US Pre-Market Brief | 2026-09-07 (ET) Monday

Today is Labor Day in the US, and NYSE, Nasdaq and the bond market are closed all day, so there is no pre-market session and no 09:30 open. This piece is therefore a holiday-edition pre-market brief written for the Tuesday, 2026-09-08 open.

• Coverage window: 2026-09-04 (Friday) 16:00 ET regular-session close → 2026-09-07 about 08:15 ET
• Covers: Friday's close and after-hours → the weekend (9/5–9/6) → today's Asian and European sessions → today's Labor Day futures session (E-mini closes early at 13:00 ET)
• Single-stock percentage moves are on a 09-04 regular-session close basis (CNBC quote API), each one verified by back-computing from the prior close
• Treasury yields are taken from the official US Treasury yield-curve CSV; the financial data for the storage and equipment names comes from primary-source SEC EDGAR 10-Q / 10-K / 8-K filings
• ⚠️ Coverage blind spot: the data runs through 9/7 08:15 ET. There are still about 25 hours between then and the 9/8 open (today's European close, the Labor Day futures segment through the 13:00 ET early close, and Monday-evening announcements) that this piece does not cover; you must sweep that window yourself before the open

---

0. One-Sentence Summary of the Day

1. Friday's big green candle was the worst-beaten part of the whole chain bouncing, not the start of a new leadership theme. All six semiconductor-equipment and AI-interconnect names are sitting in drawdowns of 29.8%–44.7% from their 6/30 highs (CRDO −44.7%, KLAC −39.6%, AMAT −38.5%, LRCX −29.8%). This is a bounce within a drawdown, not a breakout. (Volume was mixed: SNDK 1.51x, MU 1.48x and KLAC 1.56x were above average, while MRVL 0.84x and AMAT 0.94x were below; full table in §8.2.)
2. One event definitely did happen, but it cannot explain the whole move. S&P Dow Jones Indices announced that SanDisk, Dell, Palo Alto and Arista will be added to the S&P 100 on 9/21, replacing Nike, Simon Property, Colgate and Honeywell Aerospace. But the other three names added in the same batch rose only +1.50% (DELL), +1.22% (ANET) and +0.40% (PANW) — that is the order of magnitude mechanical passive money should produce. Within SNDK's +11.90%, the index event can only explain a small part.
3. This piece cannot attribute Friday's storage move to a single driver, and that is the honest conclusion. Explanations already ruled out: ① it is not Astra — Astra was released on 9/3, and on that day MU was only +0.22%, SNDK only +0.10% and SOXX +0.15%, while software IGV was up +3.41% instead (⚠️ but I was unable to find the time of day of the Astra release; if it went out after Friday's — after 9/3's — close, then 9/4 was the first tradable session and this exclusion does not hold, see §1.1①); ② it is not rates — the 2-year Treasury rose only +3bp on the day and the 30-year actually fell −1bp; ③ it is not company news — multiple outlets explicitly reported that there was no company-level news that day. What is left is sector fund flow, whose trigger is unconfirmed.
4. Leadership has not changed hands. From the 6/30 close to the 9/4 close: NVDA +15.13%, MU −11.93%, AMAT −37.11%. The claim that "money is rotating from the compute leader into the AI physical layer" is an illusion produced by a single-day cross-section; on a 10-week scale it is exactly the opposite.
5. Driver types: index rebalancing (liquidity) + macro (jobs beat → rate-hike re-pricing) + ratings and governance (Morgan Stanley downgrades Adobe, CEO change) + geopolitics (the US-Iran tanker war escalating). Rate hikes — not cuts — are the current policy baseline.
6. Prior session (9/4) close: S&P 500 7,718.60 (−0.38%), Nasdaq Composite 26,506.99 (−0.29%), Dow 53,414.25 (−0.51%), Russell 2000 2,975.65 (+0.25%), VIX 14.53. Sectors: SOXX +3.52% / Philadelphia Semiconductor Index +3.38% vs software IGV −2.23%; XLK +0.70%, XLE −0.87%, XLV −1.04%, XLF −0.79%.
7. Pre-market state today (about 08:15 ET): S&P futures −0.23%, Nasdaq 100 futures −0.09%, Dow futures −0.54%; Brent $97.47 (+1.24%), WTI $92.24 (+0.83%); gold $4,431 (−1.02%); dollar index 98.957 (−0.22%). Asia closed sharply higher today (KOSPI +4.61% to 6,995.39, Nikkei +2.12% to 66,399.84, both on a closing basis), while Europe drifted slightly lower intraday. ⚠️ There is no spot VIX reading today: Cboe is closed, SPX options do not trade, and spot VIX has no inputs. The roughly 15.25 printed by quote vendors, with a "continuously ticking" timestamp, cannot be trusted (S&P futures are only −0.23%, which on VIX's intraday elasticity to SPX corresponds to +1%–+2%, 2–4x away from +4.9%). This piece uses only the 9/4 close of 14.53 for VIX throughout.

In one sentence: the real thing to watch on Tuesday is not a "storage leadership theme" but whether one unexplained big green candle can be confirmed by a second one, the 9/10 ORCL and ADBE earnings, and the supply divergence between DRAM and NAND inside storage; when attribution is unclear, the correct posture is to downgrade rather than chase, and that is this piece's stance throughout.

---

1. News Overview

---

1.1 Five Market Narratives That Must Be Corrected First

① "Astra ignited storage" — the sequence refutes it directly.

On the day Astra was released, storage stocks barely moved (MU +0.22%, SNDK +0.10%), and the biggest gainer was software instead (IGV +3.41%).

The accurate conclusion can only go this far: prices on 9/3 did not reflect Astra; the reaction from 9/4 onward cannot be distinguished from the index event or an oversold bounce. This piece therefore does not treat Astra as the explanation for Friday's price action.

⚠️ This conclusion has two weaknesses that must be acknowledged:

1. I was unable to find the time of day (ET) of the Astra release. If it went out after the close on 9/3, then 9/4 was the first tradable session and the inference of "a full day's delayed reaction" collapses on the spot. This is the single pivot of the whole argument, and it is the one field that most needed filling and was not filled.
2. The 9/3 cross-section is self-contradictory inside this piece's own framework: §1 labels Astra "bullish hardware · bearish software", yet that day IGV was +3.41%, the strongest thing on the board. That says the market on 9/3 was not trading Astra at all — in which case "storage did not move on 9/3" can neither falsify nor confirm anything; it is an observation with close to zero information content.

This piece's treatment of Astra is therefore: acknowledge its long-run fundamental implication for memory demand, but do not use it to explain any single day's prices, and do not use it to support any name's rating. By the same token, today's surge in Asian storage stocks (Korea +4.61%) is attributed by local media to Astra; this piece accepts their price figures but not their attribution, and does not count it as any part of the bull case.

