U.S. Pre-Market Brief | 2026-08-14 (ET) Friday

Coverage window: 2026-08-13 16:00 ET regular-session close → 2026-08-14 08:15 ET (includes 8/13 after-hours earnings and announcements, the overnight Asia/Europe session, and this morning's pre-market). Events that occurred during the 8/13 session but determined that day's close (PPI, SanDisk Investor Day) are separately tagged as "pre-window · background."

Quote conventions:

• Prior close / prior-day change / pre-market price / pre-market change / pre-market volume / 52-week high-low for single stocks and ETFs come from the stockanalysis.com quote API, snapshotted 08:00–08:11 ET (a single snapshot band of 08:00–08:11 ET, not timestamped per ticker — for names like CAPR and IMXI that move violently minute to minute, those 11 minutes are enough to change the percentage, so cite with care). "Prior close" is the 8/13 16:00 ET regular-session closing price; "prior-day %" is the 8/13 regular-session change; "today's pre-market %" is the pre-market change relative to the 8/13 close. These two columns are two different things on two different trading days, and this report lists them separately throughout — never mixed.
• Index levels come from CBOE delayed_quotes (_SPX/_NDX/_RUT/_VIX) and api.nasdaq.com (COMP/NDX), with last_trade_time verified one by one = 2026-08-13T16:14:59, confirming the data is not stale.
• Treasury yields come from the treasury.gov official daily yield-curve CSV, a primary source.
• Earnings figures are taken preferentially from SEC EDGAR originals, company IR call transcripts, and original Investor Day presentation decks (each item is tagged with its file); where originals were unavailable, authoritative media reprints are used and marked "media reading."

The verification work in this report (three items, each of which changed a conclusion):

1. Every pre-market mover named by the media this morning was re-checked ticker by ticker against the quote API for pre-market volume. That re-check demolished an entire "pre-market losers list" — investing.com listed PAYC −5.44%, VRSK −3.62%, MET −2.92%, GEN −2.89% this morning, and their pre-market volumes were just 356, 123, 9, and 126 shares respectively, while the list's "pre-market price" for these four was in fact their prior day's closing price (see section 1.4). Behind MET's "−2.92%" sit 9 shares.
2. Three market-circulated claims about AMAT's earnings conflict with primary sources, and each has been corrected against the SEC 8-K original and the IR transcript (see section 5.1). Among them, "free cash flow collapsed 80% to $210 million" is last quarter's (FQ2) number pinned onto this quarter.
3. This report's first draft challenged SanDisk's "80% gross-margin target," and that challenge was overturned by our own verification; the correction is preserved in the body text. The first draft said "80% is higher than NVIDIA's and requires a special-accounting explanation"; after checking the primary financials: SanDisk's actual GAAP gross margin last quarter was 84.6%, with next-quarter guidance of 83–85% — that "80% long-term target" is 3–5 percentage points BELOW the company's current actual level. It is a discount, not a stretch. The genuinely fragile assumption sits on a different line; see section 5.2. This correction is the single most important self-correction in this report.

Data this report could not obtain: ① July retail sales are released at 08:30 ET, later than every snapshot in this report, so this report writes only the pre-release expectations and scenario mapping — there is no result here, and we do not pretend otherwise (see section 8.1); ② options implied volatility (IV) was not obtained, so the IV-crush judgments in the text are inferred from price action and are not options data; ③ VIX9D / VIX1D retrieval failed (the API silently returned 0.0), so we cannot assess whether front-end volatility still carries no event premium; yesterday's values are not carried over; ④ index futures (@ES/@NQ) quotes were not obtained, so pre-market index direction is proxied by SPY/QQQ/IWM/DIA; ⑤ raw NYSE/Nasdaq advance-decline counts were not obtained, so breadth is proxied by RSP vs SPY; ⑥ current DRAM / NAND spot prices and weekly contract prices were not obtained (they require paid sources TrendForce / DRAMeXchange) — this is the single most important gap in the report, because it is the only independent variable in the memory theme. The only related reading obtained is one second-hand quarterly direction: TrendForce says Q3 CY26 NAND contract prices are up +10–15% quarter over quarter, versus more than +100% in 1H26 (i.e., the second derivative of the increase has turned negative) — that figure is a media/research-house reading, not primary, and quarterly rather than current, so this report lists it only in section 9.1 as a "verification point" and does not use it as the basis for any conclusion; ⑦ the demand side has zero current data in this report — no hyperscaler capex guidance, no AMAT backlog / book-to-bill / deferred revenue, no memory-channel or manufacturer inventory days, and no quantification of new supply from CXMT/YMTC/Samsung/Hynix. Memory and WFE take up most of this report's length, and their demand-side evidence is empty — that has been honestly downgraded in the "persistence" ratings in section 2.

---

0. Today in One Sentence

The one sentence worth taking away today: the market is paying up for a promise that only starts being delivered in 2028, while selling an excellent set of results that can be verified one quarter from now.

For the two events of last night, the fundamental direction and the share-price direction are exactly inverted:

• Applied Materials (AMAT) delivered the cleanest guidance raise of this earnings season so far, and it is −4.97% in the pre-market this morning ($508.00, pre-market volume 354,000 shares, snapshot 08:10 ET). FQ4 revenue guidance $10.25 billion ± $500 million vs consensus $9.54 billion (+7.4%), non-GAAP EPS $4.02 ± $0.20 vs $3.71 (+8.4%). The key is not the midpoint but the range: the low end of revenue is $9.75 billion and the low end of EPS is $3.82 — even the worst case in the guidance is above sell-side consensus. Current-quarter revenue $9.115 billion (+25%, a record), non-GAAP EPS $3.50 (+41%, a record), non-GAAP operating margin 34.0% (a record), the 13th consecutive quarter of year-over-year gross-margin expansion, single-quarter operating cash flow $3.037 billion (an all-time high). There is no flaw to pick in these results, and the stock is falling anyway.
• SanDisk (SNDK) reported not a single realized quarterly number yesterday; it only held an Investor Day and gave an FY2028–FY2030 long-term financial model. It closed 8/13 at +13.67%, is up another +6.91% pre-market this morning ($1,633.69, pre-market volume 1.29 million shares), roughly +21.5% over the two days.

There is a fully defensible explanation for AMAT's decline that has nothing to do with fundamentals: valuation. Primary-source calculation: TTM P/E 46.1x, P/S 13.76x, P/FCF 75.5x, while the respective peaks of those three over the past four years were 24.41x, 6.35x, 31.84x — the current readings are 1.89x, 2.17x, and 2.37x those historical peaks. AMAT's valuation is not inside the past four years' distribution; it is outside it. Over the past 52 weeks the stock rose 183.65%, roughly two-thirds of which came from multiple expansion. A +7.4% guidance raise cannot hold up a multiple that has already doubled. What is falling is the valuation, not the fundamentals — those two things must be stated separately today.

But AMAT does have one genuine flaw, and it is not the same disease that CSCO / COHR had yesterday — that distinction matters. On a nine-month cumulative basis: non-GAAP net income +19% year over year, while free cash flow is −2% ($3.580 billion vs $3.655 billion), so cumulative FCF / non-GAAP net income is only 51%. Yesterday, CSCO's and COHR's cash-flow problems came from inventory (+80% and +79.5% year over year respectively); AMAT's inventory is actually improving — up 13.0% year over year, slower than revenue's +24.8%, with days on hand falling from 141.3 to 131.9. AMAT's leak is in receivables: up from $5.185 billion to $7.691 billion year to date (+33.2%), DSO from 71.9 days to 76.8 days, tying up $2.506 billion in that line alone. The theme "AI hardware's book profits look less and less like cash" got its third sample today, but through a different mechanism — do not write the three companies' causes as one just because the conclusion matches.

⚠️ A self-correction that must go in the body text: our first take on SanDisk was wrong

The first draft read: "an 80% non-GAAP gross margin is higher than NVIDIA's (about 73–75%); for a NAND supplier to set such a target demands a special-accounting explanation." That challenge sounds hard-nosed and fits everyone's common sense that NAND is a commoditized, capital-heavy industry. It is wrong, and wrong in a way the primary financials refute in one second:

That 80% long-term target is 3 to 5 percentage points below the 84.6% the company just delivered and below the 83–85% guided for next quarter. Moreover, non-GAAP and GAAP gross margins are exactly equal (the only adjustment is $6 million of stock-based compensation, 0.07% of revenue) — the joint venture's manufacturing costs have not been moved out of COGS; the accounting maneuver we suspected does not exist.

But the correction must not overshoot either, and a second layer has to be added here: 80% is a discount only when compared with a cycle-peak quarter. Read the table in full — FY2023 was 7.1%, FY2025 was 30.1%, and the simple average across the four-year range is about 36%. So an FY28–30 through-cycle target set at 80% assumes those three years look close to today's peak, not close to the four-year average. The correct formulation is therefore not "it is discounted," but: "it is a discount relative to the current peak, and extremely aggressive relative to any year of the company's own past four other than the peak" — whether it is a discount depends entirely on where FY28–30 sit in the cycle, and that is precisely this report's unknown variable.

