SK hynix, based in South Korea, makes about half of the world’s HBM, the stacked memory that sits beside every Nvidia AI chip. Its operating margin reached 76% last quarter. Since July its shares have also traded in New York as ADRs (ten ADRs equal one Korean share). Figures are in won, converted at about ₩1,327 per dollar.
Today’s close against the report’s numbers, updated every weekday. The analysis is as of October 22, 2026; the rating as of October 22, 2026. Both change only when the report is revised.
Per ADR, in dollars: the cycle model’s bear-to-bull range, the probability-weighted value, the price, its range since the July listing and the analysts’ consensus target. The copper line marks the base case; the fine dotted line is today’s close.
Written on October 22, 2026, ahead of its third-quarter results on October 29 in Seoul. Figures are from the company’s filings and releases unless noted. Not investment advice. The author holds no SK hynix shares or ADRs.
SK hynix makes DRAM, the working memory in every computer, and NAND flash for storage. Its edge is HBM: high-bandwidth memory made by stacking DRAM chips, which Nvidia and other AI chipmakers place next to each accelerator. SK hynix was first to supply HBM at scale to Nvidia and held about 50% of the market in the second quarter, ahead of Samsung (33%) and Micron (18%). It began mass shipments of the newest generation, HBM4, for Nvidia’s Vera Rubin platform this year.
It is the same industry as Micron, with the same cycle: three companies make nearly all the world’s DRAM, and prices rise and fall with the balance between their new factories and demand.
Second-quarter revenue rose 257% from a year earlier to ₩79.3 trillion and operating profit 557% to ₩60.5 trillion, a 76% margin. Net profit of ₩93.9 trillion was even higher than operating profit because of large gains outside its operations. HBM4 mass shipments began in the quarter and ramp through the second half; the company has signed long-term agreements of about five years, with deposits and volume commitments, with about ten customers.
Korean analysts expect third-quarter operating profit of about ₩77 trillion, a sixth straight record, as DRAM prices rose about 22% and NAND about 15% in the quarter. Results come on October 29 in Seoul (the evening of October 28 in New York). On October 22 Bernstein estimated that Samsung overtook SK hynix in HBM revenue in the third quarter, $8.7 billion to $7.6 billion, and the ADRs fell about 8%.
| ₩ trillion | 2023 | 2024 | 2025 | H1 2026 |
|---|---|---|---|---|
| Revenue | 32.8 | 66.2 | 97.1 | 131.9 |
| Operating profit | −7.7 | 23.5 | 47.2 | 98.2 |
| Operating margin | −24% | 35% | 49% | 74% |
| Net profit | −9.1 | 19.8 | 42.9 | 134.3 |
| Revenue in US$, at ₩1,327 | $25B | $50B | $73B | $99B |
SK hynix reports in won under Korean accounting standards, so the site’s automatic SEC tables do not cover it; these figures come from its results releases. The swing from a ₩7.7 trillion operating loss in 2023 to ₩98 trillion of operating profit in the first half of 2026 is the memory cycle in one row. First-half 2026 net profit includes large non-operating gains. Dollar figures use today’s exchange rate, so earlier years are shown at a stronger won than they were.
At the end of June SK hynix had ₩88 trillion of cash against ₩18.6 trillion of debt. In July it listed ADRs on the Nasdaq under SKHY, issuing up to 17.79 million new shares (about 2.4% more) for up to about $29 billion; this model adds that to net cash, about $80 billion in all. Capital spending will reach the high ₩40 trillions this year (about $35 billion). New capacity comes from the M15X fab, pulled forward, and the first Yongin cleanroom in early 2027.
Management says it does not expect its investment plans to cause oversupply soon and that tight conditions should last “for a considerable period”. It is weighing more shareholder returns, with analysts expecting buybacks of more than ₩30 trillion.
The same method as Micron: free cash flow for 2027–2030 at each scenario’s prices, then an exit value on mid-cycle earnings, all discounted at 14% (a high beta plus Korea and memory-cycle risk), plus net cash, over about 7.4 billion ADR-equivalents (an estimate of 740 million Korean shares outstanding after the July offering, net of treasury shares).
| Bear | Base | Bull | |
|---|---|---|---|
| Revenue, 2027 / 2028 | $280B / $230B | $330B / $350B | $380B / $450B |
| Revenue, 2029 / 2030 | $150B / $120B | $240B / $180B | $400B / $350B |
| Free cash flow, four years | $228B | $417B | $763B |
| Mid-cycle revenue and operating margin | $150B at 25% | $210B at 35% | $320B at 45% |
| Exit multiple | 10× | 12× | 14× |
| Value per ADR | $60 | $111 | $215 |
| Probability | 25% | 50% | 25% |
Weighted 25/50/25: $124. This year’s revenue should be about ₩340 trillion ($256 billion), an estimate from the first two quarters and analysts’ third- and fourth-quarter forecasts; the base case grows it another 30% in 2027 as HBM4 ramps, then has prices fall as new supply arrives from 2028, to a mid-cycle margin still far above anything before AI.
