US Pre-Market Brief | 2026-08-27 (ET) Thursday

Coverage window: 2026-08-26 16:00 ET regular-session close → 2026-08-27 08:45 ET (includes the 08:30 economic data and the 08:35–08:43 supplemental scan) Pre-market quote timestamps: main tables 08:00–08:08 ET; a few names carry supplemental 08:35–08:43 ET readings (flagged in the body). Source: stockanalysis.com, cross-verified against the CNBC quote API (10/10 agreement). Baseline is the 2026-08-26 close. Basis of financial figures: all key data for NVDA, CRM, OKTA, CRWD and SMCI are taken from the original SEC 8-K / EX-99.1 filings, not from media paraphrase. 📌 08:35 ET update: today's 08:30 ET initial jobless claims are out — actual 203,000, below the 208,000 expected (prior 207,000); continuing claims 1.778 million, below the prior 1.796 million (this brief did not obtain its consensus figure). Claims stronger than expected → a tighter labor market → reinforces the pricing of a hike this year. The body below has been updated with the actual figures, replacing the earlier "pending release."

---

0. One-Line Summary

1. The strongest catalyst is NVIDIA, but what moved the stock was not the press release — it was the call. NVDA FY27Q2 revenue $96.221 billion (+106%), data center $89 billion (+117%), Q3 guidance $108 billion ±2%. After the release, Bloomberg's headline read "forecast beats but the stock falls"; what flipped the stock was CFO Colette Kress on the 17:00 ET call saying "we expect fiscal 2028 revenue to grow approximately 70%" — versus sell-side expectations for FY2028 of roughly 45% before the print. That sentence is not in the press release, not in the 8-K, and is not formal guidance. Jensen Huang himself conceded on the call: "We have never made a forecast or given guidance more than a year out."
2. Within the top 20 of the pre-market movers list this brief covered, software is stronger than semis (names ranked 21 and below were not covered because the data source's pagination failed, so stronger names outside the sample cannot be ruled out; this brief ran a supplemental scan with a second data source, adding STX/WDC/DLTR and others — see §7). OKTA +17.7%, CRM +10.5%, CRWD +9.5% (all 08:00–08:08 ET readings), while AVGO is up just +1.19% and AMD +1.16%. NOW (+2.6%) and PANW (+5.1%), neither of which reported, rose in sympathy, supporting "sector repricing" rather than isolated events — but their pre-market notional values are only about $24.9 million and $16.6 million respectively, one-sixth to one-tenth of CRM's ($156 million), so the strength of that evidence must be discounted.
3. But the three software prints differ enormously in quality; they must not be bought as a bundle. After a line-by-line check against the original SEC 8-Ks: CRWD is a genuine beat and an organic one; OKTA is a genuine beat, but roughly half the EPS growth comes from the non-GAAP tax rate changing from 26% to 21%; CRM's "80% beat" was manufactured by a $2.613 billion mark-up on its Anthropic stake — the company's non-GAAP definition does not exclude that gain, and stripping it out leaves a true beat of only +3.1%, with GAAP operating income flat year over year. See §5.
4. The deepest thread today is that profit is migrating from the compute vendor to the memory vendors — and both ends have first-hand numbers. Upstream: NVDA states verbatim in its 8-K that its $279 billion of supply commitments are "primarily related to memory purchases," and it cut gross margin guidance from 75.0% to 74.0%, with management saying it troughs at about 71–72% in FY27Q4, attributing it to memory costs. Downstream: HP disclosed that RAM now accounts for 35% of PC bill-of-materials cost, with PC revenue +18% while unit shipments fell −16%, and segment margin going 5.2% → 4.6%. One is the most profitable chip company on earth giving margin away; the other is the PC leader being squeezed — and the attribution points to the same place.
5. Pre-market state: S&P 500 and Nasdaq futures +0.5%, Dow futures slightly lower; 10Y Treasury 4.66% (+1bp, a second straight rise); 2Y 4.23%; VIX about 15.0; Brent crude about $86 (a third straight decline).
6. The 08:30 ET data is today's first real variable, and it leans hawkish. Initial claims 203,000, below the 208,000 expected (stronger than expected); continuing claims 1.778 million, below the prior 1.796 million (consensus not obtained). Against a backdrop of core PCE just running hot, this further supports the pricing of "one to two more hikes this year" — note that the reason the 10Y rose yesterday was precisely "stronger-than-expected economic data reinforcing expectations of a hike before year-end"; today's data is a second brick in the same wall. That is a direct headwind for the high-multiple software names leading today's gains (OKTA forward PE 34, CRWD 150). The biggest calendar risk is still tomorrow: Chair Warsh's debut Jackson Hole keynote, 8/28 10:00 ET.

