U.S. Pre-Market Brief | 2026-08-06 (ET) Thursday

Coverage window: 2026-08-05 (Wed) 16:00 ET regular-session close → 2026-08-06 08:10 ET (20:10 Beijing time on August 6). Scope: Wednesday after-hours earnings (16:00–20:00 ET), the overnight Asian and European sessions, and this morning's pre-market (including pre-market earnings released before 08:10 ET). All timestamps are Eastern Time (ET). Single-stock pre-market quote snapshots were taken at 07:59–08:05 ET, sourced from stockanalysis.com; pre-market quotes move constantly — actual opening prints govern.

---

0. One-Line Summary for Today

1. Strongest catalyst = SanDisk's "big beat but not-good-enough guidance" detonating the global memory chain: SNDK Q4 revenue $8.965B (8.0% above consensus), adjusted EPS $39.25 (14.6% above consensus), gross margin 84.6% — a record — plus a new $14 billion buyback authorization; but Q1 FY27 revenue guidance of $10.30–10.80B runs against a consensus that is conflicted across three sources ($10.62B / $10.82B / $11.16B), implying gaps of 0.7% / 2.5% / 5.5% — this report adopts no single value, and confirms only the direction: the midpoint of revenue guidance is below market expectations. That one line drove a stock up roughly 500% YTD and up 33x over one year down −10.26% pre-market. This is the origin point of every trade today.
2. Strongest theme = the second wave of the memory/DRAM chain ebbing (the strongest downside theme): WDC −14.70% pre-market, SNDK −10.26%, STX −5.27%, MU −4.96%; overnight the Korean KOSPI closed at 6,296.38, down 301.88 points (−4.58%), with SK Hynix −10.37%, Samsung Electronics −6.30%, and Kioxia around −10%, while the Nikkei fell 617 points. Note: this is not the first crack — the AI memory trade already broke once on July 28–29, and harder than most people remember: the KOSPI fell −10.8% in a single day on 7/28 to close at 6,023.66, evaporating more than 18% over two days, making July the worst single month on record for the KOSPI (worse than −30%); Samsung −13.4%, SK Hynix −14.7%. Today is the second leg after the bounce, so the theme stage should be judged as ebbing, not launching.
3. ⚠️ The single most counterintuitive — and most revealing — name today is WDC, not SNDK: what Western Digital released after the close on 8/5 was its own earnings (it was not dragged down by SNDK): revenue $3.75B (+44% YoY) beat, adjusted EPS $3.56 beat (consensus $3.27–3.29), Q1 FY27 guidance of $4.00–4.20B revenue and $3.85–4.15 EPS both above consensus, and it even raised the quarterly dividend to $0.15 — and it still fell harder than SNDK, whose guidance missed (−14.70% vs −10.26%).
4. So today's real trading rule is not "the beat wasn't enough," it's "even beat + raise isn't enough": what the market is selling is the second derivative. WDC's sequential revenue growth went 12.3%→9.4%, and gross-margin expansion went +390bp→+110bp; SNDK's sequential revenue went +97%→+51%→guidance +18%, and gross-margin guidance turned to sequential contraction for the first time. Absolute levels are all at records; the acceleration has all rolled over — after gains of roughly +500% / +200%+ YTD, that is enough to trigger deleveraging-style multiple compression. Under the same rule: FIG revenue +48% beat with a full-year raise → −14.03% pre-market; HUBS double beat but Q3 guidance 1.8% light → −23.26%.
5. Driver type = earnings/guidance overwhelmingly dominant + the geopolitical peace trade continuing + macro still pending: the U.S.–Iran Hormuz agreement is reportedly in its "final stage," and Brent has retreated from wartime $100+ to around $79; 08:30 ET initial jobless claims (consensus roughly 202,000–203,000) and Q2 productivity / unit labor costs have not yet been released, and tomorrow, 8/7 at 08:30 ET, is the July nonfarm payrolls report.
6. Pre-market state: Dow futures +129 points (+0.24%), S&P +11.25 (+0.15%), Nasdaq −129.5 (−0.44%) — the index divergence is itself today's conclusion: the Dow made a record high yesterday at 54,349 (+0.49%), while the Nasdaq fell 0.83% the same day. The 10-year Treasury at 4.626% (+0.9bp), VIX 15.81 (8/5 close), gold $4,328.50 (+$23.30), WTI around $75.7, Brent around $79.1–80.0.
7. ⚠️ Today's second certainty event: the SPCX lockup expiry is today. Roughly 911.5 million shares (≈ 6.9% of existing shares outstanding) enter the eligible-to-sell pool, with the first tranche capped at 20% of that. SPCX closed −13.61% at $108.27 yesterday (volume 184 million shares = 4.0x the 10-day average), just 3.3% above its 52-week low of $104.83; yet this morning it is +2.75% pre-market at $111.25 — what showed up on day one of the unlock is absorption, not a stampede (as of 08:11 ET) — but the supply gate stays open all the way to 12/8, making this the last name anyone should size up directionally today.
8. What is being sold in SPCX is the cash flow statement, not the income statement: Q2 revenue $7.814 billion beat by 14.7%, gross margin 55.3%, loss narrowed 46% YoY — but single-quarter capex of $18.382 billion (235% of revenue, 6.5x YoY) came in roughly 41% above sell-side expectations, driving free cash flow to −$15.963 billion.

⏱ Snapshot-convention note (applies throughout): single-stock quotes in this report were snapped at different moments between 07:59 and 08:19 ET, marked individually in each place; there is no "single unified convention" across the report. Sources are the CNBC quote API and stockanalysis.com.

⚠️ Pre-market drift is material, not a rounding issue: during the review pass, AMD went from −0.84% to −1.38% and HONA went from −13.1% to −17.0% (breaking below its 52-week low), in under 20 minutes. Every quote here predates the 08:30 ET initial claims and productivity data, which will reset all pre-market pricing — before the open you must go by live quotes; no price in this report may be used directly to place an order.

