US Pre-Market Brief | Tuesday, 2026-08-18 (ET)

Coverage window: 2026-08-17 16:00 ET regular-session close → 2026-08-18 08:15 ET pre-market. Includes 8/17 after-hours earnings, the overnight Asia/Europe sessions, and this morning's pre-market.

Quote basis: Closing / open / high-low / volume for single stocks and ETFs come from the stockanalysis.com quote endpoint (u timestamp = Aug 17, 2026, 4:00 PM EDT close basis); pre-market price, pre-market change and pre-market volume come from the pre-market fields of the same endpoint, read at 08:00–08:13 ET, separately annotated in each table. Futures, Treasury yields, crude, VIX and the dollar index come from the CNBC quote endpoint, read at 07:57–08:08 ET. The 8/17 Treasury yields come from the US Treasury daily yield curve (15:30 ET settlement basis). Earnings figures are taken exclusively from SEC filings or company IR press releases; links are in the body.

One methodological note for this issue (please read first): the semiconductor complex is down heavily in this morning's pre-market (SMH −3.05%, single names −3% to −6%), but using only the "vs. prior close" column will badly misread it. This issue gives, for every reversal name, both a "vs. 8/17 close" and a "two-day net change vs. 8/14 close" column. The two columns frequently point in opposite directions — which is exactly why they must sit side by side.

---

0. Today in One Line

1. This morning's pre-market is not "semis collapsing," it is "the AI hardware chain paying back yesterday's bill." SMH is −3.05% pre-market, which looks like a landslide; but on a two-day net basis (pre-market price vs. 8/14 close), SNDK is +3.38%, WOLF +4.78%, CRDO +3.91%, AMAT +0.71%, TER +1.05% — all still positive, while MU −0.38%, MRVL −0.69%, WDC −0.93% are essentially flat. The names that ran hardest yesterday have not, even after today's drop, given back their starting point. The ones genuinely broken on a two-day net basis are a different group: ARM −6.70%, GFS −5.64%, ALAB −5.17%, AMD −4.88%, QCOM −4.67% — they did not rally yesterday at all.

1b. The selloff in the memory chain is demonstrably a macro de-leveraging, not a price-cycle top — and there are three pieces of primary evidence pointing the other way. ①DRAMeXchange spot was still rising on 8/18 itself (read 06:10 ET): DDR4 16Gb 3200 average price 89.427 (+0.55%), NAND 512Gb TLC wafer +4.97%, 256Gb TLC +17.49%; ②TrendForce piece dated 8/18 01:06 ET: 512Gb TLC spot recovered to $21.125 on 8/17, +4.97% week over week, +11.6% over three weeks, driven by the CMX flash tier of NVIDIA's Vera Rubin; ③TrendForce official release 8/18: top-five NAND makers' combined 2Q26 revenue $68.87 billion, +77% QoQ, with Micron overtaking Kioxia to rank third. On the "memory prices have topped" side of the argument, there is not a single piece of primary evidence inside the window — everything that exists is counter-evidence.

1c. The window of the selloff can be located precisely at 05:00–07:30 ET, by which time all of Asia had already closed. Stocktwits' headline at 04:59 ET still read "memory stocks rallying against the tape" (MU, SNDK, Memory Stocks Buck Market Selloff); by 07:25 ET, Barron's headline had become "Micron Stock Drops. Why Bond Yields Are Hitting the Memory-Chip Boom." No pricing agency published anything in that window; the only things moving were oil, yields and futures — the timeline itself is the evidence that this is not industry-driven.

2. The only confirmed new fundamental negative comes from Fabrinet's (FN) 8/17 after-hours report, and the kill point is not the quarter just reported — it is gross margin. Quarterly revenue $1,315.8M (+45%) and non-GAAP EPS $4.10 both beat the top end of the company's own guidance; but Q1 FY27 guidance implies revenue +6.4% QoQ (midpoint) while non-GAAP EPS is only +1.8% QoQ, and the low end of $4.10 is exactly flat with the quarter just completed. Revenue accelerating while profit does not = operating leverage in reverse. FN is −10.20% pre-market and is dragging the whole optical-module chain down with it (COHR −6.04%, LITE −6.29%, AAOI −4.45%, CIEN −4.18%).

3. Klarna (KLAR) is −18.25% pre-market, but the media headline "weak full-year revenue guidance" has the wrong emphasis. Full-year revenue guidance was indeed cut from >$4.34bn to $4.08–4.16bn, but a substantial part of that is a change in accounting presentation (part of the business moves to fair-value measurement from H2), and on a like-for-like basis the take rate was actually raised (>2.80% → 2.84–2.85%), as was full-year TMD guidance. The real kill point is Q3 adjusted operating income guidance of $5–15M, i.e. −84% to −95% QoQ against the $91M just delivered, on top of the cliff in revenue growth: +44% YoY (Q1) → +27% (Q2) → guided +4~9% (Q3).

4. HD's report is "solid delivery against a low bar," not a housing turn. Adjusted EPS $4.92 vs. $4.73 expected (a like-for-like beat of 4.0%), revenue $47.86B beating by 1.3%; but all 11 full-year guidance figures are unchanged word for word, even though H1 sales of +5.3% are already running above the full-year 2.5%–4.5% guide. The composition of the +1.7% comp is traffic −1.0% × ticket +2.8%, and traffic is still deteriorating year over year. HD is +1.95% pre-market.

5. The pre-market structure is a clean defensive rotation, not broad risk-off. Dow futures −0.09% vs. Nasdaq futures −1.15% (07:58 ET); pre-market XLV +0.42%, XLP +0.65%, XLE +0.75% all green, XLK −1.63% and SMH −3.05% leading down. The spread between the index futures (1.06pp) is larger than the decline in the indices themselves.

6. The macro backdrop has not changed and has in fact tightened further — and this time it is "Japanese and US long ends making multi-year highs simultaneously": US 30Y 5.325% (08:08 ET, the 52-week high was set today), 10Y 4.736%; Japan's long-term rate touched 2.945% intraday, the highest in 30 years, and the market is starting to price a BoJ hike as soon as September; the German 10-year Bund hit its highest since 2011. Brent $91.09, VIX 15.67 (+3.16%). Trump on 8/17 refused to restart the ceasefire agreement with Iran, called negotiations "a waste of time," and threatened force if Oman obstructs Hormuz. Separately, China's July data missed on both counts (industrial output 4.5%, retail sales 0.6%). Ranking of driver types: earnings/guidance (FN, KLAR, HD, BIDU) > rates and geopolitics > ratings > index events.

