SK hynix still has half of high-bandwidth memory revenue. Samsung took a third in one quarter. Micron guided to another record. The Korean stocks closed as if the shortage were already the risk.
The last set of company numbers still describes a shortage. The stocks that sit on top of that shortage, at least the ones in Seoul, closed as if the market had already decided the shortage was the risk.
SK hynix fell 4.23% to 1,223,000 won, the lowest close since April. Samsung Electronics fell 2.98% to 211,500 won. The Kospi fell 3.73% to 5,552.74, after trading as low as 5,493.24. Foreigners sold a net 535.4 billion won. Institutions sold about 1 trillion. Individuals bought a net 900 billion. The index had already dropped about 21% across seven sessions from 6 October through the prior Thursday, risen for two days, and given the rise back. A Daishin Securities analyst quoted by Korea JoongAng Daily said concern over AI investment had not cleared, and that the prior session's gain was being sold.
The US close the night before was not that market. A broad American fund finished at $766.48, a fraction under its 50-day average. The 10-year yield was 5.14%. Semiconductors were the soft patch inside an index still near its highs. Korea was the place the memory trade actually broke. Prices below are the 19 October close in the United States and the 20 October close in Seoul. Korean dollar values use 1,327 won per dollar.
01 Who has the revenue
Counterpoint Research's quarterly tracker puts SK hynix at 50% of high-bandwidth memory revenue, Samsung at 33%, and Micron at 18%. A year earlier those shares were 64%, 15%, and 21%. In the first quarter of this year they were 58%, 21%, and 21%. Samsung's gain, and SK hynix's loss, happened in one quarter.
Most of that revenue is still HBM3E, the generation already in volume. Counterpoint said HBM4 shipments should become visible in the second half of the year, and that Samsung's share should rise further as those shipments ramp.
| Company | Q2 2025 | Q3 | Q4 | Q1 2026 | Q2 2026 |
|---|---|---|---|---|---|
| SK hynix | 64 | 56 | 57 | 58 | 50 |
| Samsung | 15 | 23 | 22 | 21 | 33 |
| Micron | 21 | 21 | 21 | 21 | 18 |
Ordinary DRAM is a wider market. In the same quarter Samsung had 38% of DRAM revenue, SK hynix 25%, Micron 24%, and China's CXMT 10%. Counterpoint said DRAM revenue rose 57% from the prior quarter and 385% from a year earlier. CXMT does not appear in the HBM table. High-bandwidth memory is still three companies. Commodity DRAM is no longer quite that closed.
02 An old map, kept on purpose
A July estimate put HBM revenue, not company revenue, on this path. The last column was that revenue as a share of each company's market value at the time. It is here because it is the table the page was built from. It is not a 2026 result.
| Vendor | 2024E | 2025E | 2026E | Share of value then |
|---|---|---|---|---|
| SK hynix | $6.5bn | $11bn | $14bn | 35–40% |
| Samsung | $4.5bn | $8bn | $12bn | 8–10% |
| Micron | $1.2bn | $2.5bn | $4.5bn | 10–15% |
The 2026 column adds to $30.5 billion. Earlier in the year, SK hynix's own newsroom quoted Bank of America putting the whole 2026 HBM market at $54.6 billion, up 58% on the year before. Those two pictures do not agree. One is a July estimate of three vendors. The other is a bank forecast, repeated by the company that would benefit if the higher number is right. Neither is an audited total. There is no public HBM-only dollar series in the pages used here that replaces the July lines. Use the table as the old map.
03 What the stocks have already done
| Name | Last | Day | Month | 3 months | Year | Market value |
|---|---|---|---|---|---|---|
| Micron | $904.90 | +2.97% | −10.9% | −6.8% | +338.1% | $1.024tn |
| SK hynix | 1,223,000 won | −4.23% | −27.6% | −41.2% | +152.6% | 891.404tn won |
| Samsung | 211,500 won | −2.98% | −14.9% | −24.3% | +117.3% | 1,367.504tn won |
The year column is the change from the close on 20 October 2025 to the latest close. Month and three-month columns are 21 and 63 sessions. Market values are the quote-page figures. At 1,327 won per dollar, 891.404 trillion won is $672 billion and 1,367.504 trillion won is $1.03 trillion.