⚠️ One trap you must avoid: several media outlets this morning reported SNDK −5.40% / WDC −5.36% as "pre-market declines" — those two numbers are actually the 8/5 regular-session closing declines (which happened before earnings were released). The true pre-market declines are SNDK −10.26% / WDC −14.70%, nearly double. This report has corrected everything to the latter.

---

0.6 ⏰ Post-finalization addendum: the 08:30 ET macro data has been released; every quote in this report predates it

Single-stock quote snapshots in this report were taken at 07:59–08:19 ET, whereas the 08:30 ET initial jobless claims, Q2 nonfarm productivity and unit labor costs have already been released.

• This report could not obtain reliable actual values: the conventions retrieved conflict with one another (productivity consensus is +0.6% in one place and +1.9% in another; actual values are cited as +2.4% in one place, with unit labor costs +1.6%), and it could not be confirmed that these numbers correspond to this particular release. This report therefore cites no actual values, and only flags for readers:
• ⚠️ This means every pre-market quote and gap% in Sections 3, 5 and 6 may already be stale. Productivity and unit labor costs bear directly on inflation and the Fed path, and are data capable of resetting pre-market pricing market-wide.
• Before the open, please re-verify every gap%-based judgment in this report against live quotes, especially the confirmation signals in Section 8 that are premised on pre-market prices.
• Tomorrow's July nonfarm payrolls at 08:30 ET on 8/7 is this week's real macro test.

---

0.5 Data This Report Could Not Obtain / Where Sources Conflict (the cost of honesty)

Listed so readers know the boundaries of this report:

---

1. News Overview

---

2. Strongest Themes, Descending

---

3. Overall Single-Stock Strength Ranking

Scores are in Section 4. The direction column states bullish or bearish explicitly. Pre-market quote snapshots 07:59–08:05 ET.

3A. Bullish / long-watch zone

3B. Bearish / avoid zone (detailed table in Section 6)

---

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

Convention notes and two self-corrections to the scoring system:

1. FIG has been removed from the §3A long-watch zone and is retained only in §3B and §6. Reason: its risk-deduction factors actually exceed the −15 cap, and the 60-point score after truncating at the cap is a systematic overstatement; moreover the same name also appears in the avoid zone, and appearing on both the long and short lists simultaneously is a self-contradiction in the scoring system that cannot be patched with a footnote. Its true score is about 57.
2. UAL/DAL have been removed from §3A and are retained only as watch items under Theme 4 in §7. Reason: catalyst directness for that entry is "low (pure macro, indirect)," fundamentals are "not verified," the pre-market quote was "not obtained," and the agreement has not been signed — if four pending items can still make the list and receive 54 points, the scoring system loses all meaning. They do not participate in scoring.
3. This table lists only names with a complete component-level source basis. The scores for AVGO (52), EBAY (55) and DUOL (46) lack component detail and should be treated as a rough ordering rather than a traceable score.

---

5. Detailed Analysis of Top Names

① NVDA (英伟达 / NVIDIA) — priority deep-dive | 83 pts

• Related news: the 08-04 SpaceX earnings call (TechCrunch / Barchart) — Musk stated explicitly that SpaceX's ground and in-orbit AI infrastructure will exclusively adopt the NVIDIA Vera Rubin architecture, and that it is working with NVIDIA on "Starmind" in-orbit compute satellites (carrying Rubin GPUs + Vera CPUs). NVDA closed 8/5 at $219.22 (+3.43%).
• Catalyst logic: this is a catalyst for order visibility, not for current-quarter EPS. The value of the word "exclusive" lies in exclusion — it is simultaneously a negative for AMD and Intel. Its impact on valuation exceeds its impact on near-term revenue.
• Theme and stage: the AI compute mainline is in a divergence phase — the entire memory chain is being sold off today while NVDA turns green against the trend, which says money is moving out of AI's "cyclical goods" (memory) and into AI's "monopoly goods" (GPU architecture). This is the most important structural judgment in this report.
• Fundamentals verification (re-checked against primary statements): market cap $5.305 trillion; latest quarter (FY27 Q1, quarter ended 2026-04-30) revenue $81.615 billion (+85.2% YoY), gross margin 74.93%, operating margin 65.6%; TTM revenue $253.491 billion, TTM EPS $6.53; TTM free cash flow $119.076 billion, FCF margin 47.0%; capex only 2.6% of revenue (extreme asset-light with no fabs); net cash $68.224 billion, debt/equity only 4.33%; ROE 114.29%. Receivables DSO fell from about 51 days to 45 days (improving collections), and inventory growth is broadly in line with revenue with no sign of stockpiling.
• ⚠️ One red flag on earnings quality (which must be stated): of the quarter's $69.903 billion in pre-tax income, "other non-operating income" accounted for $15.929 billion (22.8%), up sharply from $5.604 billion sequentially; over the same period long-term investments/AFS securities jumped from $22.251 billion to $43.364 billion. This portion is mark-to-market gains on outside AI-ecosystem equity investments, not earned by the core chip business, and cannot be extrapolated; excluding it, the core net margin is about 55% (an estimate). This also explains why FCF/net income is only 74.6%.
• Moat: the CUDA ecosystem + generational architecture lock-in + now an added "exclusive supply" paradigm of customer binding.
• Pre-market and technicals: +0.99%~+1.14% @ $221.60 (08:03–08:11 ET); 52-week range $164.07–$236.54, 7.0% from the high, 13.2% above the 200-day moving average; +15.4% over the last 5 sessions, +21.8% over the past year. Next earnings 8/26, so no IV crush risk currently.
• Valuation percentile — three numbers that must be read together, you cannot cherry-pick one: P/E (TTM) 33.6x, forward 21.9x, P/S 20.9x, P/B 27.1x. Against its own history it sits at the 15th percentile (low) over the last 3 years, the 38th percentile over the last year (neutral to low), but already the 81st percentile over the last 3 months (the expensive end of the recent range). The valuation center has indeed stepped down for 4 consecutive fiscal years (112x→52x→41x→39x→31x), but on a short horizon it is already at the upper edge of the last three months.
• Final judgment: the only "safe haven" in the AI complex today. But note three things: ① it is up 5 sessions in a row for +15.4% and sits at the top of its 3-month valuation range; ② the 21.9x forward multiple depends entirely on the sell-side fEPS of $10.01 being delivered, with the checkpoint on 8/26; ③ if memory-chain panic spreads to the entire semiconductor complex today, NVDA will struggle to stay immune. Priority deep-dive, but not a chase-the-highs entry.

