Memory semiconductors (DRAM, HBM, NAND). SKHY (KRX 000660) / MU. Run 11 September 2026. Built from SK hynix's 424B4 prospectus (audited CY2023 to 2025, Q1 2026), Q2 2026 6-Ks and earnings call; Micron's FY2025 10-K, FY2026 Q1 to Q3 results releases and call transcripts; TrendForce, Counterpoint, IDC and Gartner share data. Events swept September 2025 to 11 September 2026; most recent events checked: SK hynix 6-K of 4 September 2026 (Solidigm pre-IPO raise still under review) and TrendForce's 2Q26 DRAM ranking of 7 September 2026; Micron leadership changes of 26 August 2026, with FQ4 results due 30 September 2026. Figures in US$; SK hynix won converted at annual average Fed noon buying rates printed in the 424B4 (1,306.8 / 1,363.4 / 1,421.4 per US$ for 2023 / 2024 / 2025). SK hynix reports calendar years; Micron's fiscal year ends in late August, about four months earlier. Not a valuation and not a recommendation.
Micron's next five years rest on three cells, not one.
New capacity feeds all three: Idaho ID1 and Taiwan Tongluo in mid-2027, then Singapore HBM packaging, Hiroshima and Idaho ID2 through 2028. Part of the cost is covered by up to $6.4bn of US CHIPS grants. SK hynix still owns the single most valuable cell, HBM, and it has matching capacity coming: M15X pulled forward and Yongin phase 1 cleanroom in early 2027 (Q2 2026 call). But that cell is being redistributed. Its HBM share went from 58% to 50% in Q2 2026, with Samsung rising from 21% to 33% (Counterpoint). SK hynix did grow faster over 2023 to 2025 (index 273 vs 241 below). That is evidence of the last cycle, not the reason for the call.
SK hynix turns revenue into margin better, and the reason is structural:
Here is the tension. In the latest overlapping quarter, Micron's margins are higher: gross 84.6% vs 83.2%, EBIT 80.4% vs 76.3%. There are two reasons. Micron's mix is heavier in commodity DRAM, which reprices every quarter, while HBM is priced once a year. And SK hynix pays 10% of operating profit into employee profit-sharing (424B4). So the growth winner and the margin winner are different companies. Micron is the faster repricer with the thinner engine; SK hynix is the fortress whose edge shows at the trough, not the peak.
Micron's exposure is the price cycle. About 60% of its revenue is conventional DRAM (inferred), and its cost engine is thinner: in FY2023 it posted a minus 9.1% gross margin and a minus 37.0% EBIT margin. The untested cushion is its customer agreements, which put price floors under about 20% of DRAM volume and a third of NAND.
SK hynix's exposure is concentration. One customer was 23.9% of 2025 revenue, and by inference that is an AI accelerator maker buying HBM. Its HBM share is sliding to Samsung. Its Wuxi DRAM and Dalian NAND fabs now need an annual US licence, after its Validated End-User status was revoked on 31 December 2025 (424B4).
In an ordinary downturn, Micron breaks first. In a disruption aimed at HBM or China, SK hynix does. Both face the June 2026 US class action alleging DRAM supply restriction.
These are the product x end-market x region cells that carry more than 80% of the combined economics. Every score is argued in the tabs below.
| Cell (product x region) | SK hynix | Micron | Why (one clause, sourced) |
|---|---|---|---|
| HBM x US AI accelerators | 4 | 3 | SK hynix is dominant at 50% (Q2 2026) but ceding share to Samsung, so it scores 4 rather than 5. Micron holds 18% to 21% and targets a share close to its DRAM share (Counterpoint; June 2026 call). |
| Conventional DRAM x global, US-booked majority | 3 | 4 | Micron is gaining share with a server-weighted mix. SK hynix is holding bits back for HBM, and its DRAM share fell from 28.8% to 24.9% (TrendForce 1Q to 2Q26). |
| NAND and eSSD x global cloud and OEM | 4 | 3 | SK hynix is number 2 in NAND at 18.5%, and its eSSD revenue doubled quarter on quarter (IDC via 424B4; Q2 2026 call). Micron is number 3 after a fast 2Q26 gain (TrendForce). |
| Mobile and PC x China and Asia OEMs | 3 | 2 | SK hynix books 19.7% of revenue in China and runs fabs there. Micron's China and Hong Kong share fell to 10.1% after the 2023 CAC ban and its exit from China server DRAM (10-K; Reuters via TrendForce). |
| Auto and embedded x global | ND | 3 | Micron's auto and embedded unit made $4.75bn at a 12% operating margin, with 3-year customer agreements (10-K). SK hynix discloses no figure. |
Each score is anchored to an exhibit in the three tabs; no score ships without one. Scores are per cell and are not summed. The three lenses get their verdicts in the Three Answers above, not here. ND means not disclosed, not zero.