Put the values next to the shares and the oddity is plain. Samsung and Micron are the same size in dollars. SK hynix, the HBM leader, is the smaller company after the drop, at about $672 billion. The July line that put its HBM revenue at 35% to 40% of market value cannot be carried forward. Fourteen billion dollars would be about 2% of today's value, not a third. Either the HBM dollars are now far above that July line, or the old percentage was never a clean revenue-to-value ratio. The pages in hand do not settle it.
All three are still up sharply from a year ago, because the run into the high was violent. A stock can be in a bear phase and still show a huge twelve-month gain. SK hynix is under both its 50-day and its 200-day average. So is Samsung. Micron, at $904.90, is under its 50-day and still above its 200-day. That is a correction inside an uptrend, not the same object as the Korean charts.
04 The chart does not agree
The future here is a share fight, not a question of whether anyone wants the chips.
Half of HBM revenue is still a lead. It is also eight points below the first quarter and fourteen points below a year ago. Samsung took the recent loss. In the spring, the company's own outlook quoted Goldman Sachs saying it would keep more than half the market through at least 2026, and quoted UBS putting it near 70% of the HBM4 slots on Nvidia's Rubin platform. Read that as what the company wanted believed in the spring. The summer share print already put more than half on the line. A 70% Rubin figure is a bank forecast inside a company article. It is not a shipment.
What can be said without that forecast is narrower. HBM3E is still the money. HBM4 is supposed to show up in the second half. SK hynix's job for the next two quarters is to stop Samsung's gain, not to discover a new market.
The company underneath the share loss is not a collapse. The quote page shows second-quarter revenue of 79.32 trillion won. The earnings line on the same page is 93.92 trillion won, which is above revenue, so that net figure includes something other than the quarter's operations. Trailing twelve-month revenue is 189.17 trillion won. Cash is listed at 87.98 trillion won. This is not a firm that has lost the cycle. It is a firm that has lost some of the mix, while the cycle is still paying everyone.
The stock has already voted. Down 27.6% in a month and 41.2% in three months, 59.1% under the 52-week high, under both moving averages. The next earnings date on the quote page is 27 October. If that print shows HBM share stable around half, and HBM4 actually shipping, the stock is a leader that has already been marked down. If share is lower and Samsung's HBM4 is the reason, the markdown was early, not wrong.
05 The catch-up is no longer a plan
In the second quarter Samsung took 33% of HBM revenue, from 21% in the first and 15% a year earlier. On ordinary DRAM it is already first, at 38%.
The preliminary third quarter, reported by Reuters on 8 October, is the company-level version of the same boom. Samsung estimated operating profit of 107.4 trillion won, about $80.17 billion, against an LSEG estimate of 106.1 trillion. Reuters called it a nearly nine-fold jump and the fourth straight record operating-profit quarter. Samsung said third-quarter revenue would rise 127% to 195 trillion won. The division split comes on 29 October.
Until that split, the HBM detail is other people's work. Douglas Kim of Douglas Research Advisory, cited by Reuters, estimated that Samsung's HBM bit shipments rose close to 50% from the prior quarter. TrendForce, citing the Korean press rather than a Samsung product note, said HBM4 sales began to contribute in the quarter and tied that to Nvidia's Vera Rubin accelerators. Treat the 50% as an analyst estimate. Treat Vera Rubin as a press report until the 29 October call says it.
The part Samsung cannot hide is the other half of the company. Memory is printing the profit. Reuters said the mobile business posted a larger than expected loss of more than $1 billion in the quarter, because the same shortage that fattens the memory division raises the cost of every handset. The foundry arm is still expected to lose money. Analysts Reuters spoke to expect fourth-quarter profit growth to slow to about 8% from the prior quarter, after about 20% in the third. TrendForce, via Reuters, sees conventional DRAM contract prices up 10% to 15% in the fourth quarter, after a second-quarter jump of about 60%. The shortage is not over. The rate of price increase is no longer the cartoon version.