6b. The truth about the overnight Asian session is the opposite of the "semis collapsed everywhere" intuition, and one premise must be corrected first: Korea was closed on 8/17 (Liberation Day 8/15 fell on a Saturday, with 8/17 as the substitute holiday), so 8/18 was the first Korean trading day since 8/14 and absorbed two days of information at once. On the day, SK Hynix was up as much as +8.8% intraday but closed only +1.03%; Samsung −2.19%; KOSPI closed 6,869.83 (−1.55%), advancers-to-decliners 195:679. The Nikkei fell −2.54% to 67,460.73, but Advantest and Tokyo Electron alone subtracted roughly 836 yen from the Nikkei, 47% of the day's 1,759.52-point decline — that is the arithmetic amplification of a price-weighted index; the same day TOPIX fell only 1.05% (722 advancers vs. 783 decliners). What was genuinely narrow was Korea, not Japan. Kioxia was −7.58%, but it was +15.07% in Tokyo on 8/17, leaving a two-day net of +6.35%. Taiwan's TAIEX closed 45,308.68 (−1.20%). The −3.5% to −6% magnitudes in the US pre-market are far larger than what actually happened in Asia; this was generated inside the US session, not transmitted overnight.

7. There are two undetonated timers today: ①July housing starts and building permits at 08:30 ET (not yet released as of this issue's 08:15 ET cutoff), a direct input into HD/LOW intraday; and ②the July FOMC minutes at 8/19 14:00 ET. In addition, NVDA reports after the close on 8/26, so the entire AI chain spends this week in a "waiting for NVDA" vacuum, pushed around by rates.

---

1. News Overview

1b. Overnight Asia Closes (8/18 session, already closed)

Premise correction: Korea was closed on 8/17 (Liberation Day 8/15 fell on a Saturday, 8/17 was the substitute holiday); 8/18 was the first Korean trading day since 8/14 and absorbed two days of information at once.

How to read it: Asia that day was not "a systemic memory selloff." SK Hynix closed up, Kioxia's plunge was a partial giveback of the prior day's +15%, and the two Taiwanese memory names fell only 1.7%/2.8%. What was hit hardest was Japanese semicap (the longest-duration capex asset), not memory (the pricing-cycle asset whose contract prices are rising this quarter). The three reasons Japanese media gave themselves were "Middle East risk, rising Japanese and US rates, and falling Korean equities" — not one of them is bad news from the equipment industry itself; they also explicitly noted a sense of "relative overextension" (stretched valuation) in recently rebounding leaders such as Kioxia — it is positioning and valuation, not orders.

⚠️ A circular reference that must be acknowledged: the Japanese side lists "falling Korean equities" as one cause of its own decline, while the Korean afternoon slide was itself driven by Treasuries and oil. Japan and Korea are not two independent votes and cannot be treated as two independent pieces of evidence.

---

2. Strongest Themes, Descending

---

3. Overall Single-Stock Strength Table (split by direction, descending within each)

Pre-market read at 08:11–08:13 ET. "Two-day net" = pre-market price ÷ 8/14 close − 1 (the 8/14 close is back-solved from the 8/17 close and that day's percentage change, so it is a derived value).

Bullish Direction

Bearish Direction

A reminder on the basis of this table: the names ranked 4–5 and 15–17 in the "bearish direction" block have large pre-market declines but two-day nets that are near zero or positive. They are tagged "watch only" rather than "avoid" precisely because no new primary negative aimed at them could be found — today's decline is the mirror image of yesterday's advance. "Not found" is not "does not exist," and that statement is itself the assumption in this issue most in need of intraday verification.

---

4. Single-Stock Scoring Model (100 points)

The score is an internal relative-ranking tool for this issue, not an absolute value judgment. The risk-deduction line hits the −15 floor for KLAR, meaning the model itself can no longer fully express the magnitude of that risk.

---

5. Detailed Analysis of Top Names

1Fabrinet | AvoidFNPre-market $537.50 (−10.20%, 34,700 shares, 08:13 ET) · today

Related news: 8/17 16:05 ET, SEC 8-K Ex-99.1¹; conference call 8/17 17:00 ET.

The quarter itself (primary): revenue $1,315.8M vs. $909.7M a year earlier (+44.6%), above the top of its own $1.25–1.29B guidance; GAAP EPS $3.83 (prior year $2.42); non-GAAP EPS $4.10 (prior year $2.65, +54.7%), above the top of $3.72–3.87 guidance. Full-year FY26 revenue $4.64B (+36%), non-GAAP EPS $14.09 (prior year $10.17).

Catalyst logic (the kill point is the structure of the guide, not the quarter): Q1 FY27 guidance is revenue $1.375–1.425B (midpoint $1.40B, +6.4% QoQ), non-GAAP EPS $4.10–4.25 (midpoint $4.175, only +1.8% QoQ), with the low end of $4.10 exactly flat against the quarter just completed. Revenue accelerating sequentially while profit stalls has only one explanation: unit profitability is falling. Non-GAAP gross margin was 12.2%, −30bp year over year. Management's stated reason is "a temporary margin headwind from normal Q1 expense seasonality," and it still expects operating leverage as revenue scales — this is the single falsification exit for this thesis, and the point that must be re-checked next quarter.

Segments and customers: data center $669M (51% of revenue, +68% YoY), communications infrastructure $413M (31%, +40%), automotive/industrial/other $234M (18%, +8%). Four customers each accounted for ≥10% in FY26: Cisco 20%, NVIDIA 16%, Nokia 11%, Amazon 11%. On capacity, Building 10 in Chonburi, Thailand, completes in early 2027 and adds 2 million square feet; management says capacity under construction and in planning can ultimately support $12.5–14 billion of revenue.

Theme and stage: AI optical modules — moving from the "revenue explosion" stage into the "margin under examination" stage. This is a narrative switch, not the end of the narrative.

Fundamental verification: revenue and EPS both come from the primary SEC filing; the segment split and customer concentration come from media relays of the company's earnings presentation (Investing.com, MarketBeat, both published 2026-08-17), and the presentation itself was not checked directly. The TTM valuation multiple (P/E around 51x) comes from a secondary relay and was not independently verified — do not cite it directly.

