Jensen Huang's sweeping move crushed global memory stocks, and only ChangXin Memory Technologies was spared from the slump.

HBM has long been Jensen Huang's biggest headache. To secure HBM supplies, Jensen Huang led his team to travel across Asia back in June this year.

However, Jensen Huang's attitude changed abruptly. On August 7, he made a major decision: NVIDIA drastically cut the memory configuration of its next-generation flagship GPU, Rubin Ultra.

According to leaks, NVIDIA has internally tested at least three low-memory versions over the past few weeks. Eventually, it cut the HBM configuration of mainstream models from the originally planned 12-Hi (around 384GB) to 8-Hi 192GB, almost halving the per-card memory. The GPU die count was also reduced from 4-die to 2-die, and the power consumption dropped from 2300W to 1800W.

Right after the news broke, SK Hynix's US-listed ADR fell 4.97% on the same day, and its South Korean stock dropped by more than 10% the next day.

Not only HBM sellers suffered, but even SanDisk, which sells NAND products, was also dragged into the downturn.

Just two days before NVIDIA released the news, SanDisk released the strongest financial report in its history.

In the fourth quarter of fiscal year 2026, the company's revenue reached 8.97 billion US dollars, surging 372% year on year and 51% quarter on quarter, about 7% above market expectations. Its non-GAAP earnings per share hit 39.25 US dollars, 14% higher than analysts' forecasts.

In fact, the skyrocketing performance of memory stocks driven by rising AI hardware demand is no longer new news.

SanDisk posted a net profit of 6.9 billion US dollars for the quarter, and a full-year net profit of 11.4 billion US dollars, with a staggering 797% year-on-year increase. The company also announced a 14-billion-US-dollar share repurchase program, with gross margin remaining at a historically high level of 83% to 85%.

If you only look at these figures, this is absolutely the most brilliant performance in SanDisk's history. However, the market responded with its after-hours share price once falling by more than 8%, and finally closed down by around 5%.

It did not end there. Back in July before the earnings report was released, SanDisk's share price had already fallen by around 40% in a single month, with a maximum drawdown of more than 55% from its all-time high in June.

Booming performance plummeting share price - this seemingly contradictory scenario is actually unfolding across the global memory industry.

From Samsung to SK Hynix, from Micron to Kioxia, almost all storage chip giants are going through the exact same plot.

Profits hit record highs, while share prices keep falling. SanDisk's latest earnings report is just another repeat of this story. Its value lies in that as a pure NAND player, SanDisk's performance and guidance provide a clear mirror for us to observe the cycle position of the entire memory industry.

01

The Common Plot of Global Memory Stocks

SanDisk gave its revenue guidance for the first quarter of fiscal 2027 during its earnings call. The company expects Q1 revenue to be between 10.3 billion and 10.8 billion US dollars, with a midpoint of around 10.55 billion US dollars, while the general market expectation ranges from 10.8 billion to 11.16 billion US dollars.

The midpoint of the guidance is about 2% to 5.5% lower than market expectations, which is not a huge gap, but under the current market sentiment, this small miss has been magnified infinitely.

SanDisk's Q4 revenue increased by 51% quarter on quarter, but calculated at the midpoint of Q1 guidance, the quarter-on-quarter growth rate will plummet to around 17.6%. Meanwhile, SanDisk itself disclosed that roughly two-thirds of the quarter-on-quarter revenue growth in Q4 came from price hikes, and only one-third came from higher shipment volume. This means that once prices stop rising, the growth rate will drop at a much faster pace.

Therefore, even though SanDisk is earning more and more, its growth rate has suddenly slowed down. The stock market trades on expectations: investors buy your stock not for how much you earn now, but for how fast your earnings will grow in the future. Once the growth rate drops, expectations change, and the share price falls.

On July 29, SK Hynix released its Q2 2026 earnings report, posting revenue of 79.32 trillion won, up 257% year on year; operating profit of 60.54 trillion won, surging 557% year on year; operating margin reached 76.3%, and gross margin hit 83%.