② SNDK (SanDisk) — avoid | the most important case study in this report

• Related news (primary SEC 8-K): 08-05 16:05 ET. Q4 FY26 revenue $8,965M (8.0% above consensus), adjusted EPS $39.25 (consensus $34.24, 14.6% beat), gross margin 84.6% (a record), data center revenue $2,977M (vs $213M a year ago, 14x), a new $14 billion buyback authorization (remaining authorization $15.5 billion ≈ 7.4% of market cap). Q1 FY27 guidance: revenue $10,300–10,800M, non-GAAP EPS $44.00–46.00, gross margin 83–85%.
• Catalyst logic (key, corrected twice): this is an excellent earnings report, and what killed it is revenue guidance below market expectations. But "below by how much" cannot be settled here — consensus exists under three conventions of $10.62B / $10.82B / $11.16B, implying midpoint gaps of 0.7% / 2.5% / 5.5%, and not one source names its provider (LSEG/FactSet/Visible Alpha…) or its cut-off time. ⚠️ The draft of this report repeatedly cited $10.62B / "a 0.7% gap" as established fact; that practice has been retracted. More importantly: even the direction of the EPS guide is contested (a beat under the $44.21 convention, a miss under $45.58) — "revenue guidance is weak but the earnings guide may actually be ahead" is a completely different sentence from "an across-the-board miss."
• What is really being sold is the third derivative (the core judgment of this report): ① three consecutive declines in sequential revenue growth: +97% → +51% → guidance +18%; ② gross margin turns for the first time: 78.4% → 84.6% (peak) → guidance midpoint 84.0%, −0.6pt sequentially, the first non-expanding gross margin guide of this pricing cycle; ③ sequential adjusted EPS: +68% → +14.6%. Absolute levels are all at records; the acceleration has all rolled over.
• The +372% YoY figure has been verified as purely organic: SanDisk completed its spin-off from WDC on 2025-02-21, so the comparison base is already a standalone-company statement, with no acquisitions and no change in reporting basis; the quarterly series is continuous with no discontinuities. Company's own breakdown: of the +51% sequential gain, about 1/3 came from shipment volume and 2/3 from price increases. Over the same period Consumer was −32% sequentially — the company is actively cutting consumer-grade capacity to feed the data center.
• Theme and stage: the second leg of the ebb. The AI memory trade already broke once on July 28–29 (at that time the KOSPI fell −10.8% in a single day on 7/28, evaporating over 18% in two days and triggering circuit breakers on two consecutive days for the first time on record; Samsung −13.4%, SK Hynix −14.7%, driven by China's CXMT completing Asia's largest IPO to expand DRAM capacity, domestic Chinese lithography equipment, and doubts about the sustainability of AI demand). Today is the second leg after the bounce, not the first crack.
• Fundamentals verification: net cash $4.76 billion, zero interest-bearing debt (it repaid $1.9 billion within a year); FY26 revenue $20.248 billion (+175%), GAAP net income $11.433 billion, ROE about 92%; operating cash flow $11.671 billion, OCF/net income = 102%, with excellent accounting quality. Extreme operating leverage (a double-edged sword): of the $3.015 billion of incremental revenue, 97% dropped straight to operating profit — because the increment is almost entirely price, with near-zero marginal cost; which means that once prices fall back, the pass-through rate for declining profit is likewise 97%, and that is this name's biggest structural risk.
• ⚠️ Two convention traps that must be spelled out: ① reported capex of only $177 million (0.87% of revenue) is physically impossible for a NAND maker — capacity investment runs through the Flash Ventures joint venture with Kioxia, so the 56.8% FCF margin and 18x P/FCF systematically overstate true capital intensity and cannot be used directly for valuation; ② this quarter's GAAP EPS of $43.97 is actually higher than non-GAAP $39.25 (below-the-tax-line differences), so comparisons to consensus must use $39.25; using 43.97 inflates the size of the beat.
• Valuation (with the "cheap" judgment corrected): P/E (GAAP TTM) 18.3x, and a forward P/E of only 6.7x annualizing Q1 guidance. ⚠️ But a 6–7x forward P/E is not evidence of "cheap," it is a textbook cyclical-top characteristic — the rule for commodity-like memory is "peak earnings = trough P/E," and when gross margin is 84.6% (versus 26.2% a year earlier), the denominator is a cycle extreme rather than a normal. The more credible anchors are P/S 10.3x and P/B 13.3x, which for a NAND company that historically traded below 1x P/S for long stretches remain at an extremely high historical percentile. PEG is invalid here, and the +372% driven by price cannot be treated as a sustainable g.
• The hardest piece of evidence on the bull side: Q4 contract liabilities +$731 million, corresponding to prepayments from NBM long-term contract customers; the company says it signed 5 in April and another 5 this time (including 3 new customers). Customers are locking volume, locking price and prepaying, which is the weightiest counter-evidence in the "is the price increase sustainable" debate.
• Pre-market and technicals: −10.26% @ $1,212.00 (it had already fallen 5.40% to $1,350.50 on 8/5). 42.6% below the 52-week high, but still +3,232% above the 52-week low (33x in a year). Multiple banks cut price targets into the $1,300–2,100 range (a factual statement; sell-side price targets are not a basis for judging share-price direction).
• Final judgment: avoid (do not participate today). The short side is equally crowded — net cash with zero debt plus $15.5 billion of remaining buyback authorization means bounce risk must be built into scenarios. But please also do not catch it because "P/E 18 is cheap": that denominator is computed from peak-cycle earnings. This is a stock whose fundamentals stand up but whose cycle position is questionable; what it needs is time and confirmation from NAND contract prices, not a directional bet.