Pre-market and technicals (the most important item here): FN rose 4.97% to $598.58 on 8/17 going into the print, i.e. the entire 8/17 session took place before earnings. Therefore:

• Versus the prior close: −10.20%
• Versus the pre-print baseline ($570.24, back-solved from the 8/17 close and percentage change): −5.74%

In other words about 4.5 percentage points of the decline is giving back pre-print front-running, and the true earnings reaction is about −5.7%. Using −10.2% to describe the market's verdict on this report overstates it by nearly double. The 52-week range is $272.49–$748.89 and the pre-market price sits slightly above the middle. Pre-market volume is only 34,704 shares, the thinnest of any key name in this issue, so the certainty of that quote is materially lower than for COHR/LITE.

Final call: avoid (today). Not because the company got worse — the quarter was a genuine beat and full-year revenue is +36% — but because the pricing basis switched from "revenue growth" to "can margins hold," and that question cannot be settled before the next report. Combined with an extremely thin pre-market quote and high reversal risk after a gap, today is not the place to establish a view.

---

2Klarna | Avoid | Pre-market $15.95KLAR−18.25% · 1.88 million shares · 08:13 ET

Related news: 8/18 ~07:00 ET, Klarna IR earnings release original PDF²; conference call 08:30 ET. As of this issue's cutoff, SEC EDGAR had not yet received the corresponding 6-K, so the data above comes from the company's IR original, not a secondary summary.

On the surface this is a comprehensive beat: GMV $36,648M (+18%, guidance $35.5–36.5bn), revenue $1,042M (+27%, guidance $960–1,000M), TMD $446M (+42%, guidance $375–395M), adjusted operating income $91M (+214%, guidance $30–50M) — all four above the top end of its own guidance.

So why is it down 18%? Ranked by strength of evidence:

1. [Strongest] Q3 adjusted operating income guidance of $5–15M, i.e. −84% to −95% QoQ against the $91M just delivered. Any valuation modeled by annualizing Q2 is void on the spot. The company's explanation is that Q3 is a "deliberate investment quarter," carrying the largest product launch in its history, with marketing landing ahead of the volume it drives, and the highest share-based payment quarter of the year.
2. [Very strong] The growth cliff: revenue YoY went Q4'25 +38% → Q1'26 +44% → Q2'26 +27% → Q3 guidance +4~9%; GMV YoY fell from +33% to +18%. The pricing basis for the multiple is the growth curve, and a single-quarter beat cannot offset a turn in the curve.
3. [Very strong] The quality of the Q2 beat is questionable: the income-statement line Gain on sale of consumer receivables was $69M in Q2'26 versus $0 in Q2'25; the corresponding fair-value adjustment was −$46M, so the net contribution was +$23M versus −$23M a year ago — a year-over-year swing of $46M, equal to 35% of the entire $131M increase in TMD. Stripping it out, revenue was about $973M (+18.2%), landing back inside the guidance range rather than above the top end, and essentially equal to GMV's +17.5% — that is, zero take-rate expansion (ex-item, the take rate fell from Q1's 2.83% to 2.65%, −18bp). The company itself attributes the sequential rise in ex-US TMD to "a back-book sale executed this quarter together with the launch of the German forward flow."
4. [Strong] Full-year profit is bet on a single Q4: full-year AOI guidance is $280–300M; subtracting H1's $159M and Q3's $5–15M leaves an implied single quarter of $106–136M in Q4, 1.6–2.1x the entire 2025 AOI ($65M). Execution risk is highly concentrated.
5. [Medium] First admission that Germany is weakening: the company states explicitly that "guidance assumes Germany remains soft in the second half rather than recovering," citing German H1 real retail sales growth of less than 1%. Germany is its largest volume market.

Verdict on the media headline: Investing.com's "weak full-year revenue guidance" has the direction right but the attribution wrong. Full-year revenue guidance was indeed cut from >$4.34bn to $4.08–4.16bn (−4.1% to −6.0%), but the company states that roughly 10bp of GMV of that comes from a presentation change, with part of the business moving to fair-value measurement from H2; on a like-for-like basis, revenue-to-GMV guidance was raised from >2.80% to 2.84–2.85%, and full-year TMD guidance was likewise raised from >$1.61bn to $1.62–1.65bn. If this were genuinely just "weak full-year revenue guidance," a decline of this size would not hold up.

Credit quality: not the kill point this time (counter-evidence). Q2 credit loss provisions were $192M, or 0.52% of GMV, below Q1's 0.55%, Q4'25's 0.65% and Q3'25's 0.72%; US Fair Financing 30+ day delinquencies fell about 20bp sequentially and Pay Later about 30bp; cumulative net charge-offs are "within the expected 3–4% range." The "BNPL credit blow-up" narrative points the opposite way from this quarter's data — do not apply it here. But note: the company states that the fair-value switch will mechanically lower the provision-to-GMV ratio in H2, so that improvement cannot be extrapolated.

Fundamentals and valuation: TTM revenue $3.82bn (+33.1%), TTM net income −$198M; cash and equivalents $2,672M (year-end 2025 $3,803M, −29.7%), consumer deposits $11,673M (−10.2%); total assets $17,766M / total equity $2,666M. At the pre-market price, P/S is about 1.55x, forward P/S about 1.44x, P/B about 2.23x, and market cap / FY26 AOI midpoint about 20.5x. Of the $9M of Q2 net income, only $4M is attributable to shareholders of the parent (the other $5M to minority interests), and H1 attributable to shareholders was −$1M — "a swing to profitability" does not yet hold on an attributable basis.

Pre-market and technicals: the IPO price was $40.00 (2025-09-10), so the pre-market price is −60.8% versus the IPO; against the 52-week high of $57.20 the drawdown is −72.6%, and the 52-week low of $12.06 is not far away. It already fell 6.16% yesterday, making today the second consecutive down day. Pre-market volume of 1.88 million shares makes this one of the most solid quotes in the list.

Final call: avoid. This is not "oversold" — the valuation basis switched from growth to profit delivery, and the company has bet all of that delivery on a Q4 that has not happened. The sell-side consensus target of $24.55 (22 analysts) was published before the report and does not yet reflect today's guidance; it cannot be used as a reference. The 08:30 ET call's explanation of the Q3 profit guide and the implied Q4 figure is the only pricing verification point today.