There is no doubt that these figures have refreshed the historical record of the storage industry, even exceeding NVIDIA's 75% gross margin level. As a hardware vendor, SK Hynix's gross margin level is now on par with that of software companies like Meta.

Moreover, SK Hynix posted 60 trillion won in operating profit for a single quarter, exceeding its total full-year profit of 47.2 trillion won in 2025.

But on the day the earnings report was released, SK Hynix's share price once plunged by more than 19%, hitting its largest single-day drop in history. It fell by a cumulative 25% this week, with its market value evaporating by about 308 trillion won, equivalent to 1.45 trillion yuan.

The reason is also "below expectations". The market previously forecast SK Hynix's Q2 revenue to be 85 trillion won, but the actual figure was only 79.3 trillion won, a gap of around 5.7 trillion won. The reasons include delayed HBM4 shipments, weak product mix, and lower-than-expected proportion of high-end products. In the spotlight, shortcomings are magnified infinitely, creating various reasons for sell-offs.

Samsung Electronics' situation is exactly the same. Samsung's Q2 earnings report released at the end of July shows that its operating profit reached 89.5 trillion won, up 1814% year on year; net profit was 71.6 trillion won, up 1299% year on year.

These figures are also at a historical level, but Samsung's share price continued to fall after the earnings report was released, with a cumulative decline of around 17% in July, and its market value evaporating by about 244.8 trillion won. Calculated from the June high, the combined market value loss of Samsung and SK Hynix has exceeded 4 trillion yuan.

The market does not price in current profits, but the changes in profits. When a company's profit rises from 100 million to 1 billion US dollars, with a growth rate of 900%, its share price will skyrocket; when profit rises from 1 billion to 6 billion US dollars, with a growth rate of 500%, the share price may still rise; but when the market expects the profit growth rate of the next quarter to drop from 500% to 200%, even if the absolute profit is still hitting new highs, the share price will start to fall.

Because the top of a cyclical stock never comes when profits are at their highest, but the moment when the growth rate starts to slow down.

The so-called "guidance cliff" refers exactly to this: performance is still growing, but the growth rate suddenly drops off a cliff, and the market's expectations for the future are reversed instantly.

SanDisk's Q1 guidance is the first clear signal of this cliff in the NAND track. The quarter-on-quarter growth of 51% in Q4 suddenly drops to 17% in Q1, which is such a large deceleration that the market believes the NAND price hike cycle has come to an end.

As mentioned earlier, two-thirds of SanDisk's Q4 growth came from price hikes, which means prices contributed to the vast majority of the incremental revenue.

But prices cannot rise forever. All kinds of hardware products on the market have already seen skyrocketing prices due to AI grabbing production capacity. If prices continue to rise, downstream customers' purchasing power will reach the limit, and alternative solutions will keep popping up, after all, no one has unlimited money.

Although Samsung and SK Hynix's HBM business is still growing, its growth rate has already started to slow down, and the delayed shipment of HBM4 has cast a shadow over the growth pace of high-end products.

TrendForce predicts that DRAM contract prices will still rise 13% to 18% quarter on quarter in the third quarter of 2026, and NAND flash contract prices will rise 10% to 15% quarter on quarter, but NAND's increase is far lower than the previous optimistic expectation of 21%.

TrendForce says that NAND Flash supply will become loose in the second half of 2027, and prices may face correction pressure. The situation of DRAM is slightly better, as HBM continues to squeeze out production capacity and AI server demand remains strong.

However, TrendForce also believes that this good situation will not last long. After all, the market has its limits, and DRAM is not a consumable, so there is no reason for the market to keep constant demand for it.

02

Jensen Huang Seems Angry

If SanDisk's guidance cliff is the signal of the NAND cycle peaking, NVIDIA's memory configuration cut for Rubin Ultra is absolutely a devastating blow to the memory market.

TrendForce later released a report pointing out that NVIDIA is evaluating multiple solutions in parallel, including HBM4e 8hi, HBM4 12hi, and HBM4 8hi, and the final specification will not be confirmed until after verification in the second half of 2026.

Right after this news came out, Samsung and SK Hynix were completely stunned.

HBM is the category with the highest unit price and the thickest profit among storage chips, and it is also the core engine of South Korean manufacturers' profit growth in this super cycle.