Samsung and Micron have both told investors they expect the imbalance to last into 2028. Reuters put the risks next to that sentence: Chinese competition, and a slowdown in AI spending. CXMT is already 10% of DRAM. It is not in the HBM table. If it gets there, the three-company assumption breaks. If the buyers pause, the 2028 line was hope.
The stock did not celebrate an $80 billion quarter. On the morning of the guidance it slipped 0.2% while the Kospi fell 0.5%. By 20 October it was 43.5% under the 52-week high and under the 200-day average, with the year still up 117.3%. A Mirae Asset analyst told Reuters the market had moved on from whether earnings were up to whether the rise would last. The chart has answered that, for now, with a no.
A dollar value of about $1.03 trillion, level with Micron, is not a distress price for a company that just flagged $80 billion of quarterly operating profit. It is a refusal to capitalise a peak. Whether it is a peak is the remaining argument. The 29 October split is the first time investors will see HBM in Samsung's own divisions rather than in Counterpoint's table.
06 The profit, not the lead
Micron is the one of the three that can be bought in dollars without also buying the Kospi. That is a convenience. It is not the same thing as owning the HBM leader.
The fiscal fourth quarter, ended 3 September and reported on 30 September, came in at $54.23 billion of revenue, against $41.46 billion the quarter before and $11.32 billion a year earlier. Full-year revenue was $133.19 billion, against $37.38 billion. Non-GAAP earnings were $33.42 a share. The guide for the November quarter is revenue of $61.5 billion, plus or minus $1.5 billion, and non-GAAP earnings of $38.15, plus or minus a dollar. Non-GAAP gross margin was guided at about 86.25%. CNBC had the street at $57 billion and $35.40 before that guide.
Inside the quarter, cloud memory did $16.28 billion at an 83% gross margin. The core data-center unit did $18.00 billion at a 90% gross margin. CNBC put DRAM revenue at $39.8 billion, up 343% on the year and 73% of sales. This is no longer an HBM story with some DRAM attached. The whole memory book is earning a scarcity margin, because wafers that go into HBM are wafers that do not go into ordinary DRAM, and ordinary DRAM has run out.
That is why the share slip matters less to the income statement than it looks, and still matters to the franchise. Counterpoint has Micron at 18% of HBM revenue, down from 21% in each of the four prior quarters in the table. On the call, operating chief Manish Bhatia said that about a year ago HBM share had reached the company's broader DRAM share, and that the aim now is to stay around that DRAM share rather than to hit one number. DRAM share in the same Counterpoint quarter is 24%. On their own yardstick they are behind, while the company is throwing off cash. He also said the company does not have a line of sight to when supply and demand balance.
The product they will talk about as theirs, rather than as a share statistic, is a custom part. Technology chief Scott DeBoer said they have been working with Nvidia for more than a year on an HBM4E they call NV HBM, and he called it the first major custom HBM product in the market. Custom does not replace a 50% share. It does give Micron a reason to be in the next Nvidia box even if Samsung takes more of the standard stacks.
Bhatia said more than 75% of fiscal 2027 shipments are already committed, which has let the allocation talks move out to 2028. An analyst on the call referred to 26 strategic agreements and 35% of sales by 2030. Bhatia said those agreements cover both DRAM and NAND, that DRAM volume is a little under that, about 35%, and that NAND bit volume is a little over it. HBM prices for calendar 2027 were moved up to narrow the profit gap with ordinary DRAM. Ordinary DRAM has become so profitable that HBM has to catch up to it. The scarcity rent has spread. That is good for the next few quarters. It is also how memory cycles look late, when every grade of chip earns a stupid margin and customers sign long deals because they are afraid.
The factories will not fix this winter. Reported capital spending for the year was $27.37 billion. CNBC described a $250 billion plan for two HBM campuses, one broken ground in Clay, New York, in January, and a first new Boise fab scheduled to come online next year. Those are different clocks. A fab that opens next year does not change the shortage this quarter. A $250 billion campus figure, as CNBC stated it, is not the $27 billion already spent.
