus | Diff (%) | |
|---|---|---|---|---|---|
| Revenue | 16,918 | +74.5 | +148.8 | 11,420 | +48.1 |
| Net profit | 5,640 | — | — | 1,751 | +222.1 |
| EPS (NT$) | 3.02 | — | — | 1.02 | +196.5 |
KGI sits well above sell-side consensus on all 1H26 numbers — the gap widens through 2H26 as the wafer-in ramps deliver bit shipment.
14. Key risks (KGI's framing — abbreviated)
KGI explicitly lists only one risk: delayed delivery of capacity-expansion equipment. (ASML Immersion DUV at 18-month lead time is the gating bottleneck for the 3Q27 step from 10k to 17k MLC/TLC wafer-in.)
[Wiki reviewer caveats — not in KGI:]
- Sell-side initiations with 203% upside warrant skeptical reading.
- KGI 2026 EPS NT$30.04 vs consensus NT$4.59 = 6.5× the street. Either KGI is way ahead of the curve or far over-extrapolating.
- eMMC ASP/Gb +505% YoY (2026) is heroic; embeds full pass-through of contract prices and no demand destruction.
- Bit-per-Wafer 3–9× 19nm assumes successful ramp of new 48L/96L 3D TLC NAND in 2Q–3Q26 — yield curves on new NAND nodes are notoriously bumpy.
- Macronix has lost money 2023–25 (cumulative ~NT$8.2 bn); market is being asked to underwrite a 56× turn in one year.
15. Source provenance
- Publisher: KGI Securities Investment Advisory (凱基證券投資顧問股份有限公司)
- Authors: Hanhsuan Shen (沈漢軒, hanhsuan.shen@kgi.com), Felix Pan (潘俊宏, Felix.Pan@kgi.com)
- Format: 18-page Mandarin equity research initiation
- Source URL on report: https://investment.kgisia.com.tw/Portal/Report/Index/Zh/R
- Source location in vault: ../../raw/equity-research/macronix-kgi-2026-03-12.pdf
- Disclaimer note: KGI is part of the KGI Financial Group; report contains standard sell-side disclaimers re investment banking relationships and conflicts of interest.
Cross-references in this vault
- Canonical ticker page: 2337
- Memory hierarchy primer: data-center-memory-types — covers SRAM/DRAM/HBM/NAND tiers; this report extends the picture into the eMMC/MLC sub-segment.
- Source page: source-illyquid — Collyer Bridge surfaced Macronix in their 17 April 2026 APAC wrap, which is what flagged it onto Pink's watchlist before this KGI report dropped.
- Watchlist: watchlist — 2337 row, source-attributed to illyquid/CB.
- Inv-q reading queue: inv-q — both the original CB article and now this KGI synthesis logged.