Why it can pull this off: the cost base is quasi-fixed. SanDisk's Q4 cost of revenue was $1.383 billion versus $1.403 billion a year earlier — down 1.4% in absolute terms, while revenue over the same period went from $1.901 billion to $8.965 billion. Costs have been compressed to 15.4% of revenue, so gross margin is far less price-sensitive than intuition suggests: if NAND prices halve from here, gross margin is still about 69%.

So where is the genuinely fragile assumption? On the first line of the long-term model, the line almost nobody is discussing: "revenue growth mid-to-high teens, consistent with bit growth" — those words, "consistent with bit growth," are equivalent to assuming zero NAND ASP decline for the next three years. And on the very same page of the very same deck, SanDisk itself writes out NAND's historical mechanism:

"Flash was priced as a commodity: falling ASPs offset rising volumes, holding revenue growth to ~4%" —SanDisk 2026 Investor Day deck, original text

Falling prices offset rising volumes, holding long-run revenue growth to about 4% — that mechanism has run for a decade, and the long-term model requires it to be permanently broken. NAND ASPs have never been flat for three consecutive years historically.

This yields the most important conclusion in this report on the memory theme, and it points in a different direction from what everyone is discussing:

If NAND ASPs decline a cumulative 40% by FY30, then under a constant unit-cost assumption SanDisk's gross margin is still about 83% (the "80% target" is met), while revenue CAGR collapses to 2.9% (the "mid-to-high teens target" fails outright).

In other words, this model contains a failure path in which "gross margin is fully intact while the revenue assumption breaks first." The 80% gross margin that became the headline is the least price-sensitive line in the model — it is protected by a quasi-fixed cost base. What actually carries the risk is the revenue-growth assumption almost nobody is discussing. (The implied cost assumptions and boundary conditions of that calculation are in the table notes of section 5.2; it is a mechanism demonstration, not a forecast.)

The same logic applies to Micron, and there the primary evidence is more direct. A breakdown of Micron's FQ3'26 10-Q original text shows: DRAM +67% quarter over quarter, of which ASP contributed about +60%, while bits actually shipped grew only "low single digits"; NAND +99% quarter over quarter, of which ASP contributed about +85%, with bits up only "mid single digits."

Put differently: the profit base of this memory rally is built almost entirely on price, with almost no volume growth underneath. Micron itself states in its 10-Q risk factors: over the past five fiscal years, annual DRAM ASP swings ranged from +40% to worse than −40%, and NAND from +30% to −50%. That downside figure is a historical fact the company itself acknowledges, not anyone's assumption.

⚠️ Here we must dismantle the second error of this report's first draft, more insidious than the first: I originally treated "AMAT's results show DRAM equipment revenue +52% year over year" as bullish cross-validation of memory-cycle strength. That use gets the direction of the evidence backwards. What AMAT sells is equipment, so its DRAM revenue reflects capacity-expansion decisions customers made 6–12 months ago, and once installed that equipment produces precisely future supply. In other words, DRAM equipment revenue +52% is only a lagging confirmation of the current cycle, and a negative input for future ASPs — it is itself the mechanism by which this report's single independent variable eventually turns down. Using it to argue "memory can keep rising" is self-contradicting. The correct reading is: it proves capacity expansion is happening; it does not prove the upcycle is sustainable.

So the correct reading of the memory theme today is: current profits and cash flows are real (SanDisk Q4 gross margin 84.6%, MU FQ4 guided gross margin 86%, both primary), but all of its elasticity is anchored to one variable — and that variable is exactly the one this report could not obtain a current reading for (DRAM/NAND spot and contract prices require paid sources). Until that reading is in hand, any assertion about "whether the cycle has peaked" is unsupported — bullish and bearish alike.

The only theme with real money following it today is memory, and it is globally synchronized. Overnight, Korea's Kospi rose +2.41% to close at 6,977 (briefly breaking above 7,000 intraday), +11.5% on the week, ending a seven-week losing streak; SK Hynix +3.26%, Samsung Electronics +2.43% (⚠️ media reading, and the sources conflict with each other: another report's headline claims Samsung +3% and Hynix +5.8%, which does not match the numbers above; this report could not obtain primary exchange data to verify, so it is flagged as unverified — the Kospi's +2.41% itself is consistent across multiple sources and is usable). In the U.S., nearly every pre-market gain with real volume behind it sits on this chain: SNDK +6.91% (1.29 million shares), WDC +4.10% (217,000 shares), MU +3.15% (1.22 million shares).

But note one counter-example: semiconductor names outside memory — LRCX −0.55%, KLAC +0.02%, ASML +0.27%, TSM −0.07% pre-market, all on volumes under 30,000 shares — AMAT's sharply raised guidance did not spill over to a single equipment peer. What money is buying today is "the price of memory," not "semiconductor capex."

The driver type is "earnings + guidance + index event," not macro — but there is one tier-one event today at 08:30 ET: July retail sales (see section 8.1). The macro backdrop is yesterday's 08:30 ET July PPI: 0.0% month over month versus +0.2% expected, with June revised to −0.1%; +4.7% year over year (prior +5.5%), energy −3.1% being the main driver. That print pushed the S&P to a record close of 7,798.99 (intraday high 7,816.70; the first break above 7,800 was on an intraday basis, and the close did not hold above 7,800 — the media generally conflated these two things this morning), with the 10Y Treasury 4.68%→4.63% (−5bp) and the 2Y 4.20%→4.15% (−5bp). Note that this market's rate direction runs opposite to the past two years' intuition: current pricing for the September 16 FOMC is roughly 71% hold / 29% a 25bp hike, with essentially no cut priced in.

One-sentence conclusion: today does not lack themes; it lacks themes that have not already been front-run. The memory chain rallied all day yesterday and is gapping again this morning, so chasing in means buying the second gap; AMAT has the strongest fundamentals and the weakest price of the day, but its valuation percentile says this decline is not necessarily over. The genuinely low-risk action today is not to buy, but to wait for the first hour's volume confirmation.

---

1. News Overview

1.1 Event table (in-window: 8/13 16:00 ET → 8/14 08:15 ET)

1.2 Index pre-market (snapshot 08:10–08:11 ET; index futures quotes were not obtained this time, so the following are ETF proxies)

How to read it: all four broad benchmarks are within ±0.3% pre-market, so at the index level there is no direction today. QQQ's slight edge over DIA is consistent with strength in the memory chain, but the magnitude is too small and DIA's pre-market volume is only 10,000 shares, not enough to support a "tech leadership" characterization. All the volatility is at the single-stock level.

1.3 Yesterday's (8/13) closing picture and cross-asset

Indices (CBOE / Nasdaq official APIs, last_trade_time all 2026-08-13T16:14:59, verified not stale)

⚠️ VIX9D and VIX1D retrieval failed this time (the CBOE API silently returned 0.0 rather than an error), so this report cannot assess whether front-end volatility still carries no event premium. Yesterday's recap observation that "VIX9D was 11.09, the front end has no cushion left" cannot be carried forward or verified today; do not apply yesterday's values.

Treasuries (treasury.gov official CSV, primary)

The whole curve shifted down 3–6bp, with the belly (5Y) falling most — a textbook "softer inflation print" reaction, not a risk-off move.

Sectors (SPDR series, 8/13 regular session)

Breadth: equal-weight RSP +0.75% vs cap-weighted SPY +0.70%. ⚠️ That 5bp difference is within noise — this report just rejected a similar inference for QQQ vs DIA in section 1.2 on "the magnitude is too small," and the same yardstick must be applied consistently, so we likewise do not characterize breadth as "healthy" here. The correct statement is: equal-weight and cap-weighted were essentially in line; this report did not obtain raw NYSE advance-decline counts and cannot judge breadth.

Yesterday's two main lines: ① software IGV +3.10% — MNDY +9.70%, MDB +7.87%, NET +6.22%, DDOG +4.70%, PLTR +4.66%, CRM +4.16%, driven by a softer PPI (long-duration assets benefit) plus several software earnings beats; ② XLC +2.07% — mainly contributed by NFLX +5.43% (Ackman's stake) and META +2.78%.

Cross-asset (ETF proxies, 8/13 close → this morning's pre-market): USO (crude) −1.78% → +0.15%; GLD (gold) −1.47% → +0.56%; UUP (dollar) −0.07% → −0.18%; IBIT (bitcoin) −0.03% → −0.86%; TLT (long bonds) +0.58% → −0.35%. Cross-asset offers no directional information this morning; the only mild anomaly is bitcoin's relative weakness.