② "Leadership has rotated to the AI physical layer" — on a 10-week scale it is exactly the reverse.

Drawdowns of the six equipment and interconnect names from their 6/30 highs: KLAC −39.62%, CRDO −44.74%, ALAB −37.85%, AMAT −38.53%, MRVL −32.23%, LRCX −29.84%. Friday was a bounce inside a deep drawdown, not a leadership switch.

③ "Yields spiked" — the closing data does not support it.

The 2-year rose only 3bp and the 30-year actually fell 1bp, while the long-bond ETF (TLT) was +0.17% on the day. Using "spiking yields" to explain a 5.75 percentage-point sector spread does not work on order of magnitude. But this line also rules out another alternative explanation: rates up, the broad market down and semiconductors up hard against the tape says this was not "a high-beta squeeze driven by falling rates" but a sector-specific event.

④ "The jobs data was hot" — hot headline, weak composition. Within the +162,000, food service +59,000 and local-government education +42,000 together account for 62.3%; manufacturing +16,000; while information was −23,000. Average hourly earnings were +3.1% YoY, with no wage-price spiral signature. This is a report driven by low-wage services and government, with white-collar tech jobs being lost.

⑤ "SanDisk's 80% gross-margin target is aggressive" — the comparison base is wrong, but the conclusion is not "conservative" either. SanDisk's most recent quarter (FY26Q4, disclosed 8/5) delivered an actual non-GAAP gross margin of 84.6% (78.4% the prior quarter, 26.4% a year earlier). The Investor Day FY28–30 target of 80% is below the current level, which on its own looks conservative; but the correct comparison base is last year's 26.4% — management is saying "the structural gross-margin center of NAND has moved permanently from 26% to 80%, and still holds four years out," something that has never happened in NAND's history. It is an aggressive assumption written as a conservative number, and it leaves only 4.6 percentage points of buffer to cover an industry whose gross margin can swing 58 percentage points within 12 months.

---

2. Strongest Themes, Descending

2.1 The Most Valuable Line in This Piece: the supply response of DRAM and NAND is completely asymmetric

This comes from the AMAT 10-Q (SEC primary source) and is the only piece of evidence in this report that can pull the catch-all "storage" theme apart:

Supporting evidence: the LRCX 10-K shows storage falling from 42% to 39% of the mix (down 3pct), while foundry rose from 45% to 54% (up 9pct).

Two inferences follow that run against market consensus:

1. Buying MU and SNDK as the same trade on Friday was a mis-mapping. SNDK's (pure NAND) pricing logic finds supply-side support in the equipment data; MU's (mostly DRAM) pricing logic finds supply-side counter-evidence in the same data.
2. Expressing a storage view through KLAC/LRCX/AMAT is betting on a 25%–39% exposure, with the other 60%–75% driven by foundry/logic (advanced nodes for AI logic chips), an entirely different chain.

---

3. Overall Single-Stock Strength Ranking

Percentage moves are on a 2026-09-04 closing basis. Tuesday's open has to digest roughly 89.5 hours of accumulated news all at once (of which the last ~25 hours fall outside this piece's coverage), so gaps will be larger than on an ordinary trading day. The table below is not a buy price.

3.1 Bullish Side

⚠️ Two labels are non-monotonic versus the scores; this is a deliberate manual override, for the reasons below:

• NVDA (47 points yet labeled "watch closely"): the scoring model penalized it heavily on "catalyst directness" and "expectation gap" (no company news Friday, fully priced), but the model does not measure trend standing — since 6/30 it is +15.13% while AMAT is −37.11%; it is the genuinely strong name on this chain. The low score reflects "no event this week," not "weak fundamentals."
• MU (58 points yet ranked behind AMAT at the same score): they tie, and MU's fundamentals are clearly more solid (all four items A in §7.1-A), but its near-term catalyst is on 9/30, not this week, so it is not moved up.

3.2 Bearish Side

---

4. Single-Stock Scoring Model (100 points total)

Two special scoring rules in this piece:

1. "No company news that day + volume below the 50-day average" is judged outright to be an oversold bounce, and catalyst directness is capped at 8 points. MRVL, AMAT and LRCX all qualify for Friday.
2. Wherever TTM net income contains identifiable one-off items (MRVL, ALAB, WDC), the headline P/E is never used; a normalized order of magnitude is used instead and flagged in the table.

---

5. Detailed Analysis of the Top Names

5.1 MU Micron — Watch Only (strongest fundamentals on the board, most dangerous valuation position) | Total score 58

MU's "forward P/E of just 7.06" is the single most easily misread number in this piece. The conclusion runs against intuition: it is the most solid of the four on fundamentals, and also the most dangerously positioned on valuation.

First, take the "7x" apart. MU's forward P/E of 7.07 is not based on any realized earnings but on the FY2027 consensus. ⚠️ That consensus itself has two versions: back-computing from $1,016.59 ÷ 7.06 gives $144.0, while the FY2027 consensus I obtained is $155.03 (which corresponds to 6.56x). The two could not be reconciled, so "7x" should be used only as an order of magnitude. Whichever you take, it is roughly 2x the FY2026 expected EPS ($73.40) and 19–20x FY2025 actual EPS ($7.59). And MU's already-realized TTM GAAP P/E is 22.94x, at the 86.7th percentile of its own 5-year range — not low at all.

So the full version of "7x is cheap" is: "if Micron next year earns at a scale never seen before in semiconductor history, then it is 7x." The low P/E is not an oversight by the market; the low P/E is precisely the market's distrust of that denominator.

But is it actually a "low-P/E trap"? — the precise answer is: not yet; it is the stage before the trap. A textbook cyclical low-P/E trap requires all three at once: a low forward P/E + a high P/B percentile + earnings estimates that have already started to be cut. Right now the first two hold and the third does not — next-quarter consensus for all four names is equal to or slightly above the companies' own guidance (MU 31.28 vs guidance 31.00; SNDK 46.21 vs the top of the range at 46.00; WDC 4.04 vs 4.00; STX 7.35 vs 7.30). The accurate description is "peak-earnings pricing," not "a confirmed trap." The difference is not word games: a trap is confirmed when the direction of revisions turns over, and that has not happened. Calling it a trap now is drawing a conclusion from evidence that has not yet appeared.