Normalization. Both companies report DRAM, NAND and other, and both bury HBM inside DRAM without a separate revenue line. Micron adds end-market business units (hyperscale cloud, OEM data center, mobile and client, auto and embedded); SK hynix discloses no end-market split. The regional bases differ too. SK hynix books revenue by the location of its selling entity, Micron by customer headquarters. Both put about two-thirds in the US: 68.8% vs 64.5%. Neither basis shows where the chips are actually used, and the two cannot be reconciled below that coarse level.
How the HBM cells were sized. They are inferred: each company's HBM revenue share (Counterpoint) multiplied by Gartner's estimate of a more than $30bn HBM market in 2025. The results cross-check. SK hynix's largest customer was $16.4bn, 23.9% of revenue, against about $17.5bn inferred. Micron's 17% customer, "primarily" in its cloud memory unit, was $6.4bn, against about $6.3bn inferred.
| Cell | SK hynix rev (% of total) | Growth | Micron rev (% of total) | Growth | Margin signal / leader |
|---|---|---|---|---|---|
| Conventional DRAM x global | ~$35.2bn (51%) (inferred) | DRAM +82% CAGR | ~$22.3bn (60%) (inferred) | DRAM +61% CAGR | Micron is gaining: its DRAM revenue grew 65.5% quarter on quarter in 2Q26, against 37.9% for SK hynix (TrendForce) |
| HBM x US AI accelerators | ~$17.5bn (26%) (inferred) | more than 2x in 2025 | ~$6.3bn (17%) (inferred) | new since FY2024 | SK hynix: 56.4% share (IDC, Q1 2026); premium of more than 5x per GB |
| NAND and eSSD x global | $14.6bn (21%) | +40% CAGR | $8.5bn (23%) | +42% CAGR | SK hynix leads in high-capacity eSSD through Solidigm, which it acquired from Intel for $8.8bn |
| Auto and embedded x global | ND | ND | $4.75bn (13%) | +7% CAGR | Micron only; 12% operating margin in FY2025 |
| All other | $1.1bn (2%) | $0.3bn (1%) | Foundry and other |
| Region | SK hynix 2025 (selling entity) | Micron FY2025 (customer HQ) |
|---|---|---|
| United States | $47.1bn, 68.8% | $24.1bn, 64.5% |
| China (Micron includes Hong Kong) | $13.5bn, 19.7% | $3.8bn, 10.1% |
| Other Asia (Korea, Taiwan, Japan, rest) | $6.4bn, 9.4% | $8.5bn, 22.7% |
| Europe and other | $1.4bn, 2.0% | $1.0bn, 2.7% |
HBM. Growth here comes from accelerator volume, stack height and the premium each new generation commands.
Conventional DRAM. Growth in 2026 has been almost all price. SK hynix's DRAM prices rose by about 65% in Q1 2026 and about 30% in Q2, on flat and then high-single-digit bit growth. Micron's rose by about 65% and then about 60%, on low-to-mid single-digit bit growth (Q2 2026 call; Micron via secondary transcripts, approx., unverified). TrendForce says SK hynix's price growth was "relatively limited" because HBM makes up the largest share of its bits, while Micron "prioritized higher-priced server DRAM".
NAND. Part of SK hynix's NAND growth was bought: Solidigm came from Intel for $8.8bn, paid in 2021 and 2025. Micron's is all organic.
Both companies sell direct to hyperscalers, OEMs and accelerator makers, so the channel is not the difference; the form of the contract is.
SK hynix: its prospectus says quantities and prices are "typically determined through mutual agreement at the time of purchase" (424B4). It has now signed long-term agreements with about 10 customers, typically around five years, with deposits and pricing mechanisms "designed to address price volatility". It declines to say how much volume they cover (Q2 2026 call).