1.4 ⚠️ Falsifying this morning's "pre-market losers list" — this section decides whether you go short a decline that does not exist

This morning investing.com's pre-market losers list showed: PAYC −5.44% ($211.47), AMAT −4.78%, VRSK −3.62% ($179.20), MET −2.92% ($94.62), GEN −2.89% ($28.72). Checked one by one:

Two conclusions, and note that they differ in strength:

1. (Weaker, already narrowed by back-solving) The first draft said "the whole list treats the prior day's close as the pre-market price." After back-solving each name, that mechanism holds precisely only for PAYC; the list prices for VRSK / MET / GEN do not reconcile with any verifiable price on any day, and the cause is unknown. So the correct statement is: the list's price source is unknown and cannot be reproduced under any known convention, therefore it is unusable — not "it is uniformly one day stale." This correction is itself a reminder: when falsifying someone else's data, your own falsification mechanism must also be back-solved case by case; do not extrapolate from n=1 to n=4.
2. (Very strong, unaffected by the above) The pre-market volumes of these four are 356 / 123 / 9 / 126 shares — at that scale, any percentage is not a price, it is an odd-lot print. This point stands independently and is this section's real conclusion.

This report therefore sets and applies throughout one convention: for stocks with pre-market volume < 50,000 shares, the pre-market change is not used as a signal, only as a note. This convention is applied evenhandedly to both up and down moves — as section 1.5 below shows, it likewise rejects a batch of upside readings.

1.5 Applying the same yardstick to the upside: software's "give-back" this morning does not hold either

After IGV's +3.10% yesterday, software names are broadly a touch red pre-market this morning, which is easy to write up as "money rotating out." Tested against the same volume convention:

Six of the eight software names have pre-market volume under 50,000 shares; the three that clear the bar all move less than 0.5%. Conclusion: software has neither given back nor extended this morning — it simply has not started trading. Yesterday's 3.10% was real; this morning's minus signs are not. Any "software was a one-day wonder" claim has no evidentiary support this morning.

---

2. Strongest Themes, Descending

---

3. Overall Single-Stock Strength Ranking

Ranking basis: descending by total score from the section 4 scoring model. Every pre-market change is tagged with pre-market volume; those under 50,000 shares are marked ⚠️ and their pre-market readings are not used as signals.

3.1 Bullish / long side

3.2 Bearish / short side

---

4. Single-Stock Scoring Model (100 points)

Scoring items: source authority 0–15 / catalyst directness 0–20 / earnings elasticity 0–15 / moat and fundamentals 0–15 / expectation gap 0–10 / catalyst persistence 0–10 / trading characteristics 0–10 / risk deduction 0 to −15.

---

5. Detailed Analysis of Top Names

5.1 AMAT (Applied Materials) — the strongest fundamentals and the weakest price of the day

Related news: 2026-08-13 16:05 ET, SEC 8-K EX-99.1¹ + IR earnings-call transcript PDF².

First, three market-circulated claims corrected (all refuted by primary sources):

⚠️ Definitional trap: the press release's geographic table says "China 28%" as a share of total revenue; the CFO transcript's "China represented 26%" is a share of SSG+AGS revenue. Both are correct with different denominators, and they must not be mixed when cited.

Primary current-quarter figures: revenue $9.115 billion (+25%, a record) vs guidance midpoint $8.950 billion and consensus of about $9.00 billion; non-GAAP EPS $3.50 (+41%, a record) vs $3.40 expected; GAAP EPS $3.17; non-GAAP gross margin 50.4% (+1.5pt, the 13th consecutive quarter of expansion); non-GAAP operating margin 34.0% (+3.3pt, a record). By segment: Semiconductor Systems revenue $7.040 billion (+26.5%), non-GAAP operating margin 38.0% (+4.8pt); AGS $1.781 billion (+21.7%); Other (including Display) operating loss widened from −$4 million to −$118 million, the only clearly deteriorating item this quarter. SSG end-market mix: Foundry/Logic 67% (69% a year ago), DRAM 26% (22% a year ago), Flash 7% (9% a year ago); DRAM revenue (including HBM packaging) +52% year over year, a record.

FQ4 guidance (the most important item in this report):

The key is not the midpoint but the low end: revenue's low end of $9.75 billion > consensus $9.54 billion (+2.2%), EPS's low end of $3.82 > consensus $3.71 (+3.0%). There has not been a second guide this earnings season where even the worst case beats consensus.

⚠️ This conclusion has a fragility that must be disclosed, because it is the most reused pivot in the report. The two benchmark figures in the table above come from two different estimate databases (revenue from LSEG, EPS from FactSet), and this report could not obtain their as-of timestamps. After results, the sell side immediately re-models, so the "consensus" visible on the morning of 8/14 is not the same number as the one from 8/13 pre-market. Likewise, for "current-quarter EPS $3.50 vs $3.40 expected," this report obtained neither a single source nor a timestamp. So the claim that "even the low end beats consensus" is sensitive to the estimate source and timing, and should be used as "direction credible, magnitude unverified," not as a precise conclusion. Breakdown (CFO transcript): SSG ~$7.9 billion (+62% year over year), AGS ~$1.84 billion (+22%), Other ~$510 million, non-GAAP gross margin ~50.4%, non-GAAP opex ~$1.58 billion, tax rate ~11%. Back-solving implies a non-GAAP operating margin of 35.0%, another 1pt above Q3 — the guidance is internally consistent.

⚠️ Three places that must be discounted:

1. "+51% revenue / +85% EPS year over year" is a low-base game. FQ4'25 revenue was only $6.800 billion (−3% year over year), with a 25.2% operating margin and non-GAAP EPS of $2.17 — AMAT's worst quarter in recent years. The clean reading is +12.5% sequentially.
2. Gross margin stopped expanding sequentially (50.4% → 50.4%). Under the "13 consecutive quarters of expansion" narrative, this is the first tap of the brakes. The CFO attributed it to recent ramp costs, but that statement comes from a third-party transcript and is flagged as unverified. Whether this is a one-off ramp or multi-year dilution from doubling capacity investment, the two explanations have opposite implications for the 2027 margin model.
3. The tax rate rises from ~11% to ~13% in 2027 (about a 2pt EPS headwind); FY27Q1 is a 14-week quarter, and a sequential model that does not adjust for it will produce wrong conclusions.

Cash-flow quality (the most important negative finding on AMAT): single-quarter FCF of $2.330 billion at an 83% conversion rate is healthy. But on a nine-month cumulative basis: non-GAAP net income +19%, FCF −2% ($3.580 billion vs $3.655 billion), a cumulative conversion rate of only 51%. The gap is almost entirely in receivables — up from $5.185 billion to $7.691 billion year to date (+33.2%), with DSO deteriorating from 71.9 days to 76.8 days, tying up $2.506 billion in that line alone. At the same time, 9M capex is +34.8% year over year, and the CFO explicitly said the company will "double quarterly systems output by 2028" and "plan the next capacity expansion before 2030" — this is a multi-year structural drag. Another signal: 9M buybacks of $1.177 billion, −71% year over year, with shareholder returns at only 65% of 9M FCF, below the company's own stated "80–100% of FCF" target. After the stock quadrupled, the company itself sharply decelerated buybacks.

Fundamental verification (primary calculation): ROE ≈ 40.3%, ROA ≈ 23.2%; net cash $2.69 billion; debt/equity 0.26, current ratio 2.42, interest coverage 33x. Revenue +24.8% while GAAP opex rose only +13.7% and headcount +7.8% — revenue growth is 3.2x headcount growth; operating leverage is rock solid. The balance sheet imposes no constraint whatsoever.

Valuation (the real reason for today's decline):

⚠️ Convention note: the "1.89 / 2.17 / 2.37x" figures cited elsewhere in this report are all on the 8/13 close basis; using this morning's pre-market price they are 1.79 / 2.06 / 2.25x. The two columns must not be mixed.

The four-year P/E center is roughly 15–22x. The current 44–46x is not inside the historical distribution; it is outside it. The 52-week gain is +183.65%, about two-thirds from multiple expansion. The stock has broken below its 50-day moving average of $558.76,200-day moving average $387.30 is still 31% lower.

⚠️ But an alluring miscalculation must be guarded against here, one this report's first draft made: do not use point estimates like "44x back to 24x = another 45% downside." That arithmetic implicitly freezes EPS, while this report's core argument is precisely that AMAT's earnings are accelerating — annualizing the FQ4 guided non-GAAP EPS of $4.02 (about $16.08), the current $508 price implies a forward P/E of only 31.6x, not 44x. A report that says "earnings are being revised up" on one hand and computes downside with "frozen earnings" on the other is self-contradictory. The only defensible formulation is qualitative: the multiple remains materially above its own historical range, and that calculation is extremely sensitive to the EPS convention used (43.8x on TTM vs 31.6x on a forward basis), so this report gives no specific downside percentage.

Pre-market and technicals: 8/13 opened $548.81 → intraday high $563.76 → closed $534.54 (−2.48%), so the front-running happened before the results and had already unwound the same day; after hours ~$519.70 → pre-market 08:10 ET $508.00 (−4.97%, 354,000 shares). Cumulatively −7.32% from the 8/12 close of $548.15. It is now −31.3% from the 52-week high of $739.67. The sell side split clearly this morning: J.P. Morgan $515→$660 (raised), BofA $720→$650 (cut).