---

0.1 The Macro Premises That Must Be Established First (Otherwise the Whole Brief Will Be Misread)

Several commonsense assumptions no longer hold in 2026; let's align first:

What this actually means for today: after yesterday's hot core PCE, the 10Y has risen two sessions running to 4.66% — and that happened against a large positive NVDA surprise; the bond market did not join the equity market's excitement. Warsh's style is explicitly not to use traditional forward guidance, so tomorrow at 10:00 ET is a variance event, not a directional event: it could confirm the hiking path, or it could say nothing at all. Therefore there is an event tomorrow morning at 10:00 ET that cannot be priced in advance, and overnight exposure faces gap risk.

---

0.2 Reconciliation: The Verification Points Logged in Yesterday's (8/26) Pre-Market List

Yesterday's report turned the case for two names into specific, falsifiable numbers. Settle the account before discussing today.

What matters most from this reconciliation: the stock-level verification points logged yesterday were all accurate — CRWD's net new ARR, and CRM's $3.27 consensus and cRPO basis, happen to be exactly the numbers that decided the truth today. The $3.27 in particular became today's tool for seeing through the "80% beat" (see §5④). Writing the verification point first and deriving the conclusion from it is more reliable than writing the conclusion first.

---

1. News Overview

---

2. Strongest Themes, Descending

---

3. Overall Single-Stock Strength Table

Readings taken 08:00–08:08 ET, baseline the 8/26 close. "Pre-market notional" = pre-market shares traded × pre-market price, used to judge whether the quote is executable — a more reliable liquidity criterion than share count.

3.1 Bullish Side

3.2 Bearish Side

---

4. Single-Stock Scoring Model (100 Points)

Deduction rationale (each tied to a specific number)

• NVDA −12: Q3 gross margin guidance 74.0% (down 100bp), with management saying Q4 troughs at 71–72%; this quarter OCF / net income is only 40% (86% last quarter), FCF −56% sequentially; DSO rose from 45 days to 60 days, while last quarter's 8-K explicitly signaled "we expect DSO to return to a more normal level next quarter" — that is a management expectation that has now been falsified; net cash was halved from about $41.2 billion to $23.2 billion, with $24.9 billion of net debt issued this quarter.
• CRWD −11: Q2 ARR beat guidance by about $45 million, yet full-year ARR guidance was raised by only $64 million, implying a true second-half raise of only about $19 million; full-year non-GAAP EPS guidance was raised only about 1.6%; forward PE 150.7, PS 35.7.
• DG −9: already gapped 6.7%~10.2% in the pre-market, with the pre-market price at or through the sell-side mean target of $132.31; the logic depends heavily on inflation continuing to squeeze consumers.
• OKTA −13: Q3 non-GAAP operating margin guidance 24–25%, below this quarter's 28%; (note: the original deduction included an item for "Q3 cRPO guidance pulling back"; QC pointed out that last quarter's original guidance was also +11% while the company delivered +14% — the same conservative convention — so that deduction has been withdrawn and the total raised from 64 to 67); Q3 non-GAAP operating margin guidance 24–25%, below this quarter's 28%; the non-GAAP tax rate was changed from 26% to 21%, and restated at the old tax rate, EPS growth is +7.7% y/y rather than the reported +15.4%; the pre-market $158.21 is already above the 52-week high of $157.00.
• MU −12: PS 11.7 and PB 10.5 both break the top of the two-year range, while the "cheap" PE of 21.2 comes from peak-cycle EPS — this is the textbook low-PE trap formation, not a margin of safety; the financial data is about 3 months stale and the next print is 9/30.
• CRM −14: excluding the $2.613 billion investment gain, the true beat is only +3.1%; GAAP operating income is flat year over year ($2.331 billion vs. $2.332 billion); organic revenue growth decelerated from +8.75% to +6.38%; within the full-year revenue guidance raise, only $100 million (+0.22%) is organic, with the other $200 million from acquisitions that have not yet closed; net debt rose from $4.9 billion to $27.9 billion, and quarterly interest expense from $67 million to $473 million.