---

3Home Depot | Watch closely | Pre-market $344.46HD+1.95% · 238 · 000 shares · 08:13 ET

Related news: 8/18 06:00 ET, SEC 8-K Ex-99.1³; conference call 09:00 ET.

The quarter itself (primary): revenue $47,861M (+5.7%); comparable sales +1.7% (US +1.3%); GAAP EPS $4.79 (prior year $4.58); adjusted EPS $4.92 (prior year $4.68, +5.1%). Gross margin 33.67% (+26bp), but SG&A grew 8.5%, far above revenue growth, so the expense ratio deteriorated 45bp and operating margin fell a net 19bp to 14.29%.

Consensus basis (important): the company disclosed both GAAP $4.79 and adjusted $4.92; sell-side consensus is built on the adjusted basis ($4.73). So $4.92 vs. $4.73, a beat of 4.0%, is a like-for-like comparison; misusing GAAP $4.79 would compress the beat to +1.3%. Revenue of $47.86B vs. $47.23B expected is a beat of 1.3%. Comps of +1.7% were essentially in line, not a beat item.

Catalyst logic — the word "reaffirm" needs unpacking: comparing the guidance line items in the three releases of 2026-02-24 (Q4), 2026-05-19 (Q1) and 2026-08-18 (Q2) item by item, all 11 figures are identical word for word: total sales +2.5%4.5%, comps flat+2.0%, about 15 new stores, gross margin about 33.1%, operating margin 12.4%12.6%, adjusted operating margin 12.8%13.0%, tax rate 24.3%, net interest expense about $2.3B, GAAP EPS growth flat4.0% (off a base of $14.23), adjusted EPS growth flat4.0% (off a base of $14.69), and capex of about 2.5% of sales.

The only substantive change is in the preamble sentence, newly added this quarter: "Guidance includes IEEPA tariff refunds, which are expected to partially offset unplanned fuel, energy and other product input costs for the year." The numbers did not move, but the composition of the path to them did — inside the same EPS range there is now a slice of non-operating income that depends on a policy/judicial outcome and that the company has not quantified, replacing a slice of operating cost that has already deteriorated. This is "a decline in the quality of the reaffirmation," and it is invisible on the surface.

The key tension: H1 sales are +5.3% with comps +1.2%, already running above the full-year 2.5%–4.5% guide, yet the company raised nothing. Backing out the implied H2: sales growth of −0.5%~+3.7% (a deceleration of 2–6pp) and operating margin of 11.5%~12.0% (versus 13.2% in H1). Part of that is the mechanical effect of the acquisition base entering the comparable set in H2, but the magnitude is larger than pure seasonality.

Growth quality: revenue +5.7% against comps of only +1.7% means about 4.0pp (seven-tenths of the growth) came from acquisitions (SRS/GMS), new stores and FX — not organic. The +1.7% comp is composed of traffic −1.0% × ticket +2.8%, so 100% of the growth comes from price, while traffic keeps deteriorating year over year (−0.4% a year ago → −1.0% this year; H1 −0.5% → −1.2%). The CFO's own words are that demand is concentrated in "smaller projects" — that is evidence against, not for, a "housing thaw." Also note: the Q1 release explicitly disclosed the FX contribution to comparable sales (+55bp), but the Q2 release did not, leaving the 40bp gap between the company-wide +1.7% and the US +1.3% unverified.

One blemish in cash flow: H1 operating cash flow was $11.42B (+27.4%), which looks excellent; but net income grew only 0.9%, and the working-capital swing alone was $2.39B, i.e. 97% of the $2.45B increase in OCF — the cash-flow improvement comes almost entirely from working-capital release rather than earnings improvement and cannot be extrapolated. Free cash flow was $9.70B, with dividend coverage of 2.09x; the dividend is safe.

Governance (this has to be written in): the 8-K Item 5.02¹⁰ of 2026-08-12 disclosed that Chairman and CEO Ted Decker is on temporary medical leave, expected to return within a few months, with CFO Richard McPhail designated interim chief executive. This explains an anomaly: there is no CEO quote in this quarter's release (both the Q1 and Q4 releases were voiced by Decker). Choosing to reaffirm rather than adjust guidance while the CEO is absent is the lowest-risk option, which further reduces the information content of "reaffirmation = confidence in the fundamentals."

Valuation: at the pre-market $344.46, P/E (TTM GAAP $14.29) is about 24.1x, in the 60.7th percentile of the past 5 years (5-year median 23.1x); forward P/E about 23.0x; P/S 2.02x; dividend yield 2.71%. Meanwhile consensus FY26 adjusted EPS growth is only +1.83%. The market is paying an above-median multiple for near-zero earnings growth; what it is paying for is the option value of "housing will eventually thaw" — and this report offers no evidence of a thaw.

Pre-market and technicals: +1.95%, less than the 4.0% EPS beat, so not an excessive reaction; the 52-week range is $289.10–$426.75 and the current price sits in the lower-middle of it. Pre-market volume of 238,000 shares makes the quote reliable.

Final call: watch closely, not upgraded to priority deep-dive. Reasons: the beat is solid and like-for-like comparable, the pre-market reaction is restrained, and the defensive profile fits today's rotation; but the valuation starting point is already in the 60th percentile of five years, growth comes from price rather than traffic, guidance embeds an unquantified tariff refund, and the CEO is absent. The most valuable incremental information today is whether the 09:00 ET call quantifies the IEEPA refund; it is not in the 8-K, only the call can provide it.

---

4百度BIDUAvoid | Pre-market $96.63 (−7.19%, 732,000 shares, 08:12 ET) · Baidu

Related news: earnings at ~05:00 ET on 8/18, Investing.com report⁴ (published 05:10 ET).

The numbers: revenue RMB 31.33B, −4% YoY, below consensus of RMB 31.95B; EPS RMB 7.22, far below the RMB 9.84 expected. By segment: AI cloud infrastructure RMB 7.3B (+50%), within which GPU cloud +283% (up from +184% last quarter — accelerating); AI applications RMB 2.5B (+3%); AI-native marketing services RMB 2.6B (flat). Online marketing services RMB 13.1B, −19% YoY. Adjusted operating income RMB 3.8B (12% margin), adjusted EBITDA RMB 6.2B (20%).