Final judgment: watch closely, not priority deep-dive. Fundamentals are impeccable and even the low end of guidance beats, but this decline is driven by multiple compression, and the multiple is still materially above the top of its own four-year range. "Strong fundamentals" is not a reason to conclude "the decline is over," and that is the easiest mistake to make today.

5.2 SNDK (SanDisk) — the number that became the headline is in fact the sturdiest line in the model

Related news: 2026-08-13 SanDisk 2026 Investor Day (original 98-page deck³). FY2028–FY2030 long-term model: revenue growth mid-to-high teens "consistent with bit growth," non-GAAP gross margin about 80%, non-GAAP operating margin about 75%, opex about 5% of revenue, adjusted FCF margin about 50%, capital intensity mid-single-digit % of revenue, with a commitment to return 100% of excess cash.

① First, correcting this report's mistaken first-draft challenge

The first draft said "an 80% gross margin is higher than NVIDIA's, requires a special-accounting explanation, and we suspect JV costs have been moved out of COGS." After checking the primary financials, every layer of that challenge fails:

How it manages this: the cost base is quasi-fixed. Q4 cost of revenue was $1.383 billion versus $1.403 billion a year earlier — down 1.4% in absolute terms, while revenue over the same period went from $1.901 billion to $8.965 billion (+372%). Full-year COGS was $5.776 billion vs $5.143 billion, up just 12.3%, against revenue up 175%. Unit-cost declines from BiCS node migration offset bit growth.

This leads to a counterintuitive but important corollary — gross margin is far less price-sensitive than everyone's intuition (assuming bits and unit costs unchanged):

⚠️ This table is a constant-cost extrapolation, and it excludes three things that inevitably appear in a real memory downcycle: ① inventory write-downs; ② underutilization charges — and SanDisk in particular has a clause that makes this unavoidable (see ④ below: it must pay half of the JV's fixed costs whether or not it takes output; when it deliberately lowered utilization in the first nine months of FY26, $11 million went straight into COGS); ③ new depreciation — the company itself plans to raise capital intensity from 2.2% to mid-single-digit. Reference point: Micron's FY2023 gross margin was −9.11%; that is what a real downcycle looks like.

So this table can only support a far more restrained statement: under a constant-cost assumption, gross margin is the least price-sensitive item in the model — not that "gross margin is the sturdiest line."

Under the above assumptions, if NAND prices halve from here, gross margin is still about 69%. That is the mathematical basis for "the 80% target is achievable provided the cycle does not deteriorate severely."

② So where is the genuinely fragile assumption? On the first line of the long-term model

"Revenue growth mid-to-high teens, consistent with bit growth" — those words, "consistent with bit growth," are equivalent to assuming zero NAND ASP decline over the next three years.

And in the very same deck, SanDisk itself writes out NAND's historical mechanism:

"Flash was priced as a commodity: falling ASPs offset rising volumes, holding revenue growth to ~4%"

Falling prices offset rising volumes, holding long-run revenue growth to about 4%. That mechanism has run for a decade, and the long-term model requires it to be permanently broken. NAND ASPs have never been flat for three consecutive years historically.

Solving the two constraints together yields the most important conclusion in this report:

⚠️ This table has two implicit assumptions that must be spelled out, or readers will overestimate its certainty: ① back-solving each row holds FY30 COGS constant at about $7.86 billion (vs FY26's $5.776 billion), i.e. it implies "COGS rises only 36% over four years while bits grow 17% a year," equivalent to assuming unit costs fall about 8% per year for four straight years; if unit-cost declines fall short, the 83.3% in the −40% scenario does not hold either. ② The revenue CAGR is computed over four compounding periods from FY26→FY30 (1.17 × 0.6^0.25 − 1 = +2.97%); on three periods it would be −1.3%. This table is a mechanism demonstration, not a forecast.

Under the above assumptions: a cumulative 40% ASP decline still leaves gross margin at 83% (the "80% target" met), while revenue CAGR collapses to 2.9% (the "mid-to-high teens" target fails outright). That means the model contains a failure path in which "gross margin is fully intact but the revenue assumption breaks first."

In other words: the 80% gross margin that became the headline is the least price-sensitive line in the model (protected by the cost base); what actually carries the risk is the revenue-growth assumption almost nobody is discussing.

③ The NBM long-term agreements: real substance, but coverage and counterparty risk must be stated clearly

Primary figures (from the deck): 10 contracts / 8 customers; TCV $93.9 billion, RPO $91.1 billion (both at floor pricing); third-party financial guarantees $16.5 billion (of which $2.5 billion is already in the cash balance); weighted duration 4+ years; covering about half of FY27 bits and about two-thirds of FY28. Pricing is fixed plus floating, with the floating portion carrying floors and ceilings.

The supportive side: TCV $93.9 billion ÷ duration of 4+ years ≈ $18.8–23.5 billion a year; NBM covering about half of FY27 bits ⇒ implied FY27 company-wide revenue of about $37.6–47.0 billion, while annualizing FQ1'27 guidance gives $42.0 billion, which falls inside that range.

⚠️ This report's first draft wrote it up as "two independent paths landing on the same number, showing the floor price is anchored near today's price" — that is over-reading, and it has been corrected. $42.0 billion merely sits in the middle of a wide $37.6–47.0 billion range, and with a range width of nearly 25%, falling inside it does not constitute corroboration. The more fundamental problem: bit coverage ≠ revenue coverage — NBM is priced at floor, so if the floor is below the current price, the same bits correspond to less revenue, and the back-solved company-wide revenue is systematically overstated. So the inference "the floor is anchored near the current price" depends on what it is trying to prove — it is circular, and this report withdraws it. All that can be established is: the scale of NBM is real (RPO $91.1 billion), but the absolute level of the floor remains unknown.

Three reservations that must be stated alongside it:

1. Not one of the 8 customers' identities has been disclosed. Two-thirds of FY28 bits are concentrated in 8 unnamed counterparties.
2. Guarantees $16.5 billion ÷ TCV $93.9 billion = 17.6%. If NAND spot prices fall sharply, the economic loss to customers of taking delivery at the floor could far exceed the cost of walking away — default becomes the rational choice, and SanDisk collects the guarantee while losing the revenue. The deck also concedes that the guarantee amount stays constant over the contract term while RPO amortizes down, i.e. coverage is lowest early in the contract, exactly when price risk is greatest.
3. The absolute floor price has never been disclosed. "Even at floor pricing, attractive financials" is a qualitative statement that cannot be checked.

④ Two structural risks that are easy to overlook

• The JV's fixed-cost clause is a downside amplifier. In black and white in the 10-Q: "obligated to pay for half of Flash Ventures' fixed costs regardless of the output the Company chooses to purchase." Whether or not it takes the output, it pays half the fixed costs. When demand weakens, SanDisk cannot protect its P&L by cutting production — it must either dump at low prices or pay for idleness. When it deliberately lowered utilization in the first nine months of FY26, $11 million went straight into COGS. This clause is leverage in an upcycle and a noose in a downcycle.
• Capital intensity will more than double. FY26 actual capital intensity was just 2.2% ((own capex $177 million + Flash Ventures net $275 million) ÷ revenue $20.248 billion), while the long-term model itself sets it at mid-single-digit %. The company itself plans to more than double capital intensity — that is capacity expansion, and capacity expansion is the enemy of price. There are also off-balance-sheet commitments: $6.543 billion related to Flash Ventures, purchase obligations $4.970 billion; plus $856 million of prepaid facility depreciation to Kioxia not yet released (future depreciation runs through the P&L but not through cash, systematically flattering future FCF conversion).

⑤ Valuation and price position

⚠️ The share-count convention must be settled first: SanDisk has 149 million shares outstanding and 155 million diluted weighted-average shares, and market caps computed from the two differ by about 4%. This report uses shares outstanding (the first draft mistakenly used the diluted count; corrected): $1,528.11 × 149 million = market cap of about $227.7 billion; zero interest-bearing debt, cash $4.762 billion.

"Low P/E + extremely high P/B" is the textbook combination for a cycle top — the denominator of P/E is peak earnings, and the denominator of P/B is book value, which does not lie.

One fact that must be presented in parallel (it supports both sides): the intraday high of $2,354.39 was set on 6/22, before both the 8/5 Q4 blowout and the 8/13 Investor Day; the two strongest fundamental events both landed after the top, and the stock is still 30.6% below that high. The bullish reading: the market has already pre-priced a cycle collapse, and 8.5x forward P/E is a mispricing. The bearish reading: the market is using price to say it does not believe this model. Neither reading can be falsified today.

Two more worth noting: ① FY26 spent $4.524 billion on buybacks, yet shares outstanding rose from 146 million to 149 million and the diluted weighted average from 145 million to 155 million — $4.5 billion spent and the share count went up; ② the deck's "TAM from about $60 billion → CY26 $300 billion+ → CY27 nearly $500 billion" carries the source note "Source: Sandisk Internal estimates," i.e. the numerator of the entire narrative is self-attested; ③ as of verification, there is no 8-K on EDGAR corresponding to this Investor Day, so the FY28–30 targets currently live only in a PPT on the IR site and in no SEC filing (flagged as unverified; the company may file later).