③ HUBS (HubSpot) — avoid / short watch | today's cleanest negative

• Related news: 08-05 16:05 ET. Q2 revenue $911.7M (+19.8%, consensus about $898M), adjusted EPS $3.26 (consensus $3.02), a double beat; but net customer additions of about 7,000, far below the 9,000–10,000 target range; Q3 guidance of about $924.5M, roughly 1.8% below the $942.27M consensus; full year cut. Oppenheimer downgraded.
• Catalyst logic (the draft understated the severity): this is a double hit on guidance and operating metrics. ⚠️ The phrase "Q3 guidance 1.8% light" actually understates the real change: 1.8% is only the dollar gap, while the revenue growth implied by guidance falls from Q2's 20% (17% in constant currency) to about 14–15%, a 5–6 percentage-point growth cliff. Management's stated reasons are specific: ① the shift toward trial-based / outcome-based pricing for AI agents is slowing deal decisions; ② the enterprise software demand environment turned cold abruptly, with CFOs demanding more proof of value and mid-size deals gaining extra approval layers.
• ⚠️ One counter-fact that conflicts with the bear narrative but must be listed: HubSpot also authorized a $1 billion buyback. You cannot present only the parts that suit the narrative.
• Theme and stage: the origin point of Theme 2. What makes this narrative damaging is that it is demand-side, not an execution problem at HubSpot alone.
• Fundamentals verification: P/E 90.05, TTM revenue $3.45 billion (+21.1%), market cap already well off its highs (52-week range $169.63–$525.51, with the current pre-market $192 approaching the 52-week low).
• Pre-market and technicals: −23.26% @ $192.00. Note: it is already down 23%, so the risk/reward for shorting is already poor, and it is only 13% above the 52-week low of $169.63.
• Final judgment: avoid. Shorting is watch-only — after a 23% gap, this is not a place to establish a short. Its real value is as an early warning signal for NET/DDOG/TTD/TWLO tonight.

④ WDC (Western Digital) — avoid | the most counterintuitive name today, and the most important passage in this report

• Related news (primary SEC 8-K): 08-05 16:05 ET, and what WDC released was its own earnings, not a spillover from SNDK. Q4 revenue $3,747M (+44% YoY, +12.3% QoQ) beat (consensus about $3.69B), adjusted EPS $3.56 beat (consensus $3.27–3.29), non-GAAP gross margin 54.4% (+13.1pt).
• ⚠️ The guidance was also a beat: Q1 FY27 revenue guidance of $4,000–4,200M (consensus $4,010M, 2.2% higher), EPS $3.85–4.15 (consensus $3.81, 5.0% higher), gross margin guidance of 55–56%, still expanding; and it raised the quarterly dividend to $0.15 (from $0.12).
• So the question becomes: a beat on results + a beat on guidance + a dividend increase, and why is it down 14.70%, worse than SNDK whose guidance missed? Two things stack up:
  1. The second derivative rolled over too, just more gently: sequential revenue +12.3% → guidance +9.4%; non-GAAP gross-margin expansion +390bp → +110bp; sequential EPS +31% → +12%. The absolute size of the beat does not match the acceleration already implied by a +218.67% YTD move.
  2. (This one matters more, and the market may have priced it only partly) the GAAP profit engine is used up: of FY26 GAAP net income of $9,424M, $6,498M (69%) came from a non-cash revaluation gain on the "retained SanDisk equity stake" (SanDisk rose 33x in a year, and WDC booked the paper gain into its income statement along the way); and that line item was already zeroed out on the 2026-07-03 balance sheet (it was $354M a year earlier). The source of that income is gone.
• ⚠️ So "WDC's P/E is only 21x" is a fake number: the headline P/E of 21.4x is entirely polluted by that one-off non-cash paper gain, and stripping it out gives a non-GAAP TTM P/E of 50.8x (2.4x the headline), with a forward P/E of about 27.7x annualizing Q1 guidance. From FY27 onward GAAP and non-GAAP will converge sharply, and GAAP EPS faces a technical cliff. This is probably the single number most likely to be written down wrong today.
• Actual operating quality is in fact very good: operating margin went from 24.5% in FY25 to 34.5% in FY26, with 41.7% in Q4 alone, and Q1 guidance implying about 45.9% — HDD operating leverage is real and still expanding, just at a slower pace. Cash flow tells the same story: OCF/GAAP net income is only 42% (which looks like a warning), but OCF/non-GAAP net income is 101% — the 42% is purely because that $6.498 billion paper gain is non-cash, and stripping it out leaves it entirely healthy.
• The balance sheet is the most thoroughly repaired of the three: interest-bearing debt is only $1.052 billion (with zero long-term debt), versus $4.711 billion a year ago; net cash +$527 million, versus net debt of −$2.597 billion a year ago, a full deleveraging within a year; convertible bonds have been settled in cash and preferred shares converted to common.
• Pre-market and technicals: −14.70% @ $442.85 (it had already fallen 5.36% on 8/5). Down about 19% over two days; 35.1% below the 52-week high, still +610% above the 52-week low. Mizuho price target $590←$685, UBS $525←$560 (factual statements).
• Final judgment: avoid. But note that the core message of this passage is not "should WDC be shorted," it is: when a company's results, guidance and dividend are all good and it still falls 14.7%, that tells you the market is not pricing this quarter, it is pricing the valuation assumption for the entire memory chain. In this environment, "good earnings" is not a reason to buy any memory stock.

