The Memory Big Three in August: The Businesses Got Better, So Why Did the Stocks Split?

Connecting context: In August we covered the rebound flow analysis and the fourteen Korean valuations. This post lays those two over the three memory companies. To keep it easy to read, each term is explained the first time it appears.

TL;DR

• Over the course of August, the business side of the story got better. Demand, contracts, technology, and shareholder returns all moved forward.
• But the stocks split three ways. From July 31 to August 28, Samsung Electronics fell 2.1% and SK Hynix fell 3.8%, while Micron rose 13.3%.
• First, one number needs fixing. There’s a story going around that Micron rose 47% in August to reach $1,211. That’s not true. The $1,211 area was the one-year high set on June 25, and Micron’s closing price on August 28 was $932.86. Even measured from the start of the year, the real gain is a little over three times, not 284%.
• The reason the two Korean companies couldn’t rise despite good news shows up clearly in how their payouts were structured. SK Hynix bought back 40 trillion won of stock and decided to cancel all of it, and the stock rose 12.7% the day after the announcement. Samsung Electronics’ 15 trillion won buyback, by contrast, is money earmarked for employee bonuses, so the share count doesn’t shrink, and the stock fell 8.7% on the first trading day after the announcement.
• The China variable grew too. CXMT (ChangXin Memory) rose 466% on its first trading day in late July, and its market cap touched the 800 trillion won range intraday. But its index inclusion is in a secondary index, not a major one.
• To see whether today’s prices are actually cheap, you need to check whether they would hold up if earnings came down. Even if you cut 2027 expected earnings by 30%, Samsung Electronics still has 26.8% of room left and SK Hynix has 10.8%. Micron falls short by 1.4%, which is close to a wash.
• There’s more to check starting in September. Korea’s export statistics, Micron’s earnings report, and next January, when Samsung Electronics will decide the size of its remaining payout, make up the timeline for this story.

1. How the three stocks moved in August

Let’s start with a chart. It sets the July 31 closing price at zero for each stock and overlays how they moved through August 28.

All three fell together in early August. By August 10, Samsung Electronics was down as much as 12.4% and SK Hynix as much as 17.3%. Up to this point, they moved together.

The split came after that. Micron climbed as high as 22.9% in mid-August before finishing the month up 13.3%. The two Korean stocks bounced sharply on August 20 and 21, but they were pushed back down in the final week and ended the month in negative territory.

Here are the month’s results on their own. Samsung Electronics fell 2.10%, from 262,500 won to 257,000 won. SK Hynix fell 3.78%, from 1,718,000 won to 1,653,000 won. Micron rose 13.34%, from $823.03 to $932.86.

2. First, let’s correct one number

There’s a story circulating widely about Micron: that it rose 47% in August to reach $1,211, and that it’s up 284% since the start of the year. We checked, and it isn’t true.

The $1,213 area is actually the one-year high recorded on June 25. Micron didn’t come anywhere close to that level in August. Its highest point that month was the $1,011 area, on August 17, and it closed August 28 at $932.86. Today’s price is actually 23% below the June high.

The 284% year-to-date figure also looks like a calculation error. Dividing the June high by the early-January price gives about 285%, and it looks like that figure got carried over and mislabeled as the current return. Measured from last year’s final closing price, Micron is up about 227% through August 28. That’s a big gain, but it isn’t 284%.

Mistakes like this are common. When you put the high and the current price in the same table, it’s easy to mix up the rows. But this one number changes the conclusion. If Micron were up 47%, it would clearly look expensive. If it’s up 13%, the story becomes much more cautious.

3. The business news was good

Setting the stock price aside, the facts confirmed in August all point in one direction.

Demand got bigger. NVIDIA reported quarterly revenue of $96.2 billion on August 26. That’s up 106% from a year earlier, and $89 billion of it came from data centers. Its guidance for next quarter is $108 billion.