"Not a trap" absolutely does not mean "cheap":

• P/B 11.39, at the 98.0th percentile of 5 years (daily frequency, n=1,257), against a 5-year median of just 2.10. This is the only reliable P/B percentile of the four (SNDK has only 19 months of history, STX's equity was once negative, WDC's entity has changed).
• Back-solving from P/B 11.39 with the 5-year median ROE of 13% implies a normalized P/E of about 87.6x; even with the prior cycle's peak ROE of 25% it is still 45.6x. The gap between 7x and 45–88x is the entire difference between a "peak" basis and a "normalized" basis.
• The cash-flow gap (on a consistent basis): TTM net income of about $50.5 billion against TTM FCF of just $26.2 billion — half the profit has already been eaten by capex. That corresponds to TTM P/E 22.94 vs P/FCF 43.9, about 1.9x. ⚠️ Do not write it as "P/E 7 vs P/FCF 43.9" — the former is forward and the latter is TTM, and switching rulers exaggerates the gap by roughly 3x.

One counter-detail that weakens the bear case (though it has a basis problem of its own, disclosed here too): MU's ROE is 130.5% against a 5-year median of 13.0%, a 10x deviation; while P/B deviates from its median by only 5.4x. In a residual-income framework (fair P/B ∝ ROE), the market has marked P/B up by only half as much as ROE, which seems to say the market is already partially pricing mean reversion.

⚠️ But these three numbers are arithmetically inconsistent, so the argument can only count as a "hint," not a conclusion: P/E ≈ P/B ÷ ROE, and 11.39 ÷ 1.305 = 8.73, which does not match the TTM P/E of 22.94 used above. The most likely cause is that ROE uses average equity while P/B uses period-end equity, and MU's shareholders' equity went from $50.7 billion to $100.7 billion over four quarters, so the two denominators differ enormously. Until all three are recomputed over the same period on the same denominator, "the market has already priced mean reversion" should not be used as a conclusion.

Supply-side counter-evidence (the bears' hardest line): MU's quarterly capex went $5.39 billion → $6.39 billion → $7.82 billion, and the $19.60 billion spent in the first three quarters of FY26 already exceeds the $15.86 billion of all of FY25 by 24%. "Capacity expansion does not arrive until end-2027" refers to the timing of bit output, not the timing of equipment purchases — the orders have been placed and the money has been spent. Layer on AMAT's DRAM equipment revenue of +49.5%, and the DRAM supply response is already under construction.

Bull-side evidence (equally real, and a more recent reading): ① contract prices have never turned negative, with 3Q26 DRAM still +13–18%; ② inventory days (COGS basis) went 161 → 122, a healthy drawdown — the signature of a cycle top is inventory piling up, and MU shows none; ③ net cash of $23.75 billion, liabilities/assets 24.9% (lowest of the four), D/E 0.06, interest coverage 258x; ④ no one-off gains or losses at all, with GAAP and non-GAAP virtually identical, OCF/net income 1.02; ⑤ diluted share count up only +2% (the other three +8–9%); ⑥ FQ4 company guidance for gross margin of about 86%, still expanding (versus SNDK, which has already guided flat).

⚠️ One basis that is easy to misread: MU's inventory days on a revenue basis collapsed from 101.8 days to 18.8 days, which looks like "extreme scarcity" — that is an illusion created by price increases; on a COGS basis it merely went from 161 days to 122 days, far milder. Using the wrong denominator turns a mild drawdown into extreme scarcity.

Judgment: the strongest evidence on each side is talking about different clocks — the bulls are talking about "now through mid-2027," the bears about "2027–2028," and both can be right at once. But for Tuesday's trading decision, MU has no near-term catalyst (earnings are on 9/30) and has just been through a bounce with no company news, so it does not qualify as a priority name this week. Conclusion: watch only. The decision point is the 9/30 earnings: ① whether the midpoint of FQ1-27 gross-margin guidance is ≥ the realized figure (this is the first direct reading on whether "DRAM follows NAND into flat"); ② if FY27 capex guidance is ≥$35 billion, the company is closing the shortfall with its own hands.

5.2 ORCL Oracle — Priority Deep-Dive | Total score 78

• Reports FY27Q1 after the close on 9/10 (Thursday). Consensus revenue about $19.13 billion, EPS about $1.299; last quarter's RPO was $638 billion (+363% YoY, up $85 billion QoQ) and cloud infrastructure revenue was +93%, but free cash flow was −$23.7 billion. Options imply about ±11.2%.
• Catalyst logic: the only event this week that can define market sentiment on its own. ORCL is a pure bet on "whether AI capex is sustainable": RPO of $638 billion far exceeds its $457.4 billion market cap, but converting RPO into revenue requires enormous capex, which has already driven FCF to −$23.7 billion. The point of disagreement is financing capacity and depreciation cadence, not demand.
• The largest expectation gap on the board: at $158.78 the stock is 54.1% below its 52-week high of $345.72, and the market has already priced financing risk very heavily. It does not need "good" to rally, only "not worse."
• Strength after hours: +3.08% Friday to close at $158.78, and another +0.58% after hours to $159.70, one of the few large caps to firm after the bell.
• ⚠️ Basis warning: RPO of $638 billion and +363% are stock figures already disclosed last quarter, not new news. What matters this time is new bookings in the single quarter and capex/FCF guidance; do not treat an old cumulative number as a new expectation gap.
• Conclusion: priority deep-dive. Note the implied move of ±11.2% and the pronounced IV crush after the print.

5.3 SNDK SanDisk — Watch Closely | Total score 74

• The real reason for Friday's +11.90% is index inclusion, not fundamentals. S&P Dow Jones announced its addition to the S&P 100 on 9/21. All four newly added names rose that day (SNDK +11.90%, ANET +1.22%, DELL +1.50%, PANW +0.40%) and all three removed names fell (NKE −0.95%, SPG −0.98%, CL −1.47%) — the cross-section cleanly proves this was an index event.
• Media broadly attributed the move to the "$93.9B order book" and the "80% gross-margin target." Both come from the 8/13 Investor Day, three weeks before Friday; JPMorgan's Overweight/$2,250 price target was also a mid-August action. None of it was same-day news.
• The fundamentals really are the strongest: FY26Q4 revenue $8.97 billion, +51% QoQ, +372% YoY, non-GAAP gross margin 84.6%, EPS $39.25; $93.9B TCV, $16.5B of financial guarantees, average contracts of more than 4 years, with floor-price clauses.
• NAND supply-side support (this piece's exclusive plus): AMAT's NAND equipment revenue is −1.6% YoY with the mix down to 7%, and LRCX's storage mix went 42% → 39%. Nobody is adding NAND capacity, which is the hardest evidence for the durability of SNDK's price increases, and it comes from primary-source SEC filings.
• The company disclosed one quantification available nowhere else: "about 1/3 of the FQ4 sequential growth came from volume and 2/3 from price" — the most direct basis in the whole data set for "how much of the growth is price."

