Data basis (read this first)
- ⚠️ U.S. markets were still open at the time of writing. All same-day U.S. figures are live intraday prices, as-of 2026-09-10 10:32 ET (22:32 Beijing time), roughly one hour into the session. Intraday prices move, and the close may contradict this piece. Pre-registered closing criteria are given at the end.
- 09-08 and 09-09 U.S. figures are settled closing prices, sourced from same-day readings files.
- Options data is as-of the 09-09 U.S. close (U.S. equity options do not trade overnight and do not refresh intraday). This is labeled at every point of use and never mixed with 09-10 intraday prices.
- This article contains no positions, share counts, dollar amounts, or trade recommendations.
Bottom line in one sentence
Over the last three sessions, the spread between memory stocks and the semiconductor ETF has flipped sign three times while going almost nowhere in aggregate. Reading any single one of those sessions as a "regime change" would have been contradicted by the next one. This is rotation, not a trend.
The problem: three sessions, each of which looked like a regime change
To judge whether a stock is trending, its own percentage move is not enough, because most of that move comes from its sector. The correct read is the relative spread: the stock's move minus the sector ETF's move.
Here is that spread for Micron (MU) and SanDisk (SNDK) against the semiconductor ETF (SMH) over the last three sessions.
Table 1: Memory stocks' spread vs. SMH, three sessions
| Session | SMH | Micron | MU spread | SanDisk | SNDK spread |
|---|---|---|---|---|---|
| 09-08 close | +1.19% | −1.61% | −2.80 pp | −0.12% | −1.31 pp |
| 09-09 close | +0.10% | +2.75% | +2.65 pp | +1.51% | +1.41 pp |
| 09-10 intraday (10:32 ET) | −2.08% | −3.50% | −1.43 pp | −3.15% | −1.08 pp |
| Three-session simple sum | −1.57 pp | −0.98 pp |
pp = percentage points. Relative spread = stock return − SMH return. ⚠️ Spreads are computed from unrounded returns, so they may differ by 0.01 pp from subtracting the rounded percentages shown in the table.
The price chain can be verified independently: Micron 1,000.26 (09-08) → 1,027.77 (09-09) → 991.77 (09-10 intraday); SanDisk 1,737.99 → 1,764.17 → 1,708.55; SMH 573.73 → 574.29 → 562.37.
What this table shows
- Three sessions, signs of −, +, −. Three flips. Not one session continued the direction of the one before it.
- Each move was large. Two of Micron's three daily spreads exceeded 1.4 points, and the 09-09 session reached 2.65 points. Any one of them, read alone, is big enough to support a confident story about memory being abandoned or memory coming back.
- Yet the three-session sum is only −1.57 and −0.98 points. The cumulative net displacement over three days is smaller than the swing on any single one of those days.
When daily swings persistently exceed cumulative displacement, what is happening is rotation, not a trend. That is the core test in this piece.
The test: a pre-registered bucketed criterion, and today it triggered bucket three
Writing the criterion down in advance and then reconciling against it is the only defense against post-hoc storytelling. The following buckets were fixed after the 09-09 close (before the U.S. open), measuring Micron's and SanDisk's spreads against SMH.
| Bucket | Condition | Meaning |
|---|---|---|
| One | Both ≥ +1 pp | Memory leadership continues; 09-09 was not a one-day event |
| Two | Both between −1 and +1 pp | Simply tracking the sector; no independent information |
| Three | Both ≤ −1 pp | 09-09 was one-day noise; the leadership claim is withdrawn |
| Four | One ≥ +1, one ≤ −1 pp | DRAM and NAND are diverging internally; do not describe them as one |
09-10 intraday reading: Micron −1.43 pp, SanDisk −1.08 pp. Both ≤ −1 pp. Bucket three triggered.
The conclusion is that 09-09's relative strength in memory has, so far, been a single-session event. This is an intraday reading and the close may land in a different bucket; closing criteria are specified at the end.
Separating intra-sector rotation from broad de-risking
At the single-stock level these two look identical (everything is red), but they mean opposite things. Rotation means money is still in the sector and merely changing seats. Broad de-risking means money is leaving. Two measures distinguish them.
Test A: sector decline as a multiple of the broad index
| Instrument | 09-10 intraday |
|---|---|
| S&P 500 ETF (SPY) | −0.47% |
| Nasdaq 100 ETF (QQQ) | −0.79% |
| Semiconductor ETF (SMH) | −2.08% |
The semiconductor ETF is falling 4.4x as fast as the S&P and 2.6x as fast as the Nasdaq 100. The broad indices are largely holding, which means this is not a market-wide collapse in risk appetite. The pressure is concentrated inside semiconductors.
Test B: dispersion within the sector
Table 2: Semiconductor cross-section, 09-10 intraday (as-of 10:32 ET)
| Company | Ticker | Intraday | vs. SMH |
|---|---|---|---|
| Intel | INTC | −5.08% | −3.01 pp |
| Micron | MU | −3.50% | −1.43 pp |
| SanDisk | SNDK | −3.15% | −1.08 pp |
| Western Digital | WDC | −3.13% | −1.05 pp |
| AMD | AMD | −2.35% | −0.28 pp |
| Nvidia | NVDA | −2.27% | −0.19 pp |
| Semiconductor ETF | SMH | −2.08% | — |
| ASML | ASML | −1.85% | +0.23 pp |
| TSMC | TSM | −1.56% | +0.52 pp |
| Marvell | MRVL | −1.19% | +0.88 pp |
| Super Micro | SMCI | −0.46% | +1.61 pp |
| Broadcom | AVGO | −0.27% | +1.80 pp |
The spread between the best and worst name is 4.81 percentage points (Intel −5.08% to Broadcom −0.27%), against a sector move of only −2.08%. Internal dispersion larger than the sector's own displacement is the signature of seat-changing.