Final judgment: watch only. The news is real, the money is real (1.29 million pre-market shares, the highest in the memory chain), and the gross-margin target checks out as credible — but the price has already fully expressed this news (+21.5% over two days), and the revenue assumption underpinning the long-term model contradicts the historical mechanism the company itself printed on the same PPT page. The price information in this theme has been fully expressed by yesterday's and this morning's two gaps; further incremental information requires volume confirmation.

5.3 RDDT (Reddit) — the most certain mechanism, and also the easiest "buy the announcement, sell the effective date"

Related news: 2026-08-13 after the close, S&P Dow Jones Indices official announcement⁵ — Reddit will join the S&P 500 before the open on 8/18, replacing AvalonBay (AVB). AVB is being removed because it is being acquired by fellow index member Equity Residential (EQR); the merged entity will be renamed Vivmark Residential (VMRK) and remain in the index.

Catalyst logic: this is the hardest piece of logic in the report — it depends on no forecast, index funds must buy, and the timing is certain (8/17–8/18 close rebalance). Reddit's Q2 revenue, reported 7/30, was +61% year over year (media reading, this report did not obtain the SEC original; flagged as unverified), so the fundamentals merit inclusion.

⚠️ But note the shape of the event: the passive buying from index inclusion happens before 8/18, and the announcement-day move usually front-loads that demand. Historically the pattern of "big gain on announcement day → pullback around the effective date" is common for this event type. This morning's pre-market is $175.10 (+10.74%, pre-market volume 631,494 shares) — the volume is real, but the level you are buying has already run ahead of the passive money.

The 52-week range is $119.27–$282.95; at $175.10 the stock sits in the lower-middle of that range, −38% from the high.

Final judgment: watch only. The news grade is S and the mechanism is certain, but the +10.74% pre-market move has already eaten the easiest part. The verification point is whether it holds after the 8/18 effective date, not chasing it today.

5.4 MU (Micron) — it is not today's leader; it is the weakest of the four memory names

Related news: MU had no SEC filings on 8/13–8/14 and no substantive catalyst of its own. Its most recent 8-K is the FQ3 results from 6/24, and its most recent Form 4 is from 7/28.

Its only company-specific news is the $250 million AI venture fund (Micron Ventures Paradigm Fund) established on 8/13. Multiple outlets attributed the day's +4.23% to it — this report judges that to be after-the-fact stitching, not a catalyst: $250 million = 0.023% of a $1.07 trillion market cap, = 0.5% of $51.4 billion of TTM operating cash flow. If it were the cause, it could not explain why SNDK rose three times as much the same day.

The decisive cross-sectional evidence — MU was the weakest name in the memory group that day:

This set of data refutes two things at once: ① this was not a broad market/semiconductor rally (SOXX only +0.76%, NVDA +0.54%); ② it was not an MU-specific event either — if the catalyst were at MU, MU should have led rather than trailed. The true source is the SanDisk Investor Day.

⚠️ One premise must be corrected first: MU is not at a high; it is inside a wide range after pulling back from its high.

It is −41.2% from the 6/25 high (intraday basis), still 24.3% below the 52-week high, and below its 50-day moving average of $962.88. The back-to-back gains on 8/12 (+4.92%) and 8/13 (+4.23%) occurred inside a wide trading range — this is not a breakout to new highs. July's single-day swings include 7/30 +18.36%, 7/29 −9.94%, 7/21 +12.17% — this is an extremely high-volatility name, and treating two up days as a trend signal is dangerous.

Primary fundamental figures (FQ3'26, ended 2026-05-28, from SEC 8-K EX-99.1¹⁹ and the 10-Q²⁰): revenue $41.456 billion (+345.7% year over year, +73.8% sequentially); GAAP gross margin 84.6%; GAAP EPS $24.67 / non-GAAP $25.11; operating cash flow $25.388 billion. By segment: DRAM $31.328 billion (75.6%), NAND $9.943 billion (24.0%). FQ4'26 guidance: revenue $50.0 billion ± $1.0 billion, gross margin about 86%, non-GAAP EPS $31.00 ± $1.00.

★ The single most critical piece of primary evidence in this report on the memory theme — growth is almost entirely price, not shipped volume. Breakdown from the 10-Q MD&A original text:

Quarterly revenue rose 74%, while the bits actually sold rose only a single-digit percentage. This is a pure price-increase cycle. It means profit elasticity to a single variable (ASP) is close to 1:1, and MU itself states in its 10-Q risk factors: over the past five fiscal years, annual DRAM ASP changes ranged from +low-40% to −high-40%, and NAND from +low-30% to −low-50%. That downside figure is a historical fact the company itself acknowledges.

Valuation percentiles (self-built, method reproducible: 66 quarters of SEC XBRL equity data + 2,513 trading days of closing prices, aligned by filing date to avoid look-ahead bias):

A P/E percentile of 58.3% versus a P/B percentile of 98.2% — a 40-percentage-point gap — matches directionally the "cycle top = low P/E percentile + high P/B percentile" criterion. But one key stage adjustment must be made:

In 2018 the market used a 2.6x P/E to say outright, "this E is fake and will fall back." In 2026 the market uses a 22x TTM P/E to say, "this E is real and going higher." So the judgment is: we are not in the early stage where "the low-P/E trap has not yet been seen through," but in the late stage where "the de-cyclicalization narrative for a cyclical stock has been fully accepted and priced in" — in 2018 there was at least a 2.6x P/E as a cushion; now there is none.

The counter-fact that must be presented alongside it (otherwise the analysis is biased): the denominator is repairing itself at speed. BVPS rose from $64.25 in FQ2'26 to $89.18 in FQ3'26 (+38.8% in one quarter). If earnings continue at the FQ4 guided level and the price does not move at all, retained earnings alone would take P/B from 10.65 to about 4.49 within a year. This does not negate the danger of a 98th-percentile reading, but it bounds its time window: the danger of a high P/B percentile depends on whether price falls faster than book value rises.

ASP-shock sensitivity (self-built model, explicitly flagged as assumption-based: bits flat, unit costs unchanged, no write-downs):

Even under a −40% ASP shock, annualized EPS is still about $50 and the P/E returns to 19.1x — roughly today's TTM P/E. In other words, the current price already embeds "no severe cyclical downturn"; if the downturn does occur, investors do not get a discount, they get back to today's valuation starting point. This is where "a 6.6x forward P/E looks cheap" most deserves to be tested. The model is generous to bulls: it assumes no inventory write-downs and no underutilization charges, both of which inevitably appear in a real memory downcycle (see MU's FY2023 gross margin of −9.11%).

Quality and cycle position (the side favorable to bulls, also primary): TTM operating cash flow / net income = 1.02x, cash conversion with no padding; net cash swung from −$5.330 billion a year ago to +$19.646 billion, debt fell from $16.141 billion to $6.376 billion, with multiple note series fully prepaid in the first nine months of FY26. This is a textbook management action — preparing the balance sheet for the downcycle at the cycle peak — meaning that even if ASPs halve, MU has no solvency risk, only valuation risk. On supply: FY27 capex guidance is "above mid-$40B," a jump of more than 65% from FY26's roughly $27 billion; but the new bits do not arrive until 2028–29 (own fabs: first DRAM wafers at Idaho fab 1 in mid-2027, fab 2 at end-2028; Singapore HBM packaging in 2027H1; Taiwan Tongluo in mid-2027). SK Hynix's 54 trillion won ($38.1 billion) new fab does not break ground until July 2027, with the first cleanroom in production in June 2029.

Sell side: 46 analysts cover it, rated 32 strong buy / 9 buy / 5 hold / 0 sell; the average target is $1,502, but the range is $361 – $2,200, a dispersion of 6.1x. A 6x spread in targets on a trillion-dollar name covered by 46 analysts is abnormal — it is a direct measure of the absence of consensus on "is this a growth stock or a cyclical at the top." The only target change in the past 30 days is Citi on 8/7 maintaining Buy while cutting its target from $1,400 to $1,150 (−17.9%); all others were reiterations.

Final judgment: watch closely. The industry upcycle has primary cross-validation, the balance sheet is extremely clean, and liquidity is the best in the group; but today it is a follower rather than an initiator (and the weakest of the four), it has no catalyst of its own, its P/B is at the 98.2 percentile, and all of its earnings elasticity is anchored to a variable for which this report could not obtain a current reading. Participating in it is participating in the memory theme, not in MU.

5.5 CAPR (Capricor) — +96% pre-market, but this is a repair, not a breakout

Related news: 2026-08-13 16:05 ET, the company's Q2 results and update announced plans to submit a BLA amendment including 24-month open-label extension data from HOPE-3; "the FDA has indicated it is willing to accept the amendment and will extend the PDUFA action date accordingly upon receipt." This morning Cantor Fitzgerald upgraded from Neutral to Overweight, target $3.50 → $28.