⑤ SPCX (SpaceX) — watch only | today's only supply event with a fixed date

• Related news: ① the first post-listing earnings report after the close on 08-04: Q2 revenue $7.81B (+92%), net loss $541 million, capex $18.37 billion; ② Musk announced the exclusive NVIDIA adoption; ③ the first lockup expiry is today, 8/6: roughly 911.5 million shares enter the eligible-to-sell pool, with the first tranche capped at 20% of that (about 182.3 million shares near term), with smaller subsequent tranches continuing into October and a full backstop of 2026-12-08; ④ Morgan Stanley's Adam Jonas reiterated Overweight with a $300 price target.
• ⚠️ Shares actually sellable today: two readings coexist, and this report adopts no single value. This is the most important convention warning in this entry:

  Convention	Sellable today	Denominator (public float)	Relative to float
  Motley Fool: the 911.5 million shares are the eligible pool, of which 20% is sellable today	about 182.3 million shares	below 280.1 million shares	about 65%
  Forbes: the 911.5 million shares are themselves today's unlock tranche	911.5 million shares	about 639 million shares	about 140%
  The two sources disagree both on what "911.5 million" means and on the free float (280.1 million vs 639 million), a gap of a factor of 5. Until the lockup terms are verified against the original S-1/424B4, this report claims neither is correct and does not quantify today's selling pressure on that basis. ⚠️ The draft of this report asserted that "what is actually sellable is 20% of it" and used that to "correct the media"; that assertion has been withdrawn.

  Table scrolls horizontally
• One clause detail (which holds under both readings): the first release ratio is 20%, but rises to 30% if the share price is more than 30% above the offering price. SPCX is currently near its 52-week low (offering price $135, current price $111), so the lower 20% tier applies — share-price weakness has, ironically, lowered today's release ratio.
• The subsequent cadence (per Forbes): about 7% released at each of 70/90/105/120/135 days, plus up to a further 28% after Q3 earnings; by 12/8 a cumulative maximum of about 40% of company shares will be tradable, while the remaining roughly 60% — including Musk's — is expected to stay locked until mid-2027. ⚠️ The draft described 12/8 as the "full-unlock backstop," which was imprecise and has been corrected. The supply pressure is a continuous multi-month event; do not treat today as "the negative being cleared in one go."
• Fundamentals verification (re-checked): market cap $1.425 trillion. The income statement is actually a beat: revenue $7,814M beat by 14.7% (consensus about $6,810M), gross margin 55.3% (versus 43.9% a year earlier), operating loss narrowed from −$970M to −$143M, and net loss narrowed 46% YoY.
• ⚠️ What is being sold is the cash flow statement, not the income statement: single-quarter capex of $18,382M = 235% of revenue, 6.5x YoY, about 41% above sell-side expectations (of which roughly $15.83 billion went to AI, per a media convention still to be verified), driving free cash flow to −$15,963M (an FCF margin of −204%). This is the only case among the three where results beat and the stock was sold on the cash flow statement.
• Cash runway: cash and short-term investments $100.009 billion, net cash +$60.645 billion; at Q2's −$15.963 billion per quarter of FCF, that is about 6.3 quarters. But the capex trajectory is an accelerating curve of $6.9 billion → $8.9 billion → $18.4 billion. This is not a liquidity risk (a $1.4 trillion market cap has enormous refinancing capacity), it is the valuation risk that "shareholder returns keep being diluted by capital spending."
• The growth story and the uncertainty are the same thing: the first segment disclosure shows AI Solutions & Infrastructure going from $475M to $2,194M (+362%) in one quarter, contributing 55% of the entire sequential increment in a single quarter — but it has only two quarters of history. Over the same period net PP&E rose 52% in half a year to $66.854 billion, corroborating the AI data-center buildout narrative.
• Valuation: P/E is not applicable (it is loss-making, and the IPO weighted share count is distorted: Q2 EPS implies a weighted share count of about 6 billion shares versus 13.16 billion currently, so the circulating figures of PE −159x / EPS −0.68 are worthless as references). Only aggregate multiples are usable: P/S (TTM) 61.8x, P/S (Q2 annualized) 45.6x, EV/EBITDA about 234x. Even after halving 52% from the high, 45–62x P/S against a −1.8% operating margin and −$16 billion of quarterly FCF remains in extreme territory.
• Pre-market and technicals: it closed 8/5 at $108.27 (−13.61%, volume 184 million shares = 4.0x the 10-day average), 52-week range $104.83–$225.64, 52.0% below the 52-week high and only +3.3% above the low; this morning +2.75% pre-market @ $111.25 (pre-market volume already 30.75 million shares) — what showed up on day one of the unlock is absorption, not a stampede.
• Final judgment: watch only. Both sides have hard arguments: the bears have the certain supply that opens today and that supply run continues to 12/8; the bulls have the 20% tier (rather than 30%) + the MS $300 target + being only 3.2% above the low. This is the last name anyone should guess direction on today. ⚠️ A specific reminder: do not read today as "the unlock negative being cleared in one go" — on the cadence above, there are still 5 more time tranches plus up to 28% after Q3 earnings.

⑥ DASH (DoorDash) — watch closely (do not chase) | today's strongest standalone long

• Related news: 08-05 16:05 ET. Q2 revenue $4.45B (consensus $4.34B), marketplace GOV $33.1B (+36% YoY); Q3 guidance GOV $33–34B, adjusted EBITDA $950 million–$1.10 billion; multiple banks raised price targets. EPS $0.46 vs consensus $0.47, a slight miss — which is not contradictory with the revenue beat; it is a "revenue beat, EPS slightly light" combination.
• Catalyst logic: completely decoupled from the AI narrative, which is its greatest value today. It also benefits from falling oil prices (delivery costs).
• Fundamentals verification: market cap $90.3 billion, P/E 108.68, 2025 revenue $13.72 billion (+27.93%), net income $935 million (+660%). 52-week range $143.30–$285.50.
• Pre-market and technicals: +6.14% @ $220.00. It has already gapped 6%, and a P/E of 108 leaves very little margin for error.
• Final judgment: watch closely, do not chase. Only a gap-and-go with volume confirmation would be a valid signal; if the gap gets filled at the open, it says the market has no incremental money for expensive growth stocks today.