Micron has disclosed 16 agreements with the following terms (June 2026 call, approx., unverified):
| SK hynix | Micron | |
|---|---|---|
| Wafer fabs | Icheon and Cheongju in Korea; Wuxi (DRAM) and Dalian (NAND) in China (424B4) | Taiwan (the majority of 2025 DRAM output), Japan, Singapore, US (10-K) |
| China equipment access | Validated End-User status revoked 31 December 2025; annual US licence granted for 2026 | No China wafer fab; Xi'an does back-end only |
| State co-funding | Not in sources | US CHIPS grants of up to $6.4bn; Japanese and Indian subsidies (approx., unverified) |
| Behaviour in the 2023 shock | Gross margin minus 1.6%; capex cut by more than 50% | Gross margin minus 9.1% |
Over the last three quarters SK hynix has been the share donor in DRAM and HBM. Micron has taken share in conventional DRAM and NAND and given up a little in HBM. The share SK hynix lost in HBM went to Samsung, not Micron.
HBM stacking know-how (TSV and MR-MUF). It was first to mass-produce HBM3 and HBM3E, and its HBM4 yields are near HBM3E levels (424B4; Q2 2026 call). The proof is a 56.4% share (IDC, Q1 2026). Durability: medium, because Samsung gained 12 points in one quarter.
DRAM scale and node cost. Its 2025 DRAM revenue was $52.7bn against Micron's $28.6bn, and 1c-nm DDR5 is in full production. Durability: medium to high, and it shows in the gross margin.
US-anchored, state-funded capacity: up to $6.4bn in CHIPS grants and a plan to make 40% of its DRAM in the US (Micron release, 12 June 2025). Durability: medium to high, as a policy position.
Contract architecture: 16 take-or-pay agreements carrying about $100bn (approx., unverified). Durability: medium, since they have not yet been tested in a downturn.
Auto and embedded franchise ($4.75bn). Durability: high, but small.
SK hynix is the more concentrated of the two:
Micron is less concentrated but broadly exposed:
Both raised prices by 30% to 85% quarter on quarter in the first half of 2026. HBM is priced once a year at both. SK hynix says conventional DRAM strength "may also have some influence" on its 2027 HBM talks (Q2 2026 call).
Both depend on ASML EUV scanners. SK hynix is buying ₩11.9tn of them for delivery by December 2027 (424B4), and Micron's 1-gamma is its first EUV node (10-K). Both also rely on US equipment and Japanese materials.
The asymmetry is political versus geographic. SK hynix's China fabs depend on an annual US licence. Micron's DRAM output is concentrated in Taiwan. On passing through a shock, the 2023 downturn is the clearest test: SK hynix's gross margin fell to minus 1.6%, Micron's to minus 9.1%.
| Cell | Route control | Pocket price | Continuity | Outcome (share / margin) | Confirming KPI |
|---|---|---|---|---|---|
| HBM x US AI accelerators | Both sell direct on annual contracts. SK hynix reportedly holds about two-thirds of NVIDIA HBM4. | SK hynix at or above Micron (inferred) | SK hynix has the yield record; Micron is ramping faster | SK hynix: share down, margin holds. Micron: share flat, margin up. | Counterpoint HBM share |
| Conventional DRAM x global | Both direct. Micron has floors on about 20% of volume. | Micron above SK hynix in 2Q26, on server mix | Both constrained by cleanroom space | Micron: share and margin up. SK hynix: share down. | TrendForce DRAM share |
| NAND and eSSD x cloud | SK hynix through Solidigm; Micron direct | ND | SK hynix ramping 321-layer; Micron on G9 | Both up; SK hynix leads eSSD | TrendForce NAND ranking |
The tension: Micron is growing fastest in conventional DRAM, the cell with the least structural pricing power. SK hynix holds its power in HBM, the cell where its share is now falling.
1. The HBM process and yield lead (50% to 58% share vs 18% to 21%). Impact: major. Micron has chosen not to chase it. Closing it would take at least two product generations plus packaging capacity that arrives in 2027, so money alone cannot close it in 18 months.
2. DRAM scale (1.8x in 2025). Impact: moderate, and already closing. SK hynix's and Micron's 2Q26 DRAM revenue was $38.6bn vs $36.0bn (TrendForce), because Micron repriced faster.