Catalyst logic: this is a structural rather than cyclical miss. The acceleration in AI cloud is real (GPU cloud +283% and speeding up sequentially), but its absolute size (RMB 7.3B) is not yet enough to offset the shrinkage of the core advertising business (RMB 13.1B, −19%). China's property downturn and weak consumption are suppressing marketing budgets, and this is not something a quarter or two can repair. CEO Robin Li says the company is "shifting from internet-centric to AI-first," but explicitly acknowledges that online marketing "remains under pressure."

Theme and stage: China AI — an unmistakable divergence phase. On the same day, BABA is +2.97% on an AI application product while BIDU is −7.19% on advertising; both are "AI narratives," and the market only pays for the side that can monetize.

A note on the pre-market basis: the 05:10 ET report said "down about 3% pre-market," while the 08:12 ET reading is −7.19% on pre-market volume of 732,000 shares. Pre-market is cumulative, and an early snapshot is not a pre-open fact — the decline more than doubled over three hours. This issue uses the 08:12 ET reading.

Fundamental verification: the segment figures above come from authoritative media relays of the earnings release; Baidu's IR original / 6-K was not checked directly, so this is a secondary source and should be labeled as such when cited. The 52-week range is $84.82–$165.30, and the pre-market price is close to the bottom quartile of the range.

Final call: avoid. A double miss + structural shrinkage of the core business + a pre-market decline that is still widening; nothing here supports stepping in today. AI cloud is the only bright spot but is not big enough — it only becomes worth re-evaluating when its share is large enough to turn total revenue growth around.

---

5阿里巴巴BABAWatch closely | Pre-market $128.42 (+2.97%, 874,000 shares, 08:07 ET) · Alibaba

Related news: 8/18 pre-market, Bloomberg⁶ — Alipay launched a "one-stop" platform for merchants that uses AI agents to automate merchant tasks; Alibaba's HK shares rose as much as +5%.

Catalyst logic: this is a product catalyst, not an earnings catalyst, so directness is medium. The real significance is the timing: BABA reports the June quarter on 8/20, and this product launch lands two days ahead of it, which the market reads as an advance signal on cloud and AI monetization capability. Per Bloomberg, BABA has rebounded more than 40% from its June low.

Theme and stage: China AI applications — the fermentation phase, not yet crowded. The contrast with BIDU's plunge on the same day is itself evidence of divergence within the theme.

Fundamental verification: this issue did not perform primary financial verification on BABA; verification resources were prioritized for HD and KLAR, which reported today. Revenue, cloud growth, valuation percentile and the like were all not obtained — please do not cite unverified figures.

Pre-market and technicals: +2.97% on pre-market volume of 874,000 shares (the most solid quote on the bullish side of the list); the 52-week range is $91.99–$192.67 and the current price sits in the lower-middle of it.

Biggest risk: the 8/20 report is the falsification point. If front-running on the product launch meets a report that misses, the giveback will be fast; and China policy beta is high.

Final call: watch closely. It has the fullest pre-market volume and the freshest catalyst on today's bullish side, but there is a hard exam two days from now, and this issue did not complete its fundamental verification — it should not be upgraded to priority deep-dive until that verification is filled in.

---

⑥ COHR / LITE / AAOI / CIEN — the optical-module chain | Watch only

Pre-market (08:12 ET): COHR $329.99 (−6.04%, 286,000 shares), LITE $908.00 (−6.29%, 169,000 shares), AAOI $147.99 (−4.45%, 448,000 shares), CIEN $426.55 (−4.18%, 21,600 shares).

Logic: none of the four has news of its own; all are guilt by association with FN's guidance. But the two-day net basis paints a completely different picture: COHR +1.27%, CIEN −0.52%, AAOI −1.52%, LITE −1.96%. COHR rose 7.79% yesterday — that was front-running FN's report, and even after today's 6% decline it is still above its 8/14 level.

What this means: the market's medium-term pricing of the whole optical-module chain has not been overturned by FN; what is happening today is the unwinding of the "front-run FN's beat" position. What is genuinely being re-rated is the gross margin of the contract-manufacturing link (FN itself); whether module and component makers face the same cost squeeze has no primary evidence at all today.

Risk: the judgment above rests on "no new negative aimed at these four could be found." If the cost pressure on FN's call turns out to be industry-wide upstream price increases (rather than FN's own expense seasonality), then margins across the whole chain have to be revised down and this judgment is overturned.

Final call: watch only. Most of the decline is giveback rather than a breakdown, but the falsification risk is explicit and cannot be resolved today; wait for the FN call transcript and peer commentary.

---

⑦ MU / SNDK / WDC / AMAT — memory and semicap | Watch only

Pre-market (08:13 ET): MU $967.89 (−4.37%, 1.73 million shares), SNDK $1,694.26 (−5.06%, 1.12 million shares), WDC $504.05 (−5.96%, 240,000 shares), AMAT $510.74 (−4.68%, 131,000 shares), STX $939.00 (−5.61%, 70,300 shares), LRCX $326.95 (−4.91%, 74,100 shares), TER $423.20 (−4.50%, 30,000 shares).

Two-day net: SNDK +3.38%, TER +1.05%, AMAT +0.71%, MU −0.38%, WDC −0.93%, LRCX −1.63%, STX −3.54%.

Why they rose yesterday (the narrative source has been found): on 8/14 Musk reposted "memory, not compute, is the rate limiter of the Agentic era" and replied "Few realize this."; on top of that, SanDisk's 8/13 investor day gave FY2028–2030 non-GAAP gross margin of about 80% and operating margin of about 75%, and disclosed long-term agreements with 8 customers covering about 50% of FY2027 and two-thirds of FY2028 bits; and on top of that, multiple sell-side target raises (RBC raised SNDK to $1,600, Wells Fargo $1,550; New Street upgraded Micron to Buy/$1,250; BofA maintained Micron at Buy/$1,550 and gave FY2030 EPS of $200–250, 1.3–1.5x the market consensus of $160–170). Note that on 8/17 the broad US market was down for a second straight day and only memory was up — that is a "memory island" structure, not a broad sector rally.

Why they are falling today: three pieces of primary counter-evidence showing this is not an industry event

And the window of the selloff can be located precisely: at 04:59 ET Stocktwits' headline was still "MU, SNDK, Memory Stocks Buck Market Selloff"; by 07:25 ET Barron's had changed to "Micron Stock Drops. Why Bond Yields Are Hitting the Memory-Chip Boom." The selloff happened between 05:00 and 07:30 ET, a window in which all of Asia was closed and no pricing agency published anything; the only things moving were oil, yields and futures.