⑦ HONA (Honeywell Aerospace) — avoid / short watch | today's only genuine guidance cut

• Related news (corrected to "double miss + guidance cut"): 08-05 16:05 ET. Q2 revenue $4.52B (+5%), below the roughly $4.61B consensus; adjusted EPS $1.87, about 12% below the $2.12 consensus — a miss on both revenue and earnings, not in-line. ⚠️ The draft listed only the actual values without the consensus, which read as if it were in line; that has been corrected. There were also about $100 million of spin-related costs and inventory write-downs, with core profit −7% YoY.
• The guidance portion (all three items verified): full-year organic growth cut from 7–9% to 4–5% (reason: supply chain constraints); EBITDA lowered from $4.65–4.75 billion to $4.35–4.45 billion; standalone-basis full-year adjusted EPS of $7.60–7.90, far below the street's $8.90 (a gap of about 13–15%). The downgrade came from the CNBC Investing Club (Cramer's Charitable Trust), not from CNBC as a media outlet issuing a rating — the draft's wording has been corrected.
• Catalyst logic: this is the only case today where both results and guidance deteriorated substantively (SNDK's guidance merely fell short of elevated expectations, and WDC's guidance actually beat — entirely different in kind).
• Pre-market and technicals (⚠️ already broken down, missed in the draft): pre-market $169.00, about −17.0% (re-checked at 08:19 ET; the draft wrote −13.1% @ $176.99, so the quote has drifted materially). 52-week range $192.03–$297.50 — the current price is already about 8–12% below the 52-week low. This report specifically calculated that HUBS is "only 13% above its 52-week low" while omitting that HONA has already broken down, which is selective presentation and is corrected here as well. The spin-off completed in June 2026, making it a recently independent entity with no historical valuation reference.
• Final judgment: avoid. On the dimension of "how crowded the short narrative is," its fundamental basis is harder than HUBS's (double miss + a genuine guidance cut vs a guidance-only miss); but it is already down 17% and has just broken down, so the risk/reward of chasing the short is equally poor.

⑧ XYZ (Block) — watch closely | today's most overlooked long

• Related news: 08-05 16:05 ET. Q2 revenue $6.618B (consensus $6.485B), adjusted EPS $1.02 (consensus $0.87, a 17% beat); full year raised: gross profit → $12.51 billion (+21% YoY), adjusted operating income → $3.47 billion (28% margin), raised by $180 million and $130 million respectively.
• Catalyst logic: a double beat + a double raise, the only large-cap to do both today. But the after-hours reaction was choppy (up as much as +3.1% before fading, with another source citing +4.5%), which says it is not fully priced — and that is precisely where the expectation gap is.
• Final judgment: watch closely. On a day when "a beat is not enough," it is one of the few names that beat and raised; but fintech is sensitive to risk appetite, and if the broad market weakens on memory-chain panic, it will not hold up independently.

⑨ FIG (Figma) — watch only | today's most extreme "good results, bad stock"

• Related news: 08-05 16:05 ET. Q2 revenue $370.1M (+48%), beating $351.6M; EPS $0.08 (consensus $0.04); gross margin 85%, NDR 136%; full year raised to $1.463–1.467B (+39%). RBC raised its price target to $28←$22.
• Catalyst logic: the fundamentals are beyond reproach, and the stock is being dragged along wholesale by Theme 2 (software multiple compression). Pre-market −14.03% @ $24.20, already below the $28 target RBC just raised to.
• Fundamentals verification: market cap $14.87 billion, TTM revenue $1.28 billion (+43.4%); net income −$1.57 billion, EPS −3.18 (a recent IPO, including large stock-based compensation); Fwd P/E 103.65.
• Final judgment: watch only. This is the classic "catching a falling knife" setup — good fundamentals are not a reason to buy today, because what is killing it is not itself. Wait for sector sentiment to stabilize.

⑩ WBD (Warner Bros. Discovery) — watch only

• Related news: 08-06 07:00 ET pre-market. Q2 revenue $8.72B (−11%), below consensus ($9.21B and $9.29B coexist across two sources, so "about $500–570 million below consensus"); diluted EPS $0.06, better than the loss the market expected (sources give −$0.09/−$0.10/−$0.12/−$0.13); advertising revenue $1.7 billion (−22%; another source says −27%, and this report takes −22% and flags the conflict).
• ⚠️ But this is a low-quality beat: the swing to profit came mainly from operating expenses −23% YoY (because the NBA broadcast costs are gone), while net income for the quarter collapsed 91% YoY. Main causes: the loss of NBA broadcast rights (not renewed after nearly 40 years, with NBCUniversal taking over) + Supergirl grossing only $126 million worldwide (DC's worst since Catwoman in 2004).
• M&A status: the Paramount Skydance acquisition has been approved by the UK CMA, but the U.S. antitrust suit is set for trial in March 2027; Paramount CEO David Ellison says he is "absolutely willing to settle."
• Pre-market and technicals: +0.62% @ $26.13 (verified at 08:17 ET) — a big revenue miss and it is not falling. One possible reading is that the share price is already anchored by merger-arbitrage money, with fundamental price discovery ceding to the deal process; but this report has no volume or arbitrage-spread data to support that causal chain, so it is offered as a hypothesis, not a conclusion.
• Final judgment: watch only. This is already an event-driven (M&A) name rather than a fundamentals name; its risk point is the March 2027 litigation, not this earnings report.