⚠️ But SanDisk is the only one of the four with two leading turn signals showing at the same time, and that is this piece's core reservation on it:

In a quarter when NAND spot prices were +480% YoY, piling finished-goods inventory 66% higher while guiding next quarter's gross margin below the realized level — these two signals turning at once, with MU/WDC/STX showing neither, says it is company-specific rather than industry-wide.

• Three further negatives: ① FCF is structurally flattered: its quarterly capex is only on the order of $40 million, which does not mean it is genuinely asset-light — the NAND wafer capacity sits inside Flash Ventures, the joint venture with Kioxia (8 plants in Japan), an off-balance-sheet VIE, and SanDisk carries ¥149.0 billion / about $923 million of guarantee obligations, so its 79% FCF margin is not directly comparable to MU's; ② GAAP EPS of $43.97 exceeds non-GAAP EPS of $39.25, with the difference coming from an $808 million fair-value gain on its stake in Nanya Technology (non-operating); ③ $91.1B of unrecognized TCV ÷ more than 4 years ≈ ≤$22.8 billion per year, against a current annualized revenue run rate of about $35.9 billion — the locked-in portion is below the current run rate; TCV is a floor, not a ceiling.
• ⚠️ Valuation percentiles are unusable: it was only separated from WDC on 2025-02-21, giving it only about 19 months of independent history, and P/B has just 6 independent quarterly observations. Any "5-year valuation percentile for SNDK" is the result of splicing in the old SanDisk that was acquired in 2016, and cannot be trusted.
• Conclusion: watch closely. Its static fundamentals are the most impressive on the board, and the NAND supply-side evidence is the hardest bull argument in this piece; but Friday's gain is the kind that gets given back, "selling the fact" is common around the 9/21 effective date, and its own two leading indicators have already turned. Do not chase.

5.4 SKHY SK Hynix (Nasdaq ADR) — a name pending verification; this piece gives it no conclusion label

A name that is easy to miss: SK Hynix already has an ADR listed on Nasdaq (ticker SKHY), which was +8.14% at $177.00 on Friday on volume of 18.85 million shares; today the Seoul parent closed up about 7% again and the KOSPI closed +4.61%.

⚠️ But this piece uses none of its valuation data and gives it no conclusion label. The reason: there is roughly a 34% conversion discrepancy between the ADR and the Seoul parent, most likely because the ADS-to-share ratio has been quoted incorrectly. And its market cap and P/E are determined entirely by that ratio — get the ratio wrong and both numbers are wrong together. "Whether a company is listed on a given exchange and what its ADS ratio is" is a status-type fact that must be checked against primary-source exchange listings and offering documents, not inferred. Until primary confirmation is obtained, putting it in the ranking table would be irresponsible.

First thing to do next session: check the Nasdaq listing directory and the ADS ratio in the SEC 424B4, and only then decide whether it makes the table.

5.5 LRCX Lam Research — Watch Closely | Total score 63

The only one of the three equipment names with clean operating leverage: gross and operating margins have expanded four quarters running (GM 49.6% → 51.7%, OM 33.9% → 37.4%), with next-quarter guidance of revenue $8.10 billion (+20.5% QoQ, +52.1% YoY), gross margin 52.0% and EPS $2.15. Net cash $1.46 billion, ROIC 70.07%.

• ⚠️ But: a P/E of 53.41 sits at the 93rd percentile of the past 4 years, and at the same time as record-high earnings — the price has already discounted future growth, so the risk is not that earnings peak, it is whether those growth assumptions are delivered. Storage is only 39% of its equipment revenue and falling (foundry 45% → 54%), so it is more a foundry/logic beta than a storage beta.

5.6 XOM ExxonMobil — Watch Closely | Total score 60

The only "divergence-type" opportunity in this piece. The US-Iran tanker war escalated markedly over the weekend and Brent is +1.24% at $97.47 this morning; but energy stocks fell across the board on Friday: XOM −1.69%, CVX −1.29%, OXY −0.94%, XLE −0.87%.

• Supply data: Hormuz crude and product flows fell from 21.6 before the crisis to 4.9 (million barrels per day) in Q2 2026, −77.3% (EIA STEO¹⁸).
• ⚠️ This figure is in tension with this section's own price evidence, which must be stated: Hormuz normally carries roughly one-fifth of the world's seaborne crude, and if total flows really were down 77%, Brent could not be sitting at $97.47 and up only 1.24% today, and energy stocks certainly could not have fallen across the board on Friday. It is reasonable to suspect that 4.9 and 21.6 are not on the same basis (possibly a restricted transit segment, or broken out by product). Until the EIA source table is checked for basis, this piece does not use −77.3% as a directional argument, and the strength of that theme in §9② is downgraded accordingly.
• Counter-evidence: the EIA forecasts Brent averaging just $85 in Q3 and falling back to $69 in 2027, with the current price already 14.7% above the Q3 forecast. Energy stocks not following the oil price up is exactly the market pricing mean reversion.
• Judgment: watch closely. This is an event-driven hedge position, not a trend position, and it is macro-mutually-exclusive with growth stocks — the higher oil goes, the greater the rate-hike pressure.

5.7 ADBE Adobe — Avoid / Short Watch

Down −6.73% Friday to $266.51 on two negatives at once: ① Morgan Stanley cut it to Underweight, PT $240; ② it announced a CEO change. And FQ3 earnings come after the close on 9/10. It has already fallen more than 20% in each of the two preceding calendar years, is down another 18% in 2026 through Thursday, and sits −28.1% from its high.

• ⚠️ Reverse risk: TTM P/E 15.25 and a forward P/E of just 10.27 — expectations have been crushed to an extreme low. With "CEO change + downgrade + earnings" crammed into the same week, any guidance that is less than terrible could trigger a violent squeeze.
• Conclusion: avoid. Bearish directionally, but the odds ahead of earnings are not symmetric (forward P/E of just 10.27, with expectations already crushed to an extreme low).

5.8 MRVL / ALAB / CRDO — Three Names That Must Be Called Out as "Looking Strongest, Actually Weakest"

These three were among Friday's biggest gainers (+7.05% / +9.75% / +3.90%), but primary-source data shows they are the lowest-quality group in this piece:

Four pieces of evidence show these three were an oversold high-beta bounce on Friday rather than event-driven:

1. Volume: MRVL at just 0.80x average, AMAT 0.67x, LRCX 0.79x. Genuine company-level good news does not drive +7% on below-average volume.
2. Cross-section: the same day NVDA was only +0.84%, AVGO +0.21% and ANET +1.22%. If this were new news about AI networking demand, the leaders would not have sat still.
3. Beta and drawdown broadly explain the ordering of the gains: ALAB (beta 3.74) +9.75% > MRVL (beta 2.25) +7.05%, and the three sit in drawdowns of −37.9% / −32.2% / −44.7% respectively. ⚠️ But CRDO is a counterexample: its beta of 3.23 is higher than MRVL's yet it rose only +3.90% (it had just fallen 20.04% on earnings on 9/1, a drag of its own). So this line is corroboration, not a sufficient explanation.
4. Full SEC check: MRVL made no filings and issued no company announcements whatsoever on 9/4.