SK Hynix currently accounts for about 70% of the HBM market share, and Samsung accounts for around 20%.

The underlying logic for the market to give SK Hynix high valuation before was all centered on HBM: the HBM capacity of AI chips is getting larger generation by generation, the per-card value keeps rising, and the HBM growth curve will keep climbing steeply.

According to estimates by SemiAnalysis, driven by continuous HBM price hikes, the BOM cost of the Rubin Ultra rack had originally climbed from 6.6 million US dollars to 8 million US dollars, of which HBM alone accounts for about 28%.

For HBM manufacturers, the gap in revenue and profit between selling eight chips and selling sixteen chips is very large.

NVIDIA's cut can be said to have "shattered" the HBM market.

According to 404K Research's calculation, Rubin Ultra accounts for about 20% of global HBM demand in 2027. If its memory is cut from 384GB to 192GB, it is equivalent to 10% less HBM sold globally.

But the core problem is not here. Originally, Rubin Ultra was planned to use 12-layer stacked HBM4e, but now the mainstream version is downgraded to 8-layer HBM4. The unit price of HBM4e is 20% to 30% higher than that of HBM4, and 12-layer stacking has 50% more chips than 8-layer stacking.

Combining the two factors, Semi Analysis says that the HBM value corresponding to a single GPU has dropped by far more than 50%, and the HBM expenditure per rack has shrunk from 2.24 million US dollars to about 900,000 US dollars.

HBM4e is SK Hynix's most profitable product line, because its 12-layer stacked HBM4e has low yield, few competitors, and extremely strong premium capability.

If NVIDIA's main models all switch to HBM4 8-Hi, not only will the unit price drop, but the competitiveness of Samsung and Micron will also be relatively improved.

On July 10, the day SK Hynix listed on NASDAQ, its CEO Kwak Roh-jung said in an interview that 2027 will see the most severe supply shortage in the history of the storage industry, the situation of supply falling short of demand will last until after 2030, and customer demand far exceeds production capacity, even if they try their best to expand production, they cannot keep up.

The market even took for granted that HBM prices would keep rising. But now it seems that it is almost time to hit the brakes.

03

Domestic Substitution

When discussing global memory stocks, there is an unavoidable variable, that is domestic substitution.

In July, when SK Hynix and Samsung started to plunge, there were rumors that domestic storage would break the hegemony of SK Hynix and Samsung.

Changxin Technology was listed on the STAR Market on July 27, with a 465.82% increase on the first day. In less than 24 hours, its market value exceeded 3.28 trillion yuan.

To understand the real impact of domestic substitution on the global memory landscape, we first need to understand where Changxin stands.

According to data from Counterpoint Research, Changxin Technology's share in the global DRAM market rose from 3% in the first quarter of 2025 to 8% in the first quarter of 2026, and it is currently the world's fourth largest DRAM supplier.

Although this is enough to prove that domestic storage has achieved the breakthrough from 0 to 1, it is still far from challenging the status of the three giants.

In terms of technology, Changxin has now achieved mass production of DDR5 and LPDDR5 at scale, and its self-developed 4F process has broken through the 17nm process, perfectly avoiding the use of EUV lithography machines.

This technical level is roughly equivalent to that of South Korean manufacturers two years ago. It already has competitiveness in the mainstream consumer and server markets, but in the HBM field, Changxin still has a clear gap.

HBM is rather special, as it requires more advanced process and more complex packaging technology, and Changxin does not yet have the capacity for large-scale HBM mass production.

In terms of production capacity, Changxin is expanding very fast. By the end of 2026, the company's monthly wafer production capacity is expected to exceed 300,000 pieces.

According to Xinhua News Agency's report in mid-July, both Changxin and YMTC are building new factories, and their total production capacity after 2027 is expected to reach more than twice the current level.

At present, Changxin has 3 12-inch wafer fabs in Hefei and Beijing, with a monthly production capacity of about 280,000 to 300,000 pieces, which is expected to rise to 300,000 to 350,000 pieces by the end of 2026.

Apple is an extremely big customer for the global memory