⑪ AMD (超微半导体 / AMD) — watch only | of the three circulating negatives, two do not hold up against the primary 8-K

It is placed here because yesterday's AMD −7.04% was the starting point of this "beat yet fell" narrative, and the market's attribution for it used the wrong numbers on a broad scale.

• Primary data (SEC 8-K Ex-99.1, 2026-08-04): Q2 revenue $11.536 billion (+50.1% YoY); data center $6.718 billion (+107.3%, 58.2% of the total), with segment operating margin rising from 27.7% in Q1 to 31.3%; non-GAAP EPS $1.66; non-GAAP gross margin 56.24%.
• ❌ Does not hold up, #1: "Q3 guidance missed." Q3 guidance is $13 billion ± $300 million, while sell-side consensus was $12.501 billion — the guidance midpoint is about 4.0% higher. Together with Q2 revenue and Q2 EPS, that is a triple beat.
• ❌ Does not hold up, #2: "gross margin of 54% missed the 56% consensus." This compares GAAP 53.77% against a non-GAAP consensus of 56% — a convention mismatch. The true non-GAAP gross margin is 56.24%, in line with expectations. ⚠️ The draft of this report also cited this mistaken attribution, and it is corrected here.
• ✅ The one that does hold up, and where the real problem lies:
  1. Gross margin is locked down by guidance: Q3 non-GAAP gross-margin guidance is still "about 56%," flat to slightly down sequentially. And with the DC mix already at 58% and rising, the market had expected a rising AI mix to lift gross margin — the guidance kills that lever outright. The reason: Helios sells rack-level systems containing large amounts of bought-in memory/networking/rack components, so the product mix moves up while the gross-margin structure moves down.
  2. FCF has clearly deteriorated: capex/revenue jumped from 3.8% to 7.0% (+108% sequentially, nearly 3x YoY), so net income rose 66% sequentially while free cash flow fell 39% sequentially, with FCF margin going from 25.0% to 13.5%. When a fabless company shows this kind of jump in capital intensity, DCF cash-conversion assumptions need to be revised down.
  3. The drag hidden behind the DC halo: Gaming revenue −30.6% YoY, and the combined Client & Gaming margin fell from 21.2% to 15.2%.
• So why did it fall 7% on 8/5? The strongest explanation is positioning, not fundamentals: AMD rose +14.1% over the 5 sessions before earnings, of which +7.00% came on the earnings day itself (8/4); after falling 7.04% on 8/5 it closed at $482.05, just −0.5% relative to the 8/3 close of $484.64. In essence this was a round trip of front-running ahead of the news and giving it back afterward, not a fundamental repricing.
• Valuation is what should really worry you: P/E (TTM) 123.0x, P/S 19.0x, EV/EBITDA 81.8x, P/FCF 93.5x, and an FCF yield of only 1.07%. The headline P/B of 11.7x does not look high, but goodwill + intangibles of about $41.5 billion ≈ 62% of equity (a legacy of the Xilinx acquisition), leaving tangible book value of only about $25.7 billion → P/tangible book ≈ 30x. Using P/B on AMD badly understates its valuation level.
• The sharpest comparison: AMD's P/S of 19.0x is now essentially level with NVDA's 20.9x, but ROE is 10.1% vs NVDA's 114.3% (an 11x gap) and FCF margin is 13.5% vs NVDA's 47.0%. The market is already pricing AMD close to NVDA on a per-dollar-of-revenue basis, without AMD having earned NVDA-level per-dollar-of-profit.
• Final judgment: watch only. Shorting lacks a fundamental basis — a triple beat + net cash of $9.885 billion + a debt/equity ratio of 4.8% means shorting lacks a fundamental basis. But it should not be bought either: the current price is not pricing delivered results, it is pricing the shape of things after Helios/MI400 ramps in 2027, and in today's "kill the high-multiple acceleration" environment that premise is the most fragile of all.

---

6. Bearish / Avoid List

---

7. Within-Theme Ranking

Theme 1: Memory / AI-memory chain ebbing (short/avoid direction)

Theme 2: Software SaaS budgets crowded out (short/avoid direction)

Theme 3: NVIDIA's exclusive ecosystem (long direction)

Theme 4: The peace trade (direction splits)

---

8. Open-Bell Confirmation Signals

Pre-market (before 09:30 ET)

• Watch pre-market volume in WDC and SNDK: if they fall on heavy volume, it is institutional repositioning rather than sentiment and the decline will carry into the session; if volume is light, a gap-fill is more likely.
• NVDA is today's decisive name: it is currently +1.14%. If NVDA turns negative after the open, memory-chain panic has spread to the entire semiconductor complex, in which case Theme 3 (NVIDIA's exclusive ecosystem) is invalid for the day and every inference drawn from that theme should be treated as unestablished.
• The divergence between Dow futures and Nasdaq futures (+0.24% vs −0.44%) is today's core structure. If the divergence converges after the open (the Nasdaq catching up), memory is an isolated event; if it widens, the rotation trade is confirmed.

Intraday (first 30 minutes)

• DASH is the litmus test for "is there still a bid for expensive growth": it gapped +6.14% with a P/E of 108. Gap-and-go = the market is still willing to pay up for growth; gap-fill = every high-multiple name is in danger today.
• Sector ETF read-through: watch SMH (semiconductors) and IGV (software). If both fall, today is a "multiple compression day" rather than a "memory event day", and the response is entirely different.
• SPCX's opening volume: whether the unlock supply actually shows up can only be verified with volume, not guessed from price.

Options sentiment

• SNDK / WDC have post-earnings IV crush, so even getting the direction right, buying options can still lose money.
• TWLO's earnings tonight carry an options-implied move of ±14%, making it the largest single-name volatility source in software tonight.