Putting it together
- Broad indices holding + sector falling hard + high internal dispersion → rotation / momentum unwind inside semiconductors, not broad de-risking.
- The opposite pattern (indices and sector falling together, names within the sector all falling by similar amounts) would be the signature of broad de-risking.
⚠️ The limit of this inference: a rigorous diagnosis requires index implied correlation. Elevated single-name implied volatility combined with low index implied correlation is what confirms a momentum unwind. This piece uses realized dispersion only as a proxy. No implied correlation data was available, so this is a directional inference, not a confirmation. If index implied correlation subsequently turns upward, the classification must change from rotation to broad de-risking.
An observation worth recording but not yet a conclusion: the unhedged names fell hardest
Options data at the 09-09 close (U.S. options do not trade overnight; these are frozen values) shows the following put/call ratios, where a lower number means less money spent on downside protection.
| Instrument | 09-09 put/call | 09-10 intraday |
|---|---|---|
| Micron | 0.425 (lowest) | −3.50% |
| SanDisk | 0.479 | −3.15% |
| TSMC | 1.635 (highest single name; 367k open interest, the largest in the set) | −1.56% |
| Semiconductor ETF | 1.076 | −2.08% |
| Nasdaq 100 | 1.674 | −0.79% |
| S&P 500 | 0.881 | −0.47% |
The two least-protected names posted the largest intraday declines, while TSMC, where protection was most heavily bought, outperformed the sector by 0.52 points.
A structural fact is visible alongside this: protective money is concentrated on the indices (Nasdaq 100 at 1.674 and the semiconductor ETF at 1.076 are both above 1) while the single names are largely naked (all below 0.55 except TSMC). When the shock arrived, the unprotected single names absorbed the larger drawdown.
⚠️ The limit that must be stated plainly: this is a single observation, n=1, not a regularity. A low put/call ratio may simply be a restatement of crowded bullish positioning, and no causal link to subsequent declines has been established. Promoting this into a usable leading indicator would require it to repeat across multiple different names and multiple points in time, and to hold on up-sessions as well (i.e., that lightly-hedged names also rally harder). Until then it is a lead to be tested, not a finding.
Pre-registered closing criteria (fixed; not revised for intraday moves)
This piece was written intraday. The following buckets are for review after the 09-10 U.S. close, again measuring Micron's and SanDisk's spreads against SMH.
| Bucket | Condition | Conclusion |
|---|---|---|
| A | Both ≥ +1 pp | Intraday weakness reversed into the close; this article's main conclusion is void and must be rewritten |
| B | Both between 0 and +1 pp | Weakness compressed; "three sign flips" is downgraded to two and a half; rotation call stands |
| C | Both between −1 and 0 pp | Weakness present but below threshold; bucket two; no independent information |
| D | Both ≤ −1 pp | Intraday reading holds; three flips in three sessions confirmed at the close |
| E | One ≥ +1, one ≤ −1 pp | DRAM and NAND diverging; the two must be treated separately and not described as one |
Extended test (five-session window): if over the next two sessions individual daily spreads continue to print absolute values ≥ 1 pp while the five-session cumulative absolute value remains below 2 pp, the rotation-not-trend call is strengthened. If the five-session cumulative absolute value exceeds 3 pp with a consistent direction, the call should be revised to a trend.
FAQ
Are memory stocks falling on Sept 10, 2026?
Yes. As of 10:32 ET intraday on 2026-09-10, Micron is −3.50%, SanDisk −3.15%, and Western Digital −3.13%, all weaker than the semiconductor ETF at −2.08%. These are live intraday prices and the close may differ.
Was the Sept 9 rally in memory stocks the return of a leadership trend?
On the current data, no. Micron beat SMH by 2.65 points on 09-09, but gave back 1.43 points intraday on 09-10, and had already trailed by 2.80 points on 09-08. The three-session sum is −1.57 points, which establishes no direction.
How do you tell whether a sector selloff is rotation or broad de-risking?
Look at three things. First, the sector ETF's decline as a multiple of the broad index (a high multiple means pressure is concentrated in the sector). Second, the dispersion between the best and worst names inside the sector (high dispersion means seats are changing). Third, index implied correlation (low indicates a momentum unwind; turning upward indicates a shift to broad de-risking). On 09-10 intraday the first two point to rotation; this piece has no data on the third.
Why use the spread versus the sector ETF instead of the raw percentage move?
Because most of a stock's daily move is sector beta. Reading the raw move causes a sector-wide decline to be misread as a problem specific to that stock. Subtracting the sector ETF's return leaves the part that is actually information about the individual name.
What does a low put/call ratio indicate?
It indicates relatively little money spent on downside protection. It is usually read as bullish sentiment, but it also means holders have no hedge cushion if the stock falls. The 09-10 intraday observation is consistent with this, but a single observation does not establish a regularity.
Disclaimer: This is a public-layer market structure analysis. It contains no positions, trades, or trade recommendations and does not constitute investment advice. U.S. figures are live intraday prices as of 2026-09-10 10:32 ET (22:32 Beijing time) and move continuously; options data is as-of the 2026-09-09 U.S. close. All figures are drawn from live quote interfaces and same-day readings files; no estimated or inferred values were used.