---

5. Detailed Analysis of Top Names

1英伟达NVDA / NVIDIAPriority deep-dive | Score 78 | Pre-market +6.07% ($209.66 → $222.38)

Related news (all taken from the original SEC 8-K / EX-99.1)

• FY27Q2 (ended 7/26) revenue $96.221 billion (+17.9% q/q, +105.9% y/y); data center $89 billion (+18% q/q, +117% y/y), 92.5% of total revenue.
• GAAP and non-GAAP gross margins were both 75.0%; GAAP diluted EPS $2.46, non-GAAP $2.22.
• Q3 guidance: revenue $108 billion ±2%; gross margin 74.0% ±50bp; GAAP/non-GAAP operating expenses about $9.2 billion / $9.0 billion. From the 8-K: "NVIDIA's outlook does not assume any data center compute revenue from China."
• Jensen Huang, verbatim: "AI has reached an inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue."

How much of a beat the guidance really is — and one trap that must be avoided

• Pre-print consensus for Q3 revenue was about $103.8 billion ~ $104.2 billion. The $108 billion midpoint is about +3.7%~4.0% above it.
• The sturdier formulation: the low end of the guidance range, $105.8 billion (108×0.98), is still above the pre-print consensus high end of $104.2 billion — the entire guidance range sits above consensus.
• ⚠️ The trap: the Q3 consensus pulled from data terminals today (8/27) is already $108.67 billion (32 analysts) — that is a post-print, upward-revised number. Judging the guidance against it produces the exact opposite conclusion, "guidance below expectations." Beat/miss judgments must use the pre-release consensus.
• ⚠️ But this brief must apply the same discipline to itself (added after QC): the $103.8–104.2 billion above comes from same-day media paraphrase (Reuters / Investing.com basis), and this brief did not obtain the original snapshot's provider, analyst count or precise timestamp. The single most important sentence in this brief — "the entire guidance range sits above consensus" — is built on that number, so please discount it as "media paraphrase, original snapshot not obtained." Likewise, the other consensus figures cited in this brief (DG $2.01, BBY $1.38, HPQ $0.69, MRVL $2.71 billion/$0.93, BURL $11.23, P $0.58, etc.) are all media-relayed values without a stated provider or snapshot time, and are uniformly flagged to be verified.

Precisely characterizing the "FY2028 growth of about 70%"

• Across the full text of the 8-K EX-99.1 (113,510 characters), the Outlook section contains only three items: Q3 revenue, Q3 gross margin, Q3 operating expenses, plus the FY27 full-year tax rate. There is no FY2028 revenue outlook anywhere in the 8-K.
• The statement comes from CFO Kress on the 8/26 call: "We expect to grow revenue by approximately 70% in fiscal 2028." She simultaneously qualified it as a "supply-constrained outlook." Jensen Huang conceded on the spot: "We have never made a forecast or given guidance more than a year out."
• Characterization: this is management commentary, not formal financial guidance in the Reg G sense. It is not in the 8-K, has no range, no ±%, and is not subject to guidance update obligations. Placing it alongside the $108 billion as "two pieces of guidance" is wrong.