The stock is the least damaged of the three, which is not the same as healthy. $904.90, a market value of $1.024 trillion, 27.9% under the 52-week high of $1,255, under the 50-day, above the 200-day, up 338.1% from the close a year earlier. Trailing earnings of $74.33 put the multiple at 12.17 on the quote page. The quote page's average target is $1,560.57, with a low of $361 and a high of $3,000. That spread is a disagreement, not a forecast.
07 Three futures, one argument
SK hynix remains the volume leader and has the chart of a stock the market is done paying up for. Its future is whether 50% holds once HBM4 is a shipment rather than a spring promise. The date on the quote page is 27 October.
Samsung has the momentum in share, the largest ordinary-DRAM franchise, an $80 billion profit flag, and a handset division losing money because memory is too expensive. Its future is whether HBM4 turns a one-quarter jump into a durable second place, or into a challenge for first. The split is 29 October. The stock has already declined to capitalise the flag.
Micron has the earnings, the contracts, the US listing, and a custom Nvidia program. It does not have the HBM lead, and on the latest share print it lost a little more of it. Its future is to collect the shortage for as long as the wafers stay tight, and to use the Nvidia work so that a slip from 21% to 18% does not become a slip out of the platform. The new clean rooms are a later conversation. The multiple says the market is already in it.
The July table had SK hynix as the concentrated bet, Samsung as the cheap option, and Micron as the small US piece. The share table has rewritten the middle. Samsung is the share-gainer. Micron is a trillion-dollar shortage. SK hynix is still the leader, and it is the one the tape has punished hardest.
08 What would make this wrong
A pause in the buyers' spending would make the 2028 comments from Samsung and Micron look like the top. Reuters already lists that next to the shortage call. A rollover in ordinary DRAM prices, faster than the 10% to 15% fourth-quarter pace TrendForce is quoted as expecting, would do it sooner, because so much of the profit is no longer specific to HBM.
And the charts can be right even if the next two prints are strong. SK hynix and Samsung report into a Kospi that is under its 200-day average. A good quarter that does not reclaim that average is a good quarter inside a downtrend. Micron can keep guiding up and still fail the 50-day. The earnings are the case. The averages are the objection. The dates that can change either of them are 27 October, 29 October, and Micron's December quarter.
This is a reading of public figures. It is not a recommendation to buy or sell.
Half the revenue is still a lead. The Korean charts have already left.
Micron has the profit, not the volume. The next papers are 27 October and 29 October.
09 Sources
- Counterpoint Research, Global DRAM and HBM market share, Q2 2026 tracker. Share tables and the note that most HBM revenue is still HBM3E.
- Korea JoongAng Daily, 3 September 2026. The same Counterpoint shares, and the expectation that Samsung's share rises with HBM4.
- Korea JoongAng Daily, 20 October 2026. The Seoul close, the sellers, and the lowest SK hynix close since April.
- Micron, fiscal fourth-quarter and full-year results, 30 September 2026.
- CNBC, 30 September 2026. The street comparison, DRAM revenue, and the campus figure as CNBC stated it.
- Micron fiscal fourth-quarter call, Motley Fool transcript, 1 October 2026. Share target, the 75% commitment, the agreements, and NV HBM.
- Reuters, 8 October 2026. Samsung's preliminary third quarter, the mobile loss, and the 2028 comment.
- TrendForce News, 8 October 2026. The HBM4 contribution as reported in the Korean press, not as a Samsung filing.
- SK hynix newsroom, 2026 market outlook. The Bank of America $54.6 billion figure, and the Goldman and UBS lines as the company quoted them.
- Yahoo Finance, Micron quote. Close, market value, trailing earnings, multiple, and the target range.
- Yahoo Finance, SK hynix quote. Close, market value, second-quarter revenue and earnings, next earnings date.
- Yahoo Finance, Samsung Electronics quote. Close, market value, and the 52-week high.