Three price anchors that must be viewed together: $19.70 before the FDA advisory committee's negative vote → $4.21 at the 8/13 close → $8.27 (+96.44%, pre-market volume 17,849,736 shares) pre-market this morning.

In other words, today's +96% has pulled the stock less than halfway out of the post-collapse hole — it is a repair, not a breakout. Equally worth noting is the sell-side behavior itself: Cantor cut its target from $62 to $3.50 just weeks ago and today moved it to $28. This name's sell-side target has traversed $62 → $3.50 → $28 within two months, so no single target price here carries information.

Key wording analysis: "the FDA is willing to accept the amendment" ≠ "the FDA will approve." What it confirms is that the review path has not closed and the PDUFA date will move out; it confirms nothing about efficacy or safety — and the advisory committee has already voted no once.

Final judgment: watch only. The 17.85 million pre-market shares make this one of the most active events in the whole market today, but a +96% gap + a purely binary regulatory outcome + a single asset + a tiny market cap, stacked together, make it unsuitable as a list name. It is listed because it will affect sentiment across the entire DMD / gene-therapy group.

5.6 IMXI (Intermex) — the market has priced only half of this news

Related news: this morning's announcement — the New York State Department of Financial Services (NYDFS) approved Western Union's acquisition of Intermex. But the same update contains a second item: on 2026-08-13, a letter from California's DFPI suspended the approval extension granted on 7/31.

This is today's most textbook case of news being read only halfway. Pre-market $14.21 (+21.45%, pre-market volume 603,878 shares), while the offer price is $16.00/share all cash — even after New York's approval, the market still leaves this deal an approximately 11.2% discount. That discount is not a market error; it corresponds precisely to the unresolved California problem.

Three facts that must be viewed together: ① the offer price is $16.00/share, total consideration about $500 million; ② California's DFPI suspended the approval extension on 8/13; ③ the deal's original outside date was around 2026-05-11 and has passed, and the originally expected Q2 2026 close is also overdue. Reverse termination fees are $27.3 million / $19.8 million.

Final judgment: watch only. Merger arb requires professional modeling of closing probability, and the remaining 11.2% discount is precisely the market's quote for California risk — it is not free money, it is risk compensation.

5.7 GLOB (Globant) — today's cleanest negative, and a structural one

Related news: 8/13 after-hours Q2 results. Revenue $614.4 million, adjusted EPS $1.40 (revenue beat consensus slightly by 0.21%). The key is the guidance: FY26 revenue guidance was cut from $2.462–2.508 billion to $2.428–2.462 billion — the top of the new guidance ($2.462 billion) is exactly the bottom of the old. Q3 target $607–615 million. Management's attribution: weakness in new markets, oil-price pressure hitting travel clients, and lengthening North American decision cycles.

Why this matters more than the numbers themselves: Globant is a direct sample of the structural narrative that AI is disrupting IT outsourcing services. Its AI business is growing (AI Pod ARR rose from $32.8 million in March to $52.8 million), yet overall guidance was still cut — the pace of new AI revenue cannot keep up with the erosion of traditional headcount outsourcing. This is not cyclical softness; it is substitution. Spillover watch list: EPAM, ACN, INFY.

Pre-market and technicals: GLOB closed +5.81% in the 8/13 regular session (it went into its own earnings carrying a gain), and this morning is at $36.03 (−12.10%, pre-market volume 57,553 shares).

⚠️ This is the fourth sample of the same pattern this week: yesterday it was COHR (+8.24% the day before earnings) and CSCO; today it is AMAT (spiked +2.9% intraday then closed −2.48%) and GLOB (+5.81%).

But this observation must be cooled down, not escalated. n=4, all in the same week, selected after the fact, and with no control group (how many samples this week went into earnings carrying a gain and then rose? This report did not count them). The pattern is also not fully homogeneous — AMAT is "spiked intraday before earnings then closed down," which is not the same thing as COHR's / GLOB's "closed green the day before earnings." So the correct formulation is: four same-pattern cases were observed this week, worth continuing to record, but neither the sample size nor the method is sufficient to call it a rule, still less to write it into trading discipline. This report's first draft said "the sample size is already enough to write into trading discipline" — that was overreaching.

Final judgment: avoid / short watch. The negative is structural, the guidance cut is explicit, and pre-market volume clears the bar. The only reservation is that the −12% has already happened, so the short entry is poor.

5.8 INTC (Intel) — fundamentals are improving, but part of today's gain is someone else's money

Related news: Intel completed a $20 billion offering, issuing 210.5 million shares @ $95/share; the original plan was $15 billion, upsized because institutional demand reached $100 billion. Q2 revenue was +25% to $16.1 billion, the best since 2011; ASP about $1,200, +43% year over year. BofA maintains Buy with the target at $160 → $145.

Two sides: positive — $100 billion of subscription demand is an extremely strong confidence signal for a turnaround story, and revenue +25% with ASP +43% are the best operating readings in fifteen years. Negative — this is dilution. The 210.5 million new shares at $95 were already about 10% in the money at the 8/13 close of $104.56; that buy-side cohort's cost is far below that of anyone chasing today, and there is no lockup constraint. Meanwhile BofA cut its target from $160 to $145 while maintaining Buy — the "constructive but with narrowing valuation headroom" combination.

8/13 close $104.56 (+3.58%), this morning pre-market $105.55 (+0.95%, pre-market volume 1,485,977 shares — among the highest in the market).

Final judgment: watch closely. Volume is real and the fundamental inflection is credible, but the $95 cost line is structural overhead supply, and there is no new news today.

5.9 Reconciling yesterday's calls (CSCO / COHR / the AI server chain)

This section is reconciliation, not new recommendations.

Yesterday this series judged that CSCO's decline was "the market repricing the fundamentals," and explicitly wrote that "the decline may only be beginning"; while COHR's decline merely "erased its own one-day front-running, with net pricing impact close to zero."

COHR is at $332.29 (+1.55%, 195,957 shares) pre-market this morning, back above its 8/11 close. That validates the method: after-hours earnings must be benchmarked against the pre-release closing price, otherwise the unwinding of front-running gets misread as a negative. CSCO is −0.35% this morning, with the decline paused but not repaired, so the conclusion stands: avoid.

The AI server chain: yesterday DELL / HPE / SMCI were grouped as "rose yesterday, still rising today," and the 8/13 result: DELL +2.07%, HPE +1.75%, SMCI +4.12%, all three extended, so the call holds. But ANET on the same chain broke — it closed 8/13 at −3.27% ($203.62), pulling back from the 8/12 record close of $210.50. Verification result: ANET had no news-driven negative, and the decline came from profit-taking at highs, against a backdrop of $799.5 million of insider selling over the past three months with no buys at all.

This morning all four are within 1.4% pre-market (SMCI +1.38% / 1.22 million shares ✅, HPE +0.40% / 71,000 shares ✅, DELL ⚠️ +0.20% / 34,000 shares, ANET ⚠️ +0.53% / 9,760 shares) — this chain has no new information this morning. Conclusion: the whole chain is downgraded to watch only, as it has been driven by the same piece of news for three consecutive sessions with no new fuel today.

---

6. Bearish / Avoid List

---

7. Within-Theme Rankings

7.1 Memory / NAND & DRAM (today's strongest theme)

This chain has only one shared risk, and it is the same one for all four names: ASP. MU's primary 10-Q shows that of DRAM's +67% sequential move, ASP accounts for +60% with bits up only low single digits; SanDisk's COGS was flat in absolute terms while revenue rose +372%. Two companies said the same thing in different ways: this round's profit base is built on price, with almost no volume growth underneath. And this report could not obtain current DRAM/NAND spot and contract prices — the single biggest gap in the report.

7.2 Semi equipment / WFE

This table is itself today's most important observation: AMAT delivered guidance whose low end beats consensus, and its four peers are all within ±0.6% pre-market this morning on volumes below 30,000 shares — the guidance did not spill over at all. Two explanations: ① the market sees this as an AMAT share-gain story rather than an industry story; ② pre-market simply has not started pricing it. The direction of LRCX/KLAC in the first 30 minutes after 09:30 is the only way to distinguish the two.

7.3 Software / SaaS (yesterday's theme, unconfirmed today)

The whole sector has no usable pre-market reading this morning (see section 1.5). The test is deferred to after 09:30: IGV's strength relative to SMH is the only qualitative tool.

---

8. Open-Session Verification Signals

8.1 ⚠️ 08:30 ET: July retail sales — not yet released as of publication

Every snapshot in this report is from 08:00–08:11 ET, earlier than that release. This report has no result and does not pretend to.

Also at 10:00 ET: June manufacturing and trade inventories, and the preliminary August University of Michigan consumer sentiment index. Context: this is one of the last big prints the Fed chair will have before Jackson Hole on August 28.

Scenario mapping (written before the release; do not treat it after the fact as verified):

• Control group ≥ +0.5% (strong): strong consumption → rate path repriced toward hikes → 10Y up → negative for the software / long-duration assets driven by yesterday's softer PPI, with less impact on earnings-driven themes like memory.
• Control group ≈ +0.3% (in line): yesterday's pattern continues, single stocks dominate.
• Control group ≤ 0% (weak): falling rates favor long-duration assets, but this raises a second-order worry about "weakening consumption," which is usually negative rather than positive for small caps (IWM).