Every figure in this valuation comes from one model script, rerun and checked against this page each time the site is built. The XLSX model rebuilds it in live formulas: change any blue input and every year’s cash flow, the bridge to the value, the scenarios and the sensitivity table recalculate.
| To justify the price, you need | Against |
|---|---|
| About $80 billion of after-tax profit a year from 2031, valued at 12×, after the base case’s cash flows | 2025 net profit was ₩42.9T, about $32B |
| Or the bull case: revenue staying above $350 billion through 2030 | New supply from all three makers arrives from 2027–2028 |
The price sits just above our weighted value. Analysts are far more optimistic, with a consensus target of about $253, and the shares trade at about 8 times trailing earnings. As with Micron, that multiple looks cheap only if peak earnings are close to normal.
| Risk | Why it matters |
|---|---|
| The cycle | Every past memory boom ended when new fabs came online; Samsung, Micron and SK hynix are all adding capacity. |
| HBM share | Samsung is gaining ground and may have overtaken SK hynix in HBM revenue in the third quarter. |
| Customer concentration | Nvidia is the largest HBM buyer by far. |
| China | CXMT, China’s DRAM maker, is closing the technology gap at the low end. |
| Currency | A 10% stronger won cuts pre-tax profit by about ₩1.2 trillion and changes the dollar value of everything here. |
Upside: HBM4 prices up sharply through 2027, long-term agreements that smooth the next downturn, large buybacks, and a cash pile that could approach $100 billion.
SK hynix was first to HBM at scale and has the largest share of Nvidia’s supply at a 76% margin. It is still a memory company at the top of its cycle, Samsung is gaining in HBM, and the won adds currency risk.
What would change the value: Third-quarter results on October 29 in Seoul, HBM4 pricing for 2027 and any buyback announcement are the next inputs.
Ranges, not three cases. The expected value is the average of 4,000 runs of this report’s own model, with every input drawn from a range: the base case is the middle, and the bear and bull cases are the 10th and 90th percentiles. Business inputs move together; the discount rate moves on its own. The 25/50/25 weighting in the valuation section gives $124. How the ranges work
Two different questions. The rating is relative: it asks whether the stock is cheaper or dearer than the S&P 500 on the same assumptions. The absolute view asks whether it is worth its price at all: on its own the stock trades within 10% of our estimated value.
On the same assumptions every report uses (a 5.15% Treasury yield and a 4.5% equity premium), the S&P 500 itself is worth about 68% of its price. The rating compares each stock with that: Buy if it is at least 15% cheaper than the market on this basis, Sell if at least 15% dearer, Hold in between. It is the same rule for every company, whoever owns it. The rating is the same whether the index is assumed to grow 6% or 10% a year. How ratings work
Against the Street: Analysts’ consensus of about $253 rests on today’s earnings at about 8 times. This model assumes memory prices fall back after 2028, as in every past cycle.
The cycle still decides; the price leaves less to go wrong than Micron’s.
| What | Figures | Source |
|---|---|---|
| Latest results | ₩79.3T (Revenue, Q2 2026, +257% (about $60B)), 76% (Operating margin), ₩60.5T (Operating profit (about $46B)), ₩69.4T (Net cash (about $52B)) and others | Q2 2026 results; Earnings call; Q3 expectations |
| Price at writing | $131.69 (ADR close, Oct 22, 2026) | Yahoo Finance; the time of the quote is stated beside the price |
| Discount rate inputs | 10-year Treasury 5.15%, equity risk premium 4.5% | Fixed for every report so they can be compared; How ratings work |
| Valuation | Every value, case and sensitivity | This site’s model: XLSX, CSV |
| Checked by hand | Not yet | Reports are added to the audit log as they are revised |
| Input | Bear (10th percentile) | Base (median) | Bull (90th percentile) |
|---|---|---|---|
| Revenue path | $280B/$230B/$150B/$120B | $330B/$350B/$240B/$180B | $380B/$450B/$400B/$350B |
| Free-cash-flow margin path | 45.0%/35.0%/10.0%/5.0% | 50.0%/45.0%/28.0%/15.0% | 55.0%/52.0%/45.0%/40.0% |
| Normal-year revenue | $150B | $210B | $320B |
| Normal-year margin | 25.0% | 35.0% | 45.0% |
| Exit multiple | 10× | 12× | 14× |
| Simulation result | |
|---|---|
| Expected value (average of 4,000 runs) | $122.82 |
| Likely range, 10th to 90th percentile | $67.31 to $190.46 |
| Median | $112.58 |
| Chance the value is above the price | 36% |
| Chance it is cheaper than the S&P 500 | 72% |
Each input is drawn from a split normal: its bear and bull values are the 10th and 90th percentiles, so the two sides can differ. Business inputs share a common factor (correlation 0.6); the discount rate is the same in every case, so it is not drawn. Inputs not listed are held at their base value. How the simulation works.