There was also evidence that this demand is giving memory makers real negotiating power. The amount NVIDIA has committed to buy in the future jumped from $119 billion to $279 billion in a single quarter. The company also said that memory prices could push its gross margin, the share of revenue left after production costs, down to a range of 71% to 72% in the fourth quarter of next fiscal year. In plain terms, the buyer is giving up margin just to lock in memory supply.

Demand shifted from one-off orders to signed contracts. A quick word on terminology here: a long-term supply agreement locks in volume and a price floor for several years at once, so profit moves around less than it would if everything sold at the spot price of the moment. SK Hynix has officially disclosed that it signed this kind of agreement with 11 customers. Micron has landed 16 strategic customer agreements, and 14 of them carry a combined minimum revenue of about $100 billion. Customer prepayments and financial commitments add up to another $22 billion.

Technology also moved from a plan into revenue. SK Hynix began mass-production shipments of HBM4, the next generation of high-bandwidth memory, in the second quarter. Samsung Electronics said it is increasing HBM4 sales and has shipped samples of the generation after that. There have also been reports that Samsung’s HBM4 yield, the share of chips produced that are good enough to sell, is around 80%, but that figure comes from industry reporting, not an official company disclosure.

The earnings themselves were good too. SK Hynix posted second-quarter revenue of 79.3 trillion won and operating profit of 60.5 trillion won, for an operating margin of 76%. That means for every 100 won of sales, 76 won was left over as profit. Micron, for its part, posted revenue of $41.46 billion and adjusted earnings per share of $25.11 last quarter, and it’s guiding for $50 billion in revenue and an 86% gross margin next quarter.

4. So why didn’t the two Korean stocks rise?

Both companies announced large shareholder returns in August. But the market’s reaction was the opposite in each case. This is the part of August with the most to learn from.

SK Hynix’s board decided on August 19 to buy back 40 trillion won of its own stock and cancel all of it. That comes to 24.07 million shares, about 3.3% of shares outstanding, the largest cancellation ever by a Korean listed company.

It’s worth pausing on what cancellation actually means. When a company buys its own shares and retires them for good, the number of shares left in the market goes down. The same profit then gets split among fewer shares, so earnings per share go up. The stock actually fell on the announcement day itself, since regular trading closed before the disclosure came out, but it rose 12.73% on the next trading day, August 20.

Samsung Electronics’s board said on August 21 that it would return 90 to 110 trillion won to shareholders this year. In sheer size, that’s bigger than SK Hynix’s plan. Yet the stock fell 8.70% on the next trading day, August 24.

The reason lies in what the payout is made of. The 15 trillion won buyback in the announcement is earmarked to pay employee bonuses. The company does buy the shares, but instead of retiring them, it hands them to employees, so the number of shares in the market doesn’t shrink. That limits how much it can lift earnings per share. What happens to the remaining 60 to 80 trillion won will be decided next January. One brokerage estimated that a large-scale cancellation would be difficult because of ownership limits under the Act on the Structural Improvement of the Financial Industry, and that the actual cancellation will likely land around 10 to 20 trillion won. The company itself has not confirmed a number yet.

The market paid for the kind of payout, not the size of it. It rewarded the plan that actually shrinks the share count with a double-digit rally, and it punished the plan that’s a buyback in name only with a decline.

One caveat is worth adding. On August 20, the day SK Hynix rose 12.73%, Samsung Electronics also rose 9.49%. Since both stocks rose together that day, it’s hard to credit the cancellation announcement alone. The difference in kind shows up more clearly on August 24. That day, while Samsung Electronics fell 8.70%, SK Hynix fell only 3.41%.

5. A new variable: China

China’s CXMT went public on July 27. It rose 466% on its first day, and its market cap touched the 800 trillion won range intraday. On a closing basis, it’s in the 700 trillion won range. The listing raised around 12 trillion won.

There was also news of index inclusion in August. But that comes with a catch. What it joined is a broad, secondary index, not a major one, and inclusion in a major index isn’t expected until around 2027.

CXMT isn’t at the point of threatening the two Korean companies’ high-bandwidth memory business right away. A technology gap remains. But the stock market works a little differently. The mere fact that a large, state-backed competitor now exists keeps alive the doubt over how long today’s high profits can last. And now that it’s part of an index, some of the money invested in Asian semiconductors gets split off toward it.