⚠️ The price reaction in the first 60 seconds after a release is not sufficient to represent pricing; do not use the first few minutes of movement to verify any of the scenarios above.

8.2 Pre-market signals (as of 08:11 ET)

• Gaps and volumes of the strong-theme leaders: SNDK +6.91% / 1.29 million shares, MU +3.15% / 1.22 million shares, WDC +4.10% / 217,000 shares — volume-price confirmation is the only bullish structure that holds today. The test: are these three gap-and-go or gap-fill? If more than half the gap is filled within the first 30 minutes, yesterday's Investor Day move is already being distributed.
• AMAT's volume is the single most informative number of the day: 354,000 shares of pre-market selling pressure, with a clear direction. The test: $531.79 (the 8/13 intraday low) is the key level. Breaking below means the post-earnings repricing is still under way; holding it means the −5% has already digested this "good results sold."
• The index has no direction: SPY/QQQ/IWM/DIA are all within ±0.3% pre-market, so do not use the index as a reason for any position.

8.3 Intraday verification (after 09:30)

1. First 30 minutes: whether the memory trio (SNDK/MU/WDC) hold above their pre-market prices — the first test of whether the theme is real.
2. Sector ETF linkage: relative strength of SMH versus IGV — yesterday IGV (+3.10%) sharply outperformed SMH (+0.73%), a 2.37pct divergence. Whether that divergence narrows or widens today determines whether the "software vs semis" rotation is real.
3. AMAT's spillover test (this report's most recommended observation): whether LRCX / KLAC / ASML strengthen after the open in line with AMAT's guidance raise. If they strengthen → AMAT's decline is purely a single-stock valuation issue and the industry logic is intact; if they weaken in tandem → the market is de-rating the entire WFE group, which is far more serious than a problem at one stock.
4. Volume confirmation: today is Friday, and the S&P rebalance (RDDT's inclusion) is next Tuesday, so liquidity into Friday's close and on Monday and Tuesday will be distorted by the index event; do not misread rebalance-driven volume as thematic flows.

8.4 Options sentiment and risk

• ⚠️ This report did not obtain options implied-volatility data; the judgments below are inferred from price action, not from options data, so discount accordingly.
• The highest IV-crush risk is AMAT: the event has passed and the event premium is gone, so even if the stock stabilizes, option longs can still lose from falling IV.
• CAPR's IV is in an extreme state (PDUFA event + a 96% gap), so any options strategy here is essentially betting against market makers on volatility pricing.
• VIX at 14.51 is low — but VIX9D/VIX1D retrieval failed, so this report cannot judge whether front-end volatility still carries no event premium. Ahead of the retail sales release, this is a known blind spot.

8.5 Main risks

1. Gap reversals: SNDK (second gap), RDDT (+10.74%), CAPR (+96%), IMXI (+21.45%) have the four largest gaps this morning and therefore the highest gap-fill risk.
2. A lone decliner with no sector follow-through: AMAT's guidance did not spill over to any equipment peer (see section 7.2), the mirror image of the "lone riser" pattern, and equally deserving of caution.
3. The memory theme's single point of failure: the four names share one independent variable (ASP), and this report could not obtain a current reading for it. That means every bullish or bearish memory judgment today is conditional.
4. Divergence between index futures and Treasuries: yesterday the 10Y fell 5bp alongside a record index high; this morning neither has direction, and retail sales could create one instantly.
5. Liquidity distortion from Friday plus next Tuesday's index rebalance.
6. "Going into earnings carrying a gain" has been punished four times this week (COHR, CSCO, AMAT, GLOB).

---

9. Final Conclusions

9.1 The 5 stocks most worth watching today

9.2 Today's 3 strongest themes

9.3 What to avoid today

1. Anything gapping more than +10% this morning (RDDT, CAPR, IMXI) — the news is real, but the price has already fully expressed it.
2. Names whose guidance was cut or withdrawn (GLOB, INV, AMPG) — especially INV, whose "indefinite suspension of targets" is an order of magnitude more serious than "cutting guidance."
3. Any trade based on this morning's losers list (PAYC / VRSK / MET / GEN) — that list's prices cannot be reproduced under any known convention, and the four names' pre-market volumes were just 356/123/9/126 shares, so the readings are unusable.
4. Every reading with pre-market volume below 50,000 shares (the entire software sector, STX, ETON, ANET, DELL, LRCX/KLAC/ASML) — this is not bearishness; there is simply no information this morning.
5. CSCO — the repricing of gross margin and cash flow is a multi-quarter process, and after −8.4% it is still not a buy point.
6. Names "going into their own earnings carrying a gain" — punished four times this week (COHR, CSCO, AMAT, GLOB).
7. AMAT — it belongs in "do not catch the knife," not "shortable": fundamentals are accelerating and shorts face the risk of another guidance raise, which is different in nature from every other entry in this list.
8. Any conclusion claiming to "know whether the memory cycle has peaked," including this report's — until current DRAM/NAND spot and contract prices are in hand, neither bulls nor bears have evidence.

9.4 Final one-sentence judgment

Today the market simultaneously delivered a verifiable set of excellent results (AMAT, with guidance beating consensus even at the low end) and an unverifiable excellent promise (SanDisk, a long-term model that only begins delivering in FY2028–30), and it rewarded the latter while punishing the former — which tells you that what is being priced right now is not fundamentals but duration preference and positioning. And when we actually take SanDisk's promise apart, we find that the 80% gross margin that became the headline is in fact its sturdiest line (it is below the 84.6% the company just delivered), while the genuinely fragile part is the revenue assumption nobody discusses — it requires zero NAND ASP decline for three straight years, while SanDisk itself writes on the same PPT page that "falling prices offset rising volumes, holding long-run revenue growth to about 4%." The four names on this chain share one independent variable, and that variable is precisely the one this report could not obtain a current reading for. So the only low-risk action today is not to buy, but to wait for the three volume confirmations after 09:30: whether the memory trio hold their pre-market prices, whether AMAT holds $531.79, and whether LRCX/KLAC strengthen in sympathy with AMAT.

---

Data-Retrieval Failures and Implementation Notes (internal)

Channel status:

1. yfinance was not used at all — given multiple consecutive trading days of Yahoo rate limiting (YFRateLimitError), we went straight to the stockanalysis quote API this time. A sub-agent confirmed empirically that yfinance is still rate-limited (both .info and .fast_info fail). The "fundamentals-analyst + yfinance" path written into the skill was in practice replaced by SEC EDGAR + XBRL + IR transcripts/decks + stockanalysis + the Nasdaq historical quote API, and data quality was actually higher (we obtained 8-K originals, an IR transcript PDF, the original 98-page Investor Day deck, and quarter-by-quarter XBRL differencing).
2. Channels that worked this time (recommend making them standard):
   • stockanalysis.com /api/quotes/s/<sym> — the primary channel. epv (pre-market volume) is the single most important field in this report; the falsification work in sections 1.4 and 1.5 rests entirely on it. Note that the index path /api/quotes/i/SPX returns 400.
   • CBOE cdn.cboe.com/api/global/delayed_quotes/quotes/_<SYM>.json — _SPX/_NDX/_RUT/_VIX are available with fresh last_trade_time.
   • api.nasdaq.com /api/quote/<SYM>/info?assetclass=index (COMP/NDX work, SPX returns "Symbol not exists"); /api/quote/<SYM>/historical can pull 2,513 trading days, excellent for building valuation percentiles in-house.
   • treasury.gov daily yield-curve CSV — official primary source, the most reliable.
   • SEC EDGAR via direct curl (with a User-Agent) — data.sec.gov/submissions/CIK*.json to list filings + /Archives/ to pull the EX-99.1 original; data.sec.gov/api/xbrl/companyconcept/ can reconstruct cash flows via quarter-by-quarter differencing.
   • Direct links to company IR static files — investor.sandisk.com/static-files/<uuid> yielded the 98-page Investor Day PDF, and ir.appliedmaterials.com/static-files/<uuid> the earnings-call transcript PDF. This is the single biggest quality upgrade this time.
3. Channels that failed:
   • WebFetch returns 403 for sec.gov, but curl with a UA works. From now on, always use curl for SEC originals.
   • benzinga.com 403 (twice), stocktitan.net DNS timeout, ir.appliedmaterials.com WebFetch 60s timeout, investors.micron.com static-files blocked by a 301 redirect (forcing MU's prepared remarks to come from a third-party transcript).
   • The stockanalysis screener API (api.stockanalysis.com/api/screener/s/f) returns an empty shell containing only status — this is why we could not auto-generate a complete pre-market movers list this time, and had to pull tickers manually one by one.
   • CBOE _VIX9D / _VIX1D return all 0.0 rather than an error — another "silent failure" sample, honestly marked as not obtained in the body text, with yesterday's values not substituted.
   • The Census MARTS PDF is binary and WebFetch cannot parse it; and the 08:30 data genuinely had not been released at publication time (shell time verified as 08:11 ET).