3. Contract disclosure. Impact: moderate but untested. SK hynix could close this within a year simply by disclosing its terms, if they match Micron's.
Basis. Both companies present costs by function, so no by-nature reconstruction was needed. SK hynix's operating profit is gross profit minus S&A minus R&D. That matches its reported 48.6% operating margin for 2025. Micron's operating income includes restructuring and other operating items: $295m in FY2023 and $100m in FY2025. These were not stripped out, because SK hynix has no matching itemised line. Two further differences:
The windows are CY2023 to 2025 for SK hynix and FY2023 to 2025 for Micron, offset by about four months.
| % of sales, 3y avg | SK hynix | Micron | Gap | What drives it |
|---|---|---|---|---|
| COGS | 64.4 | 82.3 | -17.9 | HBM premium and DRAM scale |
| R&D | 8.3 | 14.6 | -6.3 | Scale: SK hynix spends more in dollars on a bigger base |
| SG&A | 7.2 | 4.5 | +2.7 | SK hynix's S&A carries profit-share accruals and commissions |
| Gross margin | 35.6 | 17.7 | +17.9 | Present in all three years: +7.5, +25.7 and +20.6 points |
| EBIT margin | 20.2 | -1.9 | +22.1 | Gross margin gap, partly offset by S&A |
| Year | SKHY COGS | R&D | S&A | GM | EBIT | MU COGS | R&D | SG&A | GM | EBIT |
|---|---|---|---|---|---|---|---|---|---|---|
| 2023 / FY23 | 101.6 | 11.4 | 10.5 | -1.6 | -23.6 | 109.1 | 20.0 | 5.9 | -9.1 | -37.0 |
| 2024 / FY24 | 51.9 | 6.7 | 5.9 | 48.1 | 35.5 | 77.6 | 13.7 | 4.5 | 22.4 | 5.2 |
| 2025 / FY25 | 39.6 | 6.7 | 5.2 | 60.4 | 48.6 | 60.2 | 10.2 | 3.2 | 39.8 | 26.1 |
| Latest quarter | 16.8 | 4.2 | 2.6 | 83.2 | 76.3 | 15.4 | 3.2 | 1.0 | 84.6 | 80.4 |
Latest quarter: SK hynix Q2 2026, April to June (6-K of 18 August 2026, with R&D split out from note 22), against Micron FQ3 2026, March to May (8-K of 24 June 2026). Micron guides to a gross margin of about 86% for FQ4. Capex as a share of revenue over the three years: SK hynix 25.4%, 24.1% and 28.3%; Micron (net) 45.1%, 32.3% and 36.9%.
Gross margin is the persistent gap. SK hynix led by 7.5 points in 2023, 25.7 in 2024 and 20.6 in 2025. The mechanism has three parts:
This confirms the power map: the company with the HBM moat shows it in gross margin, and the 2023 trough shows the advantage survives a downturn.
The latest quarter points the other way. Micron's margin is ahead by 1.4 points because its commodity-heavier mix reprices faster than SK hynix's annually priced HBM, and SK hynix pays profit-share. So the evidence reads as a through-cycle cost edge for SK hynix, not a peak-price edge. That lowers confidence in the verdict without reversing it.
| Metric | Threshold | By when | If it hits, it favors | Where published |
|---|---|---|---|---|
| HBM revenue share | SK hynix at or above 50% (below 40% flips) | 2Q27 data, about September 2027 | SK hynix (Micron if below 40%) | Counterpoint and TrendForce, quarterly |
| DRAM revenue share | Micron at or above SK hynix in any quarter | Through 2Q27 | Micron: confirms the growth answer | TrendForce DRAM ranking |
| Gross margin gap in the first quarter of falling DRAM prices | SK hynix lead of 10 points or more (5 or less flips) | First down-quarter, likely 2027 | SK hynix (Micron if 5 or less) | SK hynix 6-K and Micron 8-K |
| Disclosure of long-term agreement coverage | SK hynix quantifies volume covered and deposits (Micron: about 20% of DRAM, $22bn) | FY2026 20-F, about April 2027 | SK hynix if disclosed and comparable; otherwise Micron | SK hynix 20-F; Micron FY2026 10-K |
| US licence for SK hynix's China fabs | 2027 licence granted | 31 December 2026 | SK hynix (Micron if delayed) | SK hynix 6-K |