6. What does “cheap” really mean?

Memory stocks show a very low price-to-earnings multiple, the share price divided by earnings per share, where a lower number is usually read as cheaper. Using 2027 expected earnings, Samsung Electronics comes out at 3.9 times and SK Hynix at 3.8 times. That’s strikingly low next to the 10 to 20 times that’s common across the broader market.

But it’s risky to take that number at face value. Memory is an industry where earnings swing enormously between good times and bad. Right now is a good stretch, so the earnings in the denominator are inflated. When earnings are at a peak, the multiple automatically looks low.

So here’s a different approach: deliberately cut expected earnings, then recalculate from there. It’s a way of checking whether today’s price would hold up even if earnings came down.

Leave earnings untouched, and all three stocks have plenty of room to run. But cut 2027 expected earnings by 30%, and Samsung Electronics still has 26.8% of room left, SK Hynix has 10.8%, and Micron falls short by 1.4%. Cut earnings in half, and all three look expensive at today’s price.

Here’s how to read that. Samsung Electronics has the most room to absorb a serious earnings decline. SK Hynix has the best underlying business of the three, but less room than Samsung. That’s because its margin has already climbed to 76%, so it also has the most to give back when things normalize. Micron has disclosed its contracts most transparently and gets a premium for its US listing, but that advantage is already priced in, leaving it with the least room of the three.

There’s one more thing worth checking. If you instead divide the share price by net assets (book value), the order flips. On that measure, the two Korean companies are trading high by historical standards. So the same companies look cheap by earnings and expensive by assets at the same time, which can be read as a sign that the market sees today’s earnings surge as temporary.

7. What to watch starting in September

Here’s the calendar of what’s ahead, in order.

The risks are worth noting too. There’s talk that the Bank of Japan could raise interest rates in September. If the yen carry trade, money borrowed in yen to invest in other assets, unwinds, Korean stocks could get shaken too. If memory prices keep climbing, buyers’ margins will keep shrinking, and at some point they may cut orders or downgrade specifications. And the investment that the three memory makers and equipment suppliers are ramping up right now could turn into oversupply sometime after 2028.

8. To sum up

The companies and their stocks are two different stories. In August, the companies got better, and that’s confirmed by earnings, contracts, and shipment records.

The reasons the stocks split apart are also clear: the type of shareholder returns, China as a new competitor, and the difference in how money behaves in the US market versus the Korean market. Two of these have a timeline attached. Samsung Electronics’ decision on its remaining payout comes next January, and CXMT’s real capability will show up as shipments pile up.

And when a widely repeated number comes along, like Micron’s supposed 47%, it’s worth checking once. What changes a conclusion is usually not the interpretation, but a single number.

The stocks mentioned in this post are examples used for analysis and are not a recommendation to buy or sell any specific stock. Responsibility for investment decisions and their outcomes rests with the investor. Korean stock prices were verified directly against Korea Exchange data, and Micron’s stock price was cross-checked against two market data sources; all figures are as of the closing price on August 28, 2026. Micron’s year-to-date gain is about 227% measured from last year’s final closing price, and about 196% measured from this year’s first closing price. The 2027 expected earnings and the reduced figures calculated from them are consensus estimates as of publication date plus our own calculations, and they vary by brokerage and by point in time. Samsung Electronics’ 80% HBM4 yield is industry reporting, not an official company disclosure, and the estimate of a 10 to 20 trillion won cancellation is also a brokerage analysis. What is confirmed by Korea Customs Service data is a 198.8% increase in semiconductor exports from August 1 to 20; detailed figures by product were not available in public data, so they were not included in the body.

Related posts

• Good Companies and Cheap Stocks Are Different Lists: Anatomy of Fourteen Korean Valuations
• Anatomy of the Rebound: Foreign Buying Came in Two Bursts, and the Same Two Stocks Changed Hands
• The AI bottleneck really is HBM, not GPUs: but bottlenecks invite workarounds