Three further substantive negatives:

• CRDO's negative operating leverage: revenue +9.6% QoQ while GAAP operating income was −22.6% QoQ; gross margin 68.2% → 64.5% (−370bp), with guidance calling for further declines; sequential growth has flattened from +51.9% to ~10%. It fell −20.04% the day after its 9/1 earnings; the market has already ruled.
• ALAB's operating leverage is implausibly weak: revenue +104.5%, yet operating margin only expanded from 20.7% to 22.7% (+2.0pct), with gross margin 76.3% → 73.3% → guided to ~72%. Its top four customers are 82% of revenue and China ship-to is 38.9%.
• MRVL dilution: warrants issued to Google for 58,970,907 shares at a strike of $206.58 (already in the money), roughly 6.6% dilution if fully exercised.

Conclusion: avoid ALAB / CRDO, watch only on MRVL.

5.9 WDC Western Digital — Avoid | This Piece's Most Important Debunking of an "Apparently Cheap" Name

WDC's headline TTM P/E of just 19.26 makes it look like the cheapest of the four storage names. That number is an illusion.

Primary-source evidence (FY2026 10-K, filed 2026-08-14), verbatim:

"Total interest and other income (expense), net changed by $6.66 billion or 553% in 2026... primarily reflects a mark-to-market gain on our retained interest in Sandisk of $6.50 billion" "the gain on the retained interest in Sandisk being tax-free due to the Separation" "As of July 3, 2026, we no longer held shares of Sandisk common stock."

• TTM investment gains of $6.50 billion ÷ TTM net income of $9.29 billion = 70%. That is, seven-tenths of WDC's reported profit is the mark-to-market gain on the SanDisk shares it held — and that asset was completely liquidated on 7/3, is 100% non-repeatable, and was tax-free on top.
• GAAP EPS $8.21 vs non-GAAP EPS $3.56 — GAAP is 2.3x non-GAAP.
• Using the company's own FY26 non-GAAP EPS of $10.22, the true P/E = 467.46 / 10.22 = 45.7x, not 19.26x.
• Three independent pieces of evidence corroborate each other: ① one-off gains are 70% of the total; ② OCF/net income of just 0.42 (MU/SNDK/STX are 1.02 / 1.02 / 1.15) — non-cash gains necessarily depress the conversion rate; ③ an effective tax rate of 4.9%, with FY27 guidance back to about 17%.
• ⚠️ One inference that has to be self-corrected: while drafting I inferred from "the forward P/E (23.27) is above the TTM P/E (19.26)" that "consensus embeds an EPS decline." That inference is wrong. Consensus is built on an adjusted (non-GAAP) basis, so $20.09 should be compared with FY26 non-GAAP $10.22, i.e. consensus embeds +96.6% growth, not the "decline" you get by comparing it with GAAP $24.28. "Forward P/E > TTM P/E" is entirely an accounting artifact created by switching rulers in the numerator (forward on non-GAAP, TTM on GAAP). By the same token, the supposed "P/E jumping from 19 to 45" is not a jump at all — on a consistent non-GAAP basis it has been 45.7x all along.
• The share price is already confirming it: −41.6% from its 52-week high (the deepest of the four), the only one of the four below its 50-day moving average (−7.8%), up only +7.6% from the 8/7 low (the other three +13.9%–+71.3%), and +171.4% YTD (the lowest of the four).

The other side, which must be stated fairly: WDC's operating business itself has improved a great deal — FY26 operating income of $4.453 billion (a 34.5% margin) against FY25's $2.334 billion, with guidance for gross margin rising further to 55–56%, and it used the SanDisk shares to retire almost all of its debt tax-free, turning net cash positive at $390 million. Its problem is "reported profit ≠ operating profit," not "operations are deteriorating." But consensus requires its non-GAAP EPS to nearly double from $10.22 to $20.09.

Conclusion: avoid. Not because the business is bad, but because its headline TTM GAAP P/E of 19.26 is seven-tenths contaminated by a $6.50 billion, fully liquidated, non-repeatable and tax-free mark-to-market gain; on the same non-GAAP basis as the other three it is 45.7x, the most expensive of the four. The share price is already confirming it: the deepest drawdown at −41.6% from its high, and the only one below its 50-day moving average.

5.10 STX Seagate — Watch Only | Cleanest Earnings Quality, Thinnest Balance Sheet

• Best earnings quality of the four: GAAP EPS $5.58 vs non-GAAP $5.71, a difference of only $0.13, with no investment gains at all; OCF/net income of 1.15, the highest on the board; gross margin has risen monotonically for 9 straight quarters (31.8% → 52.3%) with not one setback, the most linear of the four. FQ1-27 guidance is non-GAAP EPS of $7.30±0.20 (+28% QoQ).
• ⚠️ But the balance sheet is the weakest of the four, by a wide margin: the only one in net debt at −$2.19 billion, shareholders' equity of just $2.17 billion, liabilities/assets 78.3%, D/E 1.80, interest coverage 14.9x (lowest). Key background: STX's shareholders' equity was negative right up to 2025-10-03 (the result of years of buybacks above book value), and it was this earnings cycle that repaired equity back to positive.
• ⚠️ P/B of 88.88 must be discarded: its 5-year P/B series contains negative values such as −853.38 and −66.26 (when equity was negative), and computing percentiles on a series that crosses zero is arithmetically invalid. Use EV-based measures instead: EV/EBITDA 43.24 (89th percentile), EV/FCF 62.70 (100th percentile), P/S 15.78 (100th percentile).
• Conclusion: watch only. The earnings power is credible, the cushion is the thinnest.

5.11 A Stress Test of the Four Storage Names — A Counterintuitive Conclusion

Back-solving each company's cash cost base from its own guidance (after calibration the back-computed gross margins match company guidance: MU 86.1% vs guidance of "about 86%", WDC 55.6% vs "55–56%"), holding costs fixed and cutting only price:

Revenue decline required to reach operating breakeven: MU −82.8%, SNDK −77.8%, STX −48.0%, WDC −45.6%.