[Most important section] Gross margin 75.0% → 74.0% → about 71–72%: profit is leaving NVDA

• Fact (8-K): Q3 gross margin guidance is 74.0%, down 100bp from this quarter.
• Fact (call): management said gross margin will trough at about 71–72% in FY27Q4, attributed to rising memory costs.
• Fact (8-K text): supply commitments rose from $119 billion to $279 billion, "primarily related to memory purchases."
• Magnitude estimate (derived, not company-disclosed; ⚠️ tightened after QC): 75.0% → 71.5% is 3.5pp. The base used is this brief's assumption of about $123 billion of FY27Q4 revenue — a number that comes from sell-side consensus, is not in the 8-K, and is not inside the Q3 guidance ($108 billion); it has not been confirmed by the company. On that static basis it works out to about $4.3 billion per quarter. This brief does not annualize it: 71–72% is the trough level the company cited, not a full-year run rate, and multiplying by 4 would misspecify both the revenue and the gross margin assumptions — exactly the "extrapolating from a single quarter" practice this brief criticizes SMCI for in §6, which we cannot then commit ourselves.
• This is the chain that turns "thematic association" into "verifiable transmission." It has an amount, a source and a timeline — and it comes from NVDA itself.

The accounting meaning of the $279 billion of supply commitments (it cannot be treated as demand confirmation) The 8-K gives a full year-by-year schedule (in USD billions):

	Rest of FY27	FY28	FY29	FY30	FY31	FY32+	Total
Supply and capacity	92	87	88	6	5	1	279
Cloud service agreements	3	8	7	6	4	1	29
Data center leases not yet commenced	—	1	1	2	1	20	25
Equity investments	18	3	2	2	—	—	25
Capital expenditures	7	1	—	—	—	—	8
Grand total	120	100	98	16	10	22	366

Table scrolls horizontally

1. This is an off-balance-sheet purchase obligation disclosure; it does not go on the balance sheet, and inventory and payables are recognized only on delivery.
2. The direction is cash outflow — this is money NVDA owes suppliers, not money customers owe NVDA.
3. The evidence of demand is backlog, RPO and deferred revenue. Not one dollar of the $279 billion belongs to those three categories, so it cannot be treated as demand confirmation.
4. There is precedent for the reverse risk, written right into the footnotes of the same 8-K: NVDA took a $4.5 billion charge on H20 in FY26Q1.
5. A large part of the increase in the amount is price. The company itself says it is "primarily memory," and memory is in a historic price surge. A 2.34x increase in dollars ≠ a 2.34x increase in units.

Two negatives buried by the market narrative

(a) Cash conversion collapsed in a single quarter

	GAAP net income	Operating cash flow	OCF / net income	Free cash flow
Q2 FY27	$59.688 billion	$24.077 billion	40%	$21.341 billion
Q1 FY27	$58.321 billion	$50.344 billion	86%	$48.554 billion

Table scrolls horizontally

Revenue +17.9% q/q, free cash flow −56% q/q. The drags: accounts receivable −$22.346 billion, inventory −$5.784 billion, prepaid and other −$5.497 billion, for a net working-capital outflow of about −$33.6 billion; there was also $7.771 billion of non-cash equity investment gains (which is also why GAAP EPS of $2.46 exceeds non-GAAP's $2.22).

(b) A management expectation that has been falsified

From last quarter's 8-K: "We expect DSO to return to a more normal level next quarter." Actual:

Quarter	Revenue	Accounts receivable	DSO
2026-04-26	$81.61 billion	$40.71 billion	45 days
2026-07-26	$96.22 billion	$63.06 billion	60 days

Table scrolls horizontally

It did not come down; it hit a new range high. Net cash was halved from about $41.2 billion to $23.2 billion, with $24.9 billion of net debt issued this quarter; at the same time the company holds $93.94 billion of equity investment exposure — precisely the accounting vehicle behind the "circular financing" dispute.

Valuation: a counterintuitive fact On PE / PS / PB / EV-EBITDA / P-FCF — all five metrics — NVDA sits at the very bottom of its own two-year range, because earnings have run faster than the stock.

