$Sandisk(SNDK.US) the turnaround has arrived

In-depth follow-up on semiconductor export data for May-July: Storage cycle peaks,

I. The Most Core Counter-Intuitive Truth

The market generally believes:

Storage prosperity = AI prosperity = Volume and price rise together = Cycle upward

But real customs data proves:

1. The increase in May-June was entirely due to price increases, not volume.

2. In July, there was a direct collapse with working-day average daily exports dropping by -28.5%.

3. The industry has entered classic late-cycle tail characteristics: high prices but no market, strong prices but weak volume, and volume peaking before prices.

The only iron rule for the peak of the semiconductor cycle:

Terminal shipment volumes decline first, prices are propped up by manufacturers controlling production and long-term agreements, and finally prices make up for the drop, resulting in a double kill of performance.

Currently, it completely replicates the end-phase patterns of the two major tops in 2018 and 2021.

II. Deep Deduction of Data: Why is "Volume Drop" More Terrifying Than "Price Drop"?

1. Export value is a lagging indicator, while shipment volume is a leading indicator.

- ASP (price) is contract price, locked in long-term agreements, rigidly lagging.

- Shipped Bit volume and customs average daily export intensity reflect real terminal demand and real order momentum.

A fatal divergence has now appeared:

Prices are still hitting new highs → Real delivery demand has already plummeted.

This is a typical late-cycle divergence (Top Divergence).

All major semiconductor tops have died from this structure.

2. The Real Illusion of the Increase from May to June (One-time Overdraft)

The reason June was the highest of the year:

It wasn't because demand improved, but due to two one-time overdraft factors:

1. Cloud providers concentrated delivery of HBM long-term agreement orders in Q2 (half-year centralized fulfillment).

2. Q2 DRAM contract quarterly price increase of 58–63%, pushing single-month ASP to the annual peak.

It's not sustained demand, it's centralized settlement.

So July saw a direct cliff-like drop because:

The peak of long-term agreement deliveries ended + Terminal new inventory buildup dried up.

III. Fatal Structural Negative Specific to Samsung Electronics (Exclusive Deep Logic)

1. Samsung is the biggest victim of this round of "weak volume, strong price".

- SK Hynix: HBM yield rate 90%+, core supplier to NVIDIA, has both volume and price.

- Samsung: Low HBM yield rate, share only 22%, no volume, can only rely on low-end storage price increases to support revenue.

Data translation:

SK Hynix earns from true AI incremental growth, while Samsung earns from the bubble of traditional storage price increases.

2. The plunge in July exports mainly hit Samsung.

Korean semiconductor export structure:

- High-end HBM (led by SK Hynix) volume relatively stable.

- Ordinary DRAM/NAND (Samsung's main force) demand completely extinguished.

Therefore:

The -28.5% daily average drop in July was caused by the pressure on Samsung's traditional storage business.

IV. Verification of AI Capital Expenditure and Data Correspondence (Closed-loop Solid Evidence)

The cloud provider CAPEX logic we deduced earlier:

Concentrated computing power deployment in Q2, marginal cooling in Q3, slowing new orders.

Customs data now perfectly verifies the closed loop:

1. Q2 (May-June): Concentrated delivery, export surge.

2. Early Q3 (July): Vacuum in new demand, collapse in average daily shipments.

Proof:

The marginal weakening of AI capital expenditure has moved from expectation → to actual real data.

This is a negative factor that institutions have not priced in at all.