Implication (against intuition): the apparently cheapest names, MU/SNDK, have cash costs of only 17%/22% of guided revenue, so their capacity to absorb price declines is actually the greatest; the apparently "normally valued" WDC/STX have cost ratios of 54%/52%, so a 35% price decline takes their valuations above 100x and a 50% decline pushes them straight into losses.

⚠️ This must be paired with different probability weights, or it will be misused: DRAM/NAND spot prices have historically fallen 50–70% (2019, 2022–23); HDD, by contrast, is a three-way oligopoly on contract pricing with no spot market, and −50% has no historical precedent. MU/SNDK are "high probability, low slope"; WDC/STX are "low probability, high convexity." In addition, the table's fixed-cost assumption only holds near-term — MU's TTM capex is $25.3 billion and still accelerating, and rising depreciation over the next 2–3 years will erode that cost cushion.

5.12 HPE Hewlett Packard Enterprise — Avoid (but it is the most informative name of all)

FQ3 revenue +34% YoY to $12.2B, adjusted EPS $1.11, and it raised both FY26 and FY27 guidance — and the result was two straight down sessions, another −4.48% Friday to close at $52.00. The reason is supply tightness and gross-margin pressure: memory contract prices are up more than 90% in a year, which is profit for the storage makers and BOM cost for the server/PC assemblers.

• Its value is as an independent confirmation of the theme: HPE's gross-margin complaints are, in reverse, proof that storage price increases are real and are being passed downstream.
• Conclusion: avoid HPE itself, and treat it as a confirming indicator for storage price increases.

---

6. Bearish / Avoid List

6.1 A Sector-Level Signal That Should Not Be Ignored: 5/5 "earnings equals sell"

Of the six equipment and interconnect names, the five that reported in the past 6 weeks all fell the day after earnings and all remain below their pre-earnings closing price:

⚠️ Sample disclosure: of the six, LRCX's day-after reaction was not obtained for this piece, so "5/5" means 5 out of 5 among the five for which data was obtained, not the full six-name sample; and LRCX happens to be this piece's top equipment pick, so this gap runs in a direction favorable to my own conclusion and therefore must be flagged explicitly.

Most of these prints were themselves beats with raised guidance. The only possible conclusion is: the sector had already been re-priced ahead of earnings, and good news no longer produces a new expectation gap. This, HPE's "beat yet down two days running," and the deep drawdowns in §1.1② of this piece are three facets of the same phenomenon. Any assumption that "good earnings will drive the sector higher" has been falsified 5 times in the past 6 weeks.

---

7. Within-Theme Rankings

7.1 Storage (re-ranked by the strength of supply-side evidence, not by Friday's gains)

These must be accounted for separately: "fundamental solidity" and "cycle position" are two orthogonal rankings, and compressing them into one order is misleading.

A. By fundamental solidity (earnings quality + cash conversion + balance sheet + downside resilience)

B. By cycle position (have the leading indicators turned?) — the ranking is almost the reverse

C. Ranked by supply-side evidence (⚠️ this table ranks only on supply-side evidence and does not include the weights from tables A and B)

⚠️ The three tables cannot be merged into one ranking; please read them separately: by A (fundamental solidity) MU is 1st; by B (cycle position) SNDK turned first; by C (supply-side evidence) SNDK is 1st. Table C puts SNDK ahead of MU on the single piece of supply-side evidence that NAND capacity is not being added, and does not deduct anything for the two turn signals in table B — this is a deliberate dimension-by-dimension presentation, not a composite score. If table B were included, the order of SNDK and MU would reverse. The three kinds of fragility should likewise not be compressed into one ranking: most distorted reported figures → WDC; earliest cycle inflection → SNDK; thinnest balance-sheet buffer → STX.

7.2 Semiconductor Equipment

All three sit at the 89th–93rd percentiles of the past 4 years, and at record-high earnings — this is not a low-P/E trap, it is the opposite formation: the price has already discounted all of the future growth.

7.3 AI Interconnect / Optical Modules

---

8. Open-Verification Signals (for Tuesday 9/8, 09:30 ET)

8.1 Pre-Market (04:00–09:30 ET)

1. Distinguish gap-and-go from gap-fill: Asia has already surged today (Korea +4.61%), so part of any gap-up in US stocks on Tuesday is just the overnight handoff, not new buying.
2. Pre-market liquidity must be measured in notional amount, not share count: MU is $1,016 and SNDK $1,740, so a small share count is already a large sum of money, and they cannot be compared with INTC ($95.80) on the same share-count threshold.
3. Tuesday's open must digest roughly 89.5 hours of news since 9/4 16:00 ET all at once, so the gaps will be larger than on an ordinary trading day.
4. ⚠️ This piece has a blind spot of about 25 hours: the data runs through 9/7 08:15 ET, and today's European close, the Labor Day futures segment through the 13:00 ET close, and Monday-evening company announcements are all invisible to it. You must sweep that window yourself before the open.

8.2 Intraday (09:30–10:00 ET) — This Piece's Core Criteria

First, the full 9/4 volume table (completed, including the storage and equipment names I previously omitted). Baseline = 10-day average volume; the market baseline SPY is 0.97x, showing overall market volume was normal that day, with no "general pre-holiday volume drought."

This table corrects two biases of mine from earlier:

1. The storage names (SNDK 1.51x, MU 1.48x) and KLAC (1.56x) rose on above-average volume, not on a volume-less bounce. I had previously listed only the low-volume figures for MRVL/AMAT/LRCX to support the "oversold bounce" conclusion, and had not listed the very names I most wanted to argue about — the criterion was being used selectively, and that is corrected here.
2. But "above-average volume" does not support "index-driven" either: the typical signature of an index-inclusion day is volume several times average, and SNDK is only 1.51x; meanwhile PANW (0.79x) and ANET (0.82x), added in the same batch, actually traded below average volume. Neither direction of explanation fully holds, which is exactly why I said in §0.3 that this "cannot be attributed to a single driver."
3. ⚠️ The choice of baseline changes the conclusion: on a 50-day average, LRCX is 0.79x, AMAT 0.67x and MRVL 0.80x (recent sector quiet has pulled the 10-day average down). Whenever a volume ratio is quoted the baseline must be stated, and the two bases must not be mixed.

Tuesday's criteria:

1. Volume and price move must be read together. If a leading name rises on below-10-day-average volume, treat it as an oversold bounce and do not upgrade it to a trend.
2. Whether SNDK and the other three S&P 100 additions (DELL/PANW/ANET) move in the same direction. If only SNDK rises while the other three do not, the market is trading NAND fundamentals; if all four move together, it is still index money. This cleanly separates the two drivers.
3. Whether the SOXX-IGV spread persists. If the two move back in the same direction, Friday's divergence was a one-off flow.
4. Whether NVDA resumes leadership. Given that NVDA is +15.13% since 6/30 while equipment is −37%, NVDA strengthening is the continuation of the norm, not the anomaly.