One ratio that serves as counter-evidence: on 8/17 the memory ETF (Roundhill DRAM) rose 3.39x as much as SOXX (+5.36% / +1.58%); in the 8/18 pre-market it is falling only 1.56x as much as SOXX (−5.60% / −3.60%). If this were genuinely a memory-specific pricing negative, that ratio should widen, not compress.

The real origin of the "equipment is decelerating" narrative is not today, it is AMAT on 8/14: AMAT posted a record FQ3 (revenue $9.1 billion +25%, non-GAAP EPS $3.50 +41%, Q4 guidance $10.25 billion ±$500 million, i.e. +51% YoY), yet still fell 6%–7% that day, because gross margin guidance was flat at about 50.4%, with Morgan Stanley saying "the revenue guide implies slowing system shipment growth" and UBS calling out "sequential deceleration." That day LRCX was only −0.9% and KLAC was flat, which shows it was an AMAT-specific event; but the "equipment growth will decelerate" narrative it left behind became a convenient reason to sell during today's rate shock. Primary commentary pointing the other way: Lam's CEO over the same period raised the WFE outlook to "the low $150 billion range for 2026," up from the prior $140 billion.

Structural explanation (this is the core judgment of the issue): memory sits in a "pricing cycle," equipment sits in a "capex cycle," and a rate shock hits duration. Memory revenue corresponds to contract prices that are rising this quarter; equipment revenue corresponds to capacity that only lands in 2027–2028. When the 30-year Treasury makes a 52-week high and Japan's long end makes a 30-year high, the discounting hit is distributed in decreasing order of duration — which precisely explains the layering in Asia that day: Japanese equipment names −5% to −6% and closing at their lows, while SK Hynix closed up +1.03%. Incidentally: equipment revenue speaks to future supply, not current demand — rising memory prices and falling equipment stocks are not logically contradictory.

Final call: watch only, not "avoid." The rationale has been upgraded from "no negative found" to "primary evidence pointing the other way was found." Still not upgraded to a buy, for three reasons: ① the pre-market decline is a real loss on positions and does not disappear because the attribution is macro; ② the long-end rate driver is still deteriorating today with no sign of a top; and ③ SK Hynix's intraday reversal from +8.8% to a +1.03% close shows that even Korean domestic money is cutting at this level — foreign net buying converged from roughly KRW 1.52 trillion intraday to nearly zero at the close, while institutions flipped to net selling of roughly KRW 1.19 trillion.

⚠️ A basis trap that must be spelled out: SK Hynix's closing +1.03% is survivorship bias and must not be read as "memory's resilience is confirmed." The +5.53% open was one jump handing over from the overnight US session, not an independent vote by the Korean market; relative to the open the close was −4.3%, so anyone executing at the opening price lost money today. By the same token, what matters tonight is whether SKHY (the ADR) can hold on a closing basis, not that pre-market jump.

---

8英特尔INTCAvoid | Pre-market $99.00 (−4.34%, 3.42 million shares, 08:12 ET) · Intel

One widely circulated misattribution has to be cleared up first: recurring-column pieces on several finance sites of the "Why Is Intel Stock Falling Tuesday?" type attribute today's decline to "the $20 billion offering priced at $95." That is old news. Checked against SEC EDGAR, Intel's related 424B5 prospectus supplements are dated 8/10 and 8/12, with the corresponding 8-K on 8/12, and there is no new Intel SEC filing today (8/18). The offering was priced and absorbed last week; it is not today's driver.

What is actually happening today: no primary single-stock news. Down −4.34% pre-market on 3.42 million shares, the largest pre-market volume of any name in this issue; two-day net −3.41% (yesterday +0.97%). The 52-week range is $22.77–$142.35 — extremely volatile.

Call: avoid. The reason is not fundamentals but trading structure: in a stock that has just completed a $20 billion offering, shows the largest anomalous pre-market volume in the list, and has no attributable news at all, the directional reliability of the pre-market gap is at its lowest. No news + the largest volume = the combination in which one should least establish a directional view before the open.

---

9Meta Platforms | Avoid | Pre-market $562.10META−1.21% · 290 · 000 shares · 08:12 ET

Related news: 8/18, federal court in Oakland, California, opening statements in the child-safety case⁸. The attorneys general of California, Colorado, Kentucky and New Jersey are seeking damages and demanding changes to how Facebook/Instagram operate; 29 states sued in 2023, with the remaining 25 to be tried later. According to reports, the states' damages figure is roughly $193 billion, while Meta's counsel cited a figure of $1.4 trillion. The jury is advisory, with the judge ultimately ruling. The trial is expected to run about 5 weeks.

Catalyst logic: this is not a one-off event but a source of sustained headline risk over the next 5 weeks. The opening-statement phase is typically the period of highest media exposure and the most likely to produce extreme headline numbers (such as the two damages figures above, which differ by a factor of 7). The mechanism of impact on the share price is mainly sentiment and an uncertainty premium, not near-term cash flow.

Technical position: META was already −3.54% yesterday (with this trial priced in to some degree); the 52-week range is $520.26–$790.80, and the current $562.10 is within about 8% above the 52-week low.

Final call: avoid. Five weeks of litigation headlines + already near a 52-week low + a large decline yesterday means there is no advantage in timing whatsoever. If one wants a long view, the reasonable entry point is the sentiment low after the opening-statement phase ends, not the day the trial opens.

---

10Reddit | Watch only | Pre-market $165.09RDDT+0.36% · 101 · 000 shares · 08:11 ET

Related news: S&P Dow Jones Indices announced that RDDT is formally added to the S&P 500 before the open on 2026-08-18, replacing AvalonBay (AVB, leaving on acquisition by Equity Residential).

This is a textbook "event realization" case: the inclusion was announced on 8/13 and RDDT surged that day; but on the last trading day before it took effect (8/17), RDDT fell 7.63% on volume of 37.85 million shares — passive money executes its buying on the effective date, while active money had already cashed in the expectation beforehand. Pre-market today it is +0.36%, essentially unmoved.