⚠️ Quality-check (risk-auditor) results and the corrections adopted this time:

The quality check flagged 10 high-severity issues, most of which were adopted and fixed before publication; the important ones are recorded below (all edit traces are preserved in the body text, nothing was silently changed):

1. [Directional error, most serious] Treating AMAT's "DRAM equipment revenue +52%" as bullish cross-validation of memory-cycle strength. The quality check pointed out that the direction is inverted: equipment revenue reflects capacity decisions made 6–12 months ago, and once installed it produces future supply, making it a negative input for future ASPs. This error originally appeared in four places — sections 0, 2, 7.1, and 9.2 — and was the main basis for characterizing the memory theme as "the upcycle is real." All have been rewritten.
2. The self-correction overshot. The first draft jumped from "over-suspicious of SanDisk's 80%" to "it is a discounted promise" — but 80% is a discount only against a cycle-peak quarter (84.6%); against the company's own four-year range (FY23 was just 7.1%) it is extremely aggressive. A second layer has been added. Lesson: conclusions produced by a correction must also pass primary sources and a through-cycle test.
3. "44x back to 24x = another 45% downside" implicitly freezes EPS, contradicting this report's core argument that "earnings are accelerating" (annualizing FQ4 guidance gives a forward P/E of only 31.6x). That point estimate has been deleted and replaced with a qualitative statement.
4. Section 1.4's falsification mechanism extrapolated from n=1 to n=4 — after back-solving each name, "the prior day's close treated as the pre-market price" holds precisely only for PAYC, and VRSK/MET/GEN do not reconcile. The conclusion has been narrowed. The volume point (356/123/9/126 shares) is unaffected and stands independently.
5. Section 9.1 cited a set of NAND contract-price readings while the report states they were "not obtained" — the header and section 9.1 now both note that the figure is second-hand from TrendForce at quarterly frequency, and explicitly that it is not the basis for any conclusion.
6. SanDisk's market cap mistakenly used the diluted share count (155 million vs 149 million outstanding), which turned P/B from 14.5x into 15.1x. It has been standardized to the shares-outstanding basis with the difference noted.
7. Two sensitivity tables hid key assumptions — the current-quarter table did not state that it excludes write-downs / underutilization charges / new depreciation; the FY30 table back-solves an implied "8% annual unit-cost decline over four years." Table notes have been added, and "gross margin is the sturdiest line" has been downgraded to "the least price-sensitive line under a constant-cost assumption."
8. The demand side has zero current data (no hyperscaler capex, no backlog/book-to-bill, no inventory days, no supply additions), yet the WFE theme originally carried "persistence: strong." It has been downgraded to "unverified," and those four items added to the header's not-obtained list.
9. Consensus figures mix sources with no timestamps (revenue from LSEG, EPS from FactSet), while "even the low end beats consensus" is the report's most reused pivot. A fragility note has been added.
10. Also adopted: IMXI's 11.2% discount was double-counted (scored +8 as an expectation gap and also as a risk) — that item is now 0 and the total went from 49 to 41; "going into earnings carrying a gain" was downgraded from "enough to write into trading discipline" to "n=4 with no control group, insufficient to be a rule"; the 5bp RSP vs SPY difference is no longer characterized as "healthy breadth"; the NBM "two paths landing on the same number" was identified as circular reasoning and withdrawn; "that decline does not exist" was changed to "that list's readings are unusable"; several directional/timing recommendations were neutralized in wording for compliance; "eight weeks from now" was changed to "one quarter from now."

Parts the quality check confirmed as sound: nothing in the text implies the retail sales result is known; AMAT's full guidance arithmetic (+7.44%/+8.36%/low ends/segment sums/implied 35.0% operating margin) recomputes correctly; IMXI's 11.19% discount, back-solving MU's price from P/B × BVPS, CAPR's +96.44%, and the two-day cumulative gains are all correct; the eight disclosed scoring breakdowns all sum to their totals; strip_internal() at bin/send.py:64 is confirmed to strip this section.

Raised by the quality check but not changed in this report: the epv field lacks a second-source cross-check; the absolute 50,000-share threshold lacks justification (it should become a percentage of ADV); AMAT's China revenue of $2.506 billion and its receivables increase of $2.506 billion are a numerical coincidence, as are SanDisk's and MU's GAAP gross margins both being 84.6% — both were separately traced to 8-Ks, but an independent re-check is recommended next time; the cross-asset ETFs were not subjected to the volume convention. These are left for the next report.

⚠️ Four traps hit and dodged this time:

• investing.com's entire pre-market losers list is "the prior day's close treated as the pre-market price" — the "declines" for PAYC/VRSK/MET/GEN are in fact their 8/13 gains displayed with the sign flipped. What exposed it was the epv field (356/123/9/126 shares) plus reconciliation against current prices, not news checking. Written up in section 1.4.
• ⭐ This report's most important self-correction: the first draft wrote up SanDisk's 80% gross-margin target as a red flag that was "higher than NVIDIA's and requires a special-accounting explanation," and it had already been drafted. After a sub-agent obtained the primary 8-K, it was overturned: Q4's actual GAAP gross margin is 84.6% (GAAP = non-GAAP, the only adjustment being $6 million of SBC), FQ1'27 guidance is 83–85%, so that "80% long-term target" is a discount, not a stretch. The genuinely fragile assumption is on the revenue-growth line nobody reads. Lesson: when a number "looks like it violates common sense," the first reaction should be to look up its current actual value, not to construct a challenge framework first — I nearly wrote a correct target up as a red flag and built a wrong risk ranking on it. This is the mirror image of the same class of risk as the memory note "self-refutation becomes an exemption pass": skepticism itself also has to pass a primary source.
• AMAT's FCF figure was mismatched across quarters — multiple second-hand sources wrote FQ2'26's "FCF $210 million" as this quarter's number, juxtaposing it with "record revenue" for contrast. Quarter-by-quarter XBRL differencing refuted it. It also corrected a tendency of my own: I initially planned to carry over yesterday's "AI hardware inventory pile-up" framework and apply it to AMAT, but the primary data show AMAT's inventory days are improving (141.3→131.9), with the problem in receivables. Same conclusion, different cause — lazily reusing yesterday's framework would have produced an error.
• MU's attribution stitching — multiple outlets attributed 8/13's +4.23% to that day's $250 million venture fund (= 0.023% of market cap). One cross-sectional comparison collapses it: SNDK was +13.67% the same day, three times MU. If the catalyst were at MU, MU should not have finished last. Also, one article dated 8/13 cited a price range that was actually 8/12's ("silently one day stale" appearing again).

Items left blank / unfinished in this report:

• Current DRAM / NAND spot and contract prices — the biggest gap in the report, because it is the memory theme's only independent variable, and both sub-agents failed to obtain it due to paid sources (TrendForce/DRAMeXchange). The body text flags this explicitly in three places (sections 2, 7.1, and 9.3) and states outright that "neither bulls nor bears have evidence," with no speculation used as filler.
• The identities of SanDisk's 8 NBM customers and the absolute floor price (undisclosed); whether the Investor Day targets will be filed in an 8-K.
• MU's primary FY27 consensus EPS (paywalled; the text uses an implied ~$143.70 back-solved from the 6.61x forward P/E, flagged as an inference).
• VIX9D/VIX1D, options IV, raw NYSE advance-decline counts, the specifics of ZNTL's negative, and the SEC original for RDDT's Q2 revenue.

Input notes for the next report:

1. The epv (pre-market volume) threshold of 50,000 shares proved its value today; recommend making it a standard step, and it must be applied to both the up and down sides — today it rejected both a losers list and the entire software sector's "give-back" narrative.
2. Three explicit, reconcilable calls were made today: ① AMAT's decline is multiple compression, not fundamental deterioration (verification point $531.79); ② AMAT's guidance did not spill over to equipment peers (verification point: the direction of LRCX/KLAC after the open); ③ all of the memory chain's elasticity is anchored to ASP. Tomorrow's recap must reconcile each one.
3. "Going into earnings carrying a gain" now has four samples (COHR/CSCO/AMAT/GLOB); recommend upgrading it in the next report into a standing "pre-earnings gain" warning table.
4. Build a stable retrieval channel for NAND/DRAM prices (even monthly and second-hand), otherwise every memory-theme report can only reach conditional conclusions.
5. Direct static-files/<uuid> links on company IR sites were the biggest quality discovery this time — the information density of Investor Day decks and earnings-call transcripts far exceeds any media paraphrase; recommend going to the IR site for originals first whenever an Investor Day / Analyst Day occurs.

---

⚠️ Risk disclaimer: this list is a pre-market information review and observation only and does not constitute investment advice. U.S. equities carry high volatility and pre-market gap risk, and post-earnings IV crush and guidance reversals occur; automatically generated content may contain timeliness gaps or factual errors. Rely on company disclosures / SEC filings, and do not use this directly as a basis for trading.