8.3 Options and Sentiment

• VIX: the baseline is only the 9/4 close of 14.53 (the market is closed today so there is no spot reading, see §0.7). If VIX gets above 16 on Tuesday while the S&P is still green, that is a divergence warning; 14.53 itself is at a low level, meaning the market has not priced much risk premium for this week's CPI and two earnings reports.
• IV on ADBE / ORCL: both report after the close on 9/10, so IV keeps rising from Tuesday through Thursday, with a violent IV crush after the prints (ORCL implies ±11.2%). Watch the impact of the post-earnings IV crush on options positions.

8.4 Risk Signals

• Giving back the index event: if SNDK gaps up on Tuesday and closes red, that is the early form of "selling the fact," and there is one more chance around the 9/21 effective date.
• A lone name rising with no sector follow-through: see items 2 and 3 in §8.2.
• Futures structure: this morning Dow futures at −0.54% are clearly weaker than Nasdaq 100 futures at −0.09%; if that reverses on Tuesday to Dow-strong, Nasdaq-weak, the rate-hike trade is overpowering the AI trade.
• Waiting ahead of CPI: the 9/11 CPI determines the 9/16 FOMC, so from Wednesday onward volumes will most likely shrink into a wait-and-see, and if there is going to be a move, it most likely comes Tuesday through Wednesday.
• Geopolitical tail: if the Hormuz "restricted zone" is formally declared, an oil-price jump would simultaneously hit growth valuations and deepen rate-hike expectations, the one scenario that could invalidate most of the themes at once.

---

9. Final Conclusions

① The 5 Stocks Most Worth Watching Today

② The 3 Strongest Themes Today

③ What to Avoid Today

1. High-beta interconnect names bouncing without volume (MRVL / ALAB / CRDO): Friday volumes were below or only slightly above average, normalized P/Es are about 190–200x, and CRDO already shows negative operating leverage.
2. Chasing semiconductor equipment names (KLAC / AMAT): P/Es at the 92nd / 89th percentiles of the past 4 years with earnings at record highs, and 5 out of 5 earnings-equals-sell over the past 6 weeks.
3. Treating a "forward P/E of 7x" as a cheap buy (MU): that multiple is based on the FY2027 consensus and is 20.4x FY2025 actual EPS; on a realized basis MU is 22.94x, at the 86.7th percentile of 5 years, and normalized on median ROE it is 45–88x. This is "peak-earnings pricing," not a discount.
4. The apparently cheap WDC: 70% of TTM net income is a mark-to-market gain on a SanDisk stake that has been liquidated, so the true P/E is about 45.7x, not 19.26; it is the only one of the four whose forward P/E exceeds its TTM P/E, and consensus already embeds an EPS decline.
5. Per-seat-billing enterprise software (ADBE / WDAY / NOW): ⚠️ but expectations are already extremely low, and avoid ≠ short.
6. Downstream hardware assemblers (HPE): memory price increases are a cost to them, not revenue.
7. Companies treating investment gains as operating improvement (GME): $238 million is an eBay mark-to-market gain, and revenue is actually down 18–20%.
8. The three names removed from the S&P 100 (NKE / SPG / CL): mechanical selling pressure ahead of 9/21.

④ The Final One-Sentence Judgment

To this day there is no single driver that can explain the full extent of Friday's big green candle. The index inclusion definitely happened, but the other three names in the same batch rose only 0.40%–1.50%; Astra did not move storage stocks at all on the day it was released; rates moved just 3 basis points that day. When attribution is unclear, the reasonable posture is to downgrade rather than add. The longer time frame does not support a "leadership switch" either: since 6/30, NVDA is +15.13% while AMAT is −37.11%, and leadership has never changed hands; the equipment and interconnect group has had 5 earnings prints and 5 next-day declines over the past 6 weeks, and good news no longer produces an expectation gap. This piece therefore refuses to chase anything: the highest-attention name is ORCL, which has its own earnings catalyst, the hardest primary-source supply evidence is in NAND (SNDK, though its own two leading indicators have already turned), and what actually decides this week is the 9/10 ORCL and ADBE earnings and the 9/11 CPI.

---

Data-Sourcing and Pipeline Notes (Internal)

1. yfinance / Yahoo completely unusable

• Both query1 and query2 returned HTTP 429 on /v8/finance/chart; changing UA, changing host and retrying with intervals still returned 429, so this is judged to be IP-level rate limiting on this machine (not leftover processes). Two sub-agents independently reproduced the same failure (YFRateLimitError).
• Backup channels that failed in testing:
  • stockanalysis.com/api/symbol/.../history direct → HTTP 400 (path has changed)
  • stooq.com/q/d/l/ → JS PoW anti-scraping. This time I reproduced its SHA-256 four-leading-zero challenge in Python and successfully passed /__verify, but it still returned Access denied afterwards → an IP-level block; the PoW is not the real gate, so do not waste any more time on stooq in future.
  • CNBC ts-api.cnbc.com/harmony/app/bars/... and bootstrap/chart.json → HTTP 404 (endpoints do not exist)
• The historical-price channel that finally worked: using WebFetch to pull the HTML page at stockanalysis.com/stocks/<t>/history/ directly (?period=3M can be appended). This route works and should be locked in as the backup historical-price approach for this slot.
• Real-time quote channel: CNBC quote.cnbc.com/quote-html-webservice/restQuote/...; note that the root keys returned this time were FormattedQuoteResult → FormattedQuote, not the QuickQuoteResult used historically — carrying over the old key gives a straight KeyError.

2. The most important self-correction of this run (must be recorded)

The core thesis of the first draft (since completely overturned) was: "OpenAI's Astra ignites the AI physical layer, storage and equipment become the new leadership theme, NVDA underperforms," and on that basis MU was listed as the top recommendation.

Three independent pieces of evidence overturned all of it:

1. Sequence: after obtaining the 9/1–9/4 daily bars, I found that on Astra's release day (9/3) MU was only +0.22%, SNDK +0.10% and SOXX +0.15%, while IGV was +3.41%. Software rose more than semiconductors. The timing of the causal chain does not line up.
2. The real catalyst: the reason for SNDK's +11.90% on Friday was the S&P 100 inclusion announcement (effective 9/21); all four added names rose and all three removed names fell that day, and the cross-section cleanly proves it was an index event. The "$93.9B TCV" and "80% gross-margin target" that media broadly cited are both old content from the 8/13 Investor Day, and JPMorgan's $2,250 price target was also a mid-August action.
3. Time scale: from 6/30 → 9/4, NVDA +15.13%, MU −11.93%, AMAT −37.11%; all six equipment and interconnect names are in 30–45% drawdowns. "Leadership rotating to the physical layer" is an illusion produced by a single-day cross-section.