Macro timestamps (must wait)

• 08:30 ET initial jobless claims + Q2 productivity / unit labor costs — this report's snapshot (08:10 ET) predates that data, and all pre-market pricing may be reset.
• 10:00 ET wholesale inventories, 10:30 ET EIA natural gas storage.
• ⚠️ Tomorrow, 8/7 at 08:30 ET, is the July nonfarm payrolls report. Today is very likely a "wait-and-see day ahead of the data," which amplifies the impact of a single-sector event (memory) on the indices.

Risk signals

• Gold at $4,328.50 (+$23.30) is still rising under the "peace trade" narrative, which is a discordant signal — either the market does not believe the agreement will be signed, or there is other hedging demand the narrative does not cover.
• Dimon (JPMorgan CEO) warning that market leverage is unusually high — in an environment with single names gapping −14%/−23% on two consecutive days, that remark carries more weight than usual.
• Lone-gainer risk: if NVDA rises while SMH falls, do not read NVDA's strength as a sector signal.

---

9. Final Conclusions

① The 5 names most worth watching today

② Today's 3 strongest themes

③ What to avoid today

1. Every AI hardware name with "high expectations + a big run" — ⚠️ today's rule is stricter than "a beat is not enough": WDC delivered beat + raise + a dividend increase and fell 14.70% anyway. What is being priced is the second derivative (the slope of growth and gross-margin expansion), not this quarter's results, and not the absolute direction of guidance.
2. Software SaaS as a whole — before NET/DDOG give guidance tonight, there is no safe spot in this sector.
3. Energy names that have "only commodity exposure and no company catalyst" — the better the results, the more awkward, with the commodity trend running directly opposite to the earnings trend (OXY is the archetype). But do not paint with one brush: COP posted a big EPS beat of $3.23 this morning and is +1.24% pre-market, showing that energy names with independent company catalysts (divestitures, CEO transition, M&A) can outweigh the commodity headwind.
4. On the "short the memory chain" narrative, mind how crowded it is — SNDK P/E 18 + a $14 billion buyback, WDC P/E 21, and post-cut sell-side price targets still above current prices; short crowding is high, and bounce risk should be built into scenarios.
5. Before the 08:30 ET data release, every pre-market price in this report may be reset — all snapshots predate the initial claims and productivity data.

④ Final one-line judgment

What can be observed today is this: an earnings report that was "good on results, guidance and dividend" (WDC) earned −14.70%, while one that was "a big beat on results but a miss on guidance" (SNDK) earned −10.26% — the better report fell more. That tells you the market is not pricing this quarter, it is pricing the valuation assumption for the entire memory chain; what is being sold is the second derivative, not the results. Meanwhile NVDA is +1.14%.

⚠️ But please confine that sentence strictly to "a single-day cross-sectional phenomenon," and do not escalate it into a structural conclusion like "AI pricing power is shifting from cyclical goods to monopoly goods," for three reasons: ① NVDA's catalyst today comes from the 8/4 SpaceX call, which is not the same driver as today's memory selloff, and the two cannot be treated as causes of each other; ② the sample is biased — the S&P 500 also made an intraday record high on 8/5, and telling only the "Dow record, Nasdaq down" story narrates a mixed day as one-way rotation; ③ the phenomenon carries its own falsification condition (see §8): if NVDA turns negative after the open, the "monopoly goods are favored" reading fails that same day.

The only thing that can be stated with certainty: in today's environment, "good earnings" by itself is not a reason to buy any memory stock. Add that the 08:30 ET data has not been released, nonfarm payrolls come tomorrow, and the SPCX supply gate opens today, and informationally today is a "wait for confirmation" day: three key guidance conventions are unsettled, the 08:30 macro data is unreleased, nonfarm payrolls are tomorrow, and the SPCX supply gate has just opened.

---

Data-Fetch Failures and Self-Audit Log (not for clients)

1. yfinance was unusable throughout — a pure IP-level ban. Both fundamentals-verification tracks this time (NVDA/AMD, SNDK/WDC/SPCX) had .info and .fast_info all throw YFRateLimitError: Too Many Requests; further testing showed the /v8/finance/chart endpoints on both query1 and query2 returning HTTP 429, with curl_cffi spoofing ineffective — judged to be a pure IP-level CDN ban, not a leftover process.

• Working alternative paths (tested this run): the CNBC quote API (200, all fields including pre-market), SEC EDGAR 8-K Ex-99.1 (primary earnings, the most reliable), stockanalysis.com (quarterly statements).
• One exception: bare curl against the query1 chart endpoint returned 200 for one of the agent tracks, retrieving 1048 daily bars — indicating the ban is intermittent/endpoint-level, not site-wide.
• Partially retrieved but stale: quarterly_income_stmt / cashflow / balance_sheet are retrievable after backoff retries, but AMD's latest quarter only goes to 2026-03-31, with Q2 not yet ingested by Yahoo — using it directly would give data lagging by a full quarter, a "silent lag" class of failure.

2. Three substantive errors in the draft, all found by cross-verification from sub-agents:

3. Quote drift: HONA went from −13.1% to −17.0% within 20 minutes (and broke its 52-week low), and AMD from −0.84% to −1.38%. A "single unified snapshot convention for the whole report" cannot contain pre-market drift, so it has been changed to per-item timestamps.

4. Process judgment: all three sub-agents completed and all produced substantive corrections — fundamentals-analyst overturned two core facts, and risk-auditor caught the missing internal-only section (this one) and the direction of the SPCX share count. Skipping any one of them before finalizing would have shipped a report with hard errors.

---

⚠️ Risk disclaimer: this list is pre-market information gathering and observation only, and does not constitute investment advice. U.S. equities carry high volatility and pre-market gap risk, post-earnings IV crush and guidance reversals occur, and automatically generated content may have information lags or factual errors. Company disclosures and SEC filings govern; this must not be used directly as a basis for trading.