Fundamental background: according to reports, RDDT qualified under the S&P 500 profitability threshold by posting its first full year of positive GAAP net income in 2025; second-quarter 2026 revenue was about $805 million, +61% YoY. It is the second pure social-media company to enter the S&P 500 after Meta. These financial figures come from secondary reports; this issue did not check the primary filings.

Final call: watch only. The alpha of an index event was fully paid out on 8/13, the −7.63% on 8/17 was the realization, and today's inclusion leaves only mechanical buying. The 52-week range is $119.27–$282.95 and the current price sits in the lower-middle of it. Index inclusion is not a fundamental reason and should not serve as a basis for buying.

---

6. Bearish / Avoid List

---

7. Ranking Within Themes

Theme 1: AI optical modules and optical communications (direction: bearish)

Theme 2: BNPL / consumer credit (direction: bearish)

The key distinction in this theme: KLAR's problem is the cadence of growth and profit, not credit quality (its Q2 credit metrics improved across the board). Shorting AFRM/SEZL as a "BNPL credit blow-up" rests on the wrong premise — the correct association logic is the valuation-multiple anchor, not asset quality.

Theme 3: Memory and semicap (direction: neutral, no breakdown)

Theme 4: Defensive rotation (direction: bullish, but weak catalyst)

A basis warning for this theme: defensive names are broadly +0.4% to +1.0% pre-market, but pre-market volume is generally only a few thousand shares (VRTX 681 shares, JNJ 2,278 shares, ABBV 4,673 shares). Those quotes are not enough to prove money is flowing in; they only prove the selling pressure is not here. The reality of the defensive rotation has to be confirmed by volume in the first 30 minutes after the open; a pre-market quote cannot settle it.

---

8. Opening Verification Signals

Pre-market (observable as of 08:15 ET)

Three things to confirm intraday (in order of importance)

1. 08:30 ET housing starts / building permits (starts expected ~1.39 million vs. 1.427 million prior; permits ~1.38 million vs. 1.367 million prior). This is a direct input into HD's action today. A significant miss → HD's +1.95% is likely erased, and it drags on expectations for LOW's report tomorrow; a beat → combined with HD's report it creates a short-term long window in the home-improvement chain. Note the reaction function: under the current "good data = higher rates" framework, a beat on starts could be bullish for HD and bearish for the Nasdaq at the same time.

2. Whether equipment names rebound faster than memory. The core judgment of this issue is that "today's decline comes entirely from the discount rate, not fundamentals." If that judgment holds, then as soon as oil or yields pull back intraday, semicap (AMAT/LRCX/KLAC/TER, the longest duration) should rebound faster than memory (MU/SNDK/WDC, whose contract prices are rising this quarter) — because what fell was the discount rate, not the fundamentals. This is a falsifiable prediction; write it down before looking at the outcome. Conversely, if memory keeps underperforming SMH and equipment shows no elasticity, there is industry news this issue failed to find and the entire judgment should be overturned.

3. Volume confirmation in the defensive sectors. The pre-market gains in XLV/XLP/XLE rest on extremely thin volume. If in the first 30 minutes after the open the volume in these three sectors is not meaningfully above recent averages, then the "defensive rotation" is merely the absence of selling pressure, not the arrival of buying, and cannot be used as a basis for positions.

Options and sentiment

• KLAR, FN and BIDU, the three earnings names, all face IV crush today: the reports are out and implied volatility will fall quickly. Even if the direction is right, buying options can still lose money to IV compression — views on these three should be expressed through the common stock, not options.
• VIX at 15.67 is low, so hedging is broadly cheap, but that also means the market is underpricing the risks listed in this issue (a long-end breakout, geopolitics, a 5-week trial).

Main risks

• Reversal after a gap: FN (−10.2%), KLAR (−18.3%) and BIDU (−7.2%) have all gapped heavily, and pre-market gaps are frequently filled — do not chase.
• A lone gainer with no sector follow-through: BABA's +2.97% is the strongest on the bullish side, but watch whether KWEB / other China names follow; if only BABA rises, it is a single product event rather than the start of a theme.
• Distorted quotes on thin volume: the pre-market quotes for LOW (4,219 shares), VRTX (681 shares), JNJ (2,278 shares) and FN (34,700 shares) have limited reference value.
• Index futures vs. Treasuries divergence: today the two move together (stocks down, bonds down), so no divergence signal has appeared, which itself says rates are the dominant variable.
• The vacuum before NVDA's 8/26 report: the entire AI chain lacks a fundamental anchor before next Wednesday, making it easier for rates and sentiment to push it around.

---

9. Final Conclusions

① The 5 Names Most Worth Watching Today

② The 3 Strongest Themes Today

③ What to Avoid Today, and Why

1. Every company that reported this morning (KLAR / FN / BIDU) — all three have gapped heavily and all face IV crush. The direction may be right, but the entry timing is the worst possible.
2. Names with abnormal pre-market volume and no news (INTC being the archetype, at 3.42 million shares) — unattributable volume means a disorderly shareholder base, and pre-market direction is at its least reliable.
3. AI compute / IP names that genuinely broke down on a two-day net basis (ARM −6.70%, GFS −5.64%, ALAB −5.17%, AMD −4.88%, QCOM −4.67%) — unlike the memory chain, this group failed to follow when the sector surged yesterday; the relative weakness was already exposed during the rally, and today is simply continuation.
4. META — five weeks of litigation headline risk have only just begun, and it is already near a 52-week low, so there is no advantage in timing.
5. "Defensive rotation" names resting on extremely thin pre-market volume — the gains are real but the sample is too small; wait for volume confirmation after the open before discussing them.

④ The Final One-Line Judgment

Two independent things have been blended into one today. The first is real: with a report that beat on revenue but guided profit to a standstill, FN switched the pricing basis of AI optical modules from growth rate to gross margin — that is a fundamental re-rating. The second is false: the plunge in memory and equipment is not an industry event; DRAM/NAND spot prices were still rising on the day, the top-five NAND makers' second-quarter revenue was +77% QoQ, and the selloff happened precisely in the 05:00–07:30 ET window when Asia had closed and only oil and yields were moving — this is de-leveraging by duration, with equipment (capacity in 2028) hit hardest and memory (this quarter's contract prices) hit least. With the 30-year Treasury at a 52-week high and Dow futures near flat while Nasdaq futures are down 1.15%, the correct posture today is to wait for the 08:30 ET starts data and the two conference calls (KLAR 08:30, HD 09:00), rather than to misread a discount-rate shock as a cycle top in the first minute of a gapped open.