Lesson: with only a single-day cross-section in hand I wrote a self-consistent, elegant causal chain that cited a real product launch — and it was wrong. A single day's sector moves are not enough to support a conclusion like "leadership has switched"; the multi-day series must be obtained first. This time I finished the draft first and only found out after adding the daily bars — the order should have been reversed.

3. Sub-agent output and how much of it was accepted

• bull-bear-debater (storage cycle): high quality. It contributed the back-computed MU forward EPS ($144.0 / FY27 net income $164.9 billion), P/FCF 43.9, P/B 11.39, MU's quarterly capex of 5.39 → 6.39 → 7.82 (SEC 10-Q primary source), SNDK inventory days 135 → 178, TCV per year of $22.8 billion < the run rate of $35.9 billion, the existence of SKHY and its IPO proceeds, the official reason for the TrendForce deceleration being LTA caps, and server DRAM +13–18% vs mobile +8–13%. All of it was accepted and written into the main text.
• fundamentals-analyst (equipment and interconnect): high quality. It contributed the six names' drawdowns and peak dates, 5/5 earnings-equals-sell, Friday volume / 50-day average, the AMAT 10-Q DRAM/NAND end-market split (the most valuable line in this piece), the LRCX 10-K storage mix 42% → 39%, the P/E historical percentiles (KLAC 92% / LRCX 93% / AMAT 89%), the contamination of MRVL's and ALAB's net income, MRVL's Google warrant dilution, and the inconsistency between two stockanalysis pages on forward P/E basis (a 22%–39% difference). All of it was accepted and written into the main text.
• fundamentals-analyst (the four storage names MU/SNDK/WDC/STX): still had not returned at publication time. The financial data for the four storage names in the main text comes from bull-bear-debater's primary-source checks (SEC + stockanalysis) and CNBC quotes, not from that agent's output. The conclusions in §5.1 and §5.3 have not been independently re-checked by that agent; if it returns later and conflicts with the text, the primary-source filings should govern and the text should be re-verified.

4. Known basis traps (hit or nearly hit this time)

• The CNBC Treasury change_pct field is broken again: US2Y returned last 4.374% / prev 4.334% (actually +4bp) while change_pct gave −0.0781%, a self-contradictory sign. All Treasury data in this piece was switched to the official Treasury CSV.
• The read moment for .VIX: CNBC gave last≈15.25 with a timestamp of 09-07T08:11 ET that kept ticking, while previous_day_closing=14.53. Cross-checking confirms 14.53 is the 9/4 close (up 1.47% from 14.32 on 9/3). But 15.25 cannot be trusted: Cboe is closed today, SPX options do not trade, and spot VIX has no calculation inputs; moreover S&P futures are only −0.23%, which on VIX/SPX intraday elasticity corresponds to just +1%–+2%, 2–4x away from +4.9%. A ticking timestamp is a signature of a broken field, not proof of freshness (same as the _DJI/_COMP category). The first draft wrote 15.25 into the main text and set a rule on that basis in §8.3; all of that has been withdrawn, and VIX uses only the 9/4 close.
• Fiscal-year attribution for forward P/E is inconsistent: stockanalysis's statistics page and forecast page differ by 22%–39% on forward P/E for ALAB/CRDO/MRVL, while KLAC/LRCX agree. ALAB's "55.31" is in fact FY2027 (16 months out); using the verifiable FY2026 consensus it is 77x. The main text switched all three to normalized orders of magnitude with a note, and did not use headline values for cross-company comparison.
• Headline P/Es contaminated by one-off items: MRVL (a $1.901 billion non-operating item in one quarter), ALAB (a −48.9% effective tax rate), WDC (FQ4 net income of $3.069 billion > operating income of $1.606 billion). All three headline P/Es have been discarded.
• Distorted P/B denominators: KLAC's P/B of 38.2 is the result of buybacks shrinking equity; STX's equity is just $2.17 billion (negative a year ago), so its P/B of 88.9 is unusable. Both are flagged or avoided in the main text.

5. Conflicts between media figures and primary data (primary data governs)

• 24/7 Wall St. gave NVDA "+2% to $232.37", INTC "+4% to $95.41" and SOXX "+3% to $517.91" — all intraday readings; the actual closes were NVDA $230.36 (+0.84%), INTC $95.80 (+4.51%) and SOXX $519.86 (+3.52%).
• One HPE article said it "fell 5% to $27.92", which is inconsistent with its 52-week range ($64.25/$19.84) and its Friday close of $52.00; that price has been discarded, and only cross-verifiable earnings-basis figures were used.
• Several articles said SNDK's 11.9% happened on "Friday, September 5" — 9/5 was a Saturday; it was actually 9/4.
• Media differ on the nonfarm consensus (Dow Jones +53,000, others +58,000 and +65,000); the main text uses CNBC's Dow Jones survey of +53,000 with the source stated.
• The KOSPI has two readings today, +3.5% (intraday 6,921.17) and +4.61% (close 6,995.39); the main text uses the closing value with a note.

6. Still to be verified (priority for next session)

1. The roughly 34% price gap between SKHY and Seoul-listed 000660.KS — possibly a misquoted ADS ratio. Until verified it must not be used to argue either side, which is why this piece did not put SKHY in the final five.
2. KLA's storage vs foundry end-market split — neither the 10-K nor the 8-K discloses it; it appears only in the unfiled Letter to Shareholders. This determines whether KLAC is a storage beta or a foundry beta.
3. AMAT's NAND equipment mix next quarter (Q4 FY26, roughly mid-November) — this piece's most valuable judgment (DRAM expanding, NAND not) currently rests on a single observation. If the NAND mix jumps next quarter, SNDK's supply-side support collapses on the spot.
4. Whether ALAB's 9/4 move was related to speculation about S&P 500 inclusion, and whether it was ultimately passed over — if so, its +9.75% carries mechanical give-back risk. Media source, unverified.
5. Chinese supply data such as CXMT's 10% DRAM share and YMTC's 14% NAND share all come from secondary financial media and need to be re-checked against primary TrendForce/Counterpoint share tables.

---

⚠️ Risk disclaimer: this list is only a pre-market information review and watchlist and does not constitute investment advice. US equities carry high volatility and pre-market gap risk, and there is IV crush and guidance reversal after earnings; automatically generated content may contain stale information or factual errors, so company disclosures / SEC filings govern and this must not be used directly as a basis for trading.