⚠️ But NVDA must be measured with the same ruler as CRM (added after QC): this quarter's GAAP net income contains $7.771 billion of non-cash equity investment gains (about $0.32 per share), and the GAAP PE of 26.5x above does not exclude that gain. This brief spends an entire section criticizing CRM for leaving investment gains inside non-GAAP, so it cannot make an exception for NVDA — stripping it out puts the valuation percentile somewhat above "the very bottom." The non-GAAP PE of 30.0x is the cleaner basis. The close corresponds to a GAAP PE of 26.5x and a non-GAAP PE of 30.0x; fiscal-year forward PE is 22.6x. At the same moment, TSM is at the top of its own range on all three of PE/PS/PB.

⚠️ A cross-company comparison trap: starting in FY27Q1 NVDA changed its non-GAAP definition to no longer add back stock-based compensation, and retroactively restated comparable periods (FY26Q4 restated from $1.62 to $1.59). That makes NVDA's non-GAAP more conservative than peers such as AVGO/MRVL that still add back SBC, so comparing non-GAAP directly across companies will distort.

Pre-market and technicals: $222.38 (08:05 ET), previous close $209.66, +6.07%, pre-market notional about $2.7 billion — the deepest of the day, so the quote is the most executable. 52-week range $164.07–$236.54, the pre-market price is 6.0% below the 52-week high and sits at the 80.5th percentile of the range. In yesterday's regular session NVDA closed −1.59%, meaning the market went into the print cautious.

⚠️ Also note the sequential deceleration: +22.0% → +19.5% → +19.8% → +17.9% → +12.2% (Q3 guidance). A meaningful part of the +106% y/y is a base effect.

Final verdict: priority deep-dive, but do not chase the pre-market gap. This is the only name today where earnings, guidance and industry logic are all hard and valuation sits at its own historical lows. But it also carries three substantive blemishes: gross margin falling to 71–72%, cash conversion down to 40%, and a falsified DSO expectation. The thing to watch is the first 30 minutes: gap-and-go or gap-fill. If $108 billion of guidance plus a 70% outlook can only produce an open-high-close-low, it means the market's doubts about AI capex now outweigh its acceptance of the results — and that matters more than any single quarter's number.

---

② CRWD — CrowdStrike | Priority deep-dive | Score 74 | Pre-market +9.5%

Core: yesterday's logged verification point was cleanly broken, and confirmed against the original SEC 8-K

Full net new ARR series (from each quarter's original 8-K)

The acceleration is real, and it is organic. The cash flow statement shows that the six-month figure for "business acquisitions, net of cash acquired" this quarter is exactly the same as the Q1 standalone figure ($881 million) → that acquisition closed in FY27Q1, and Q2 standalone acquisition cash outflow was 0. So the $332.8 million is purely organic incremental. And it exceeded the seasonally strongest FY26Q4 ($330.7 million) — breaking a Q4 record with a Q2 carries more weight.

But two cooling factors must be placed alongside it (this is where this brief diverges most from the market narrative)

1. Low base effect: the denominator of that +50.5%, FY26Q2 ($221.1 million), is exactly the period when the "customer commitment packages" following the 2024-07-19 incident were being settled en masse and net new ARR was at its weakest. The absolute-value series is the fairer read: $330.7 million → $255.8 million → $332.8 million is a recovery to record levels, not a step change.
2. The full-year guidance raise is far smaller than this quarter's beat: Q2 ending ARR beat the guidance midpoint by about $45 million, while the full-year ARR guidance midpoint was raised by only $64 million. Netting out what has already been realized this quarter, the implied true second-half raise is only about $19 million. The company talks about a "record Q3 pipeline," but has not put it into the numbers. Full-year non-GAAP EPS guidance was also raised only about 1.6%.

Fundamentals: TTM revenue $5.396 billion (+24.2%), accelerating for 4 straight quarters. But the GAAP operating margin is −