---

Data-Collection and Execution Failures (internal)

• yfinance was intermittently rate-limited this run: both sub-agents reported that the first call to .info / .fast_info worked, but subsequent .history() / .financials / .cashflow / .balance_sheet / .earnings_estimate calls all triggered YFRateLimitError (HTTP 429). HD's and KLAR's financial data were therefore taken 100% from primary SEC filings + stockanalysis.com, and valuation percentiles were computed in-house from 5 years of daily bars pulled directly via curl from the Yahoo chart API.
• The WebSearch quota was exhausted at 08:1x ET (200/200), leaving only WebFetch and sub-agents afterwards. Specific items affected: ① the actual 08:30 ET housing starts print could not be obtained (this issue treats it as "pending," which is the correct basis, not an omission); ② sell-side rating changes for the day could not be scanned systematically, so the "ratings" news category is empty here — which does not mean there were no rating changes today.
• The stockanalysis screener API (/api/screener/s/f) returned 404, so pre-market movers were instead found via a batch quote scan of a self-built universe of ~250 liquid names (work/0818/q.py); coverage is narrower than a full-market screen and may miss small- and mid-cap movers outside the list.
• The stockanalysis historical daily API (/api/symbol/s/<sym>/history) returned a structure that did not match expectations and failed to parse; the 8/14 closes used in the "two-day net" column were instead back-solved as "8/17 close ÷ (1 + 8/17 percentage change)," making them derived values rather than direct readings (the only error comes from rounding in the percentage change, on the order of <0.01%).
• The CNBC quote endpoint's @ES.1/@NQ.1 symbols are invalid; @SP.1/@ND.1/@DJ.1 (Sep'26 contracts) were used instead; @ER2.1 (Russell 2000 futures) and @EU.1 for .DXY could not be obtained.
• Four "old article as new news" traps were identified and removed (all discarded after a second verification pass):
  1. "Dayforce (DAY) up +26.7% pre-market" — DAY was taken private by Thoma Bravo at $70/share and delisted on 2026-02-04, so the stock no longer trades; this was a 2025-08 article.
  2. Benzinga's "Why Is Intel Stock Falling Tuesday?" attributing it to a $20B offering priced at $95 — per EDGAR, the relevant 424B5 filings are 8/10 and 8/12 and the 8-K is 8/12; there is no new filing today.
  3. The CNBC "Stocks making the biggest moves premarket" search hits were the 8/14 and 8/17 editions, containing items such as "Reddit up 12%" and "Lutnick opposes Apple buying Chinese memory" — none of them from today.
  4. ts2.tech's 8/18 wrap listed CleanSpark's $6.6B data-center lease as same-day news — that deal was actually announced on 7/14 and restated with the Q3 report on 8/6; CLSK is in fact −1.53% pre-market today.
• Sub-agent execution: all four returned and were used (HD earnings verification, KLAR primary-source verification, causes of the Asian semiconductor move, memory pricing-side investigation). The conclusions of the last two are mutually independent and consistent — both point to "no industry-side negative exists inside the window; the attribution is a macro duration shock" — and each got there via a different evidence chain (one through Asian closes and index-weight decomposition, the other through DRAMeXchange/TrendForce quotes and time-window location). That is the basis on which Section 7 was upgraded from "not found" to "counter-evidence found."
• Data-collection obstacles reported by the agents: Barron's (paywall), Invezz and some CNBC pages returned 403 to WebFetch, so only headlines or search summaries were obtained for those; each such item is labeled "headline-level / secondary" in the body. TrendForce's spot weekly is paywalled, so only summary-level figures were obtained.
• Old-article traps the agents caught and discarded (8 more beyond the 4 already listed in the body): ① "SanDisk Sinks 12%... CXMT IPO Rattles Memory Stocks" is actually a 7/27 article re-surfaced by the search engine; ② "Western Digital Sinks 16%" is from 8/6; ③ "Buy or Sell the Memory Plunge?" is from 7/28; ④ "Chip stocks shed more than $1 trillion" is from 7/29; ⑤ "Asian tech stocks drop with SK Hynix plunging 10%" is from 8/6; ⑥ "SK Hynix Drops Over 8%, Samsung Falls Nearly 6%" directly contradicts the 8/18 readings (SKH +1.03%, Samsung −2.19%) and was judged an old article; ⑦ Macquarie's KRW 2.9 million target on SK Hynix is 5/15 news; ⑧ 24/7 Wall St.'s "Here Are Tuesday's Top Wall Street Analyst Research Calls" is indeed dated 8/18 but is a recurring column; the names it called out that day were verified to be Abercrombie & Fitch, America Movil and Bath & Body Works, with no memory or semiconductor names at all — the headline cannot be used to infer the content.
• A structural fact: TrendForce's spot weekly is updated on Wednesdays (the two most recent editions being 7/29 and 8/12, with the next on 8/19), so on 8/18 that weekly structurally cannot exist — if anything says "the 8/18 TrendForce weekly shows…," it has to be a mixed-up edition.
• The risk-auditor QC pass was not run: under the double constraint of an exhausted WebSearch quota at 08:1x ET (200/200) and the publishing window (must be done before the 09:30 ET open), this issue instead explicitly labels every secondary source, unverified item and unclosed assumption in the body (Section 5 ⑤⑥⑦, the basis warnings in Themes 3 and 4 of Section 7, verification point 2 in Section 8, and the Japan-Korea circular-reference warning in Section 1b). This is a fallback, not the equivalent of independent QC; the next issue should restore it.
• Verification this issue did not complete (the honest cost): ① BABA's fundamentals and valuation were not verified at all; ② the fundamentals of the optical-module chain (COHR/LITE/AAOI/CIEN) were not verified; ③ BIDU's segment figures are authoritative-media relays and were not checked against its IR original / 6-K; ④ FN's segment split and customer concentration come from media relays of the earnings presentation and were not checked against the presentation itself; ⑤ sell-side rating changes for the day could not be scanned systematically.

---

⚠️ Risk disclaimer: this list is a pre-market information review and observation only and does not constitute investment advice. US equities carry high volatility and pre-market gap risk, and post-earnings IV crush and guidance reversals are common; automatically generated content may contain stale information or factual errors. Company disclosures / SEC filings prevail, and this cannot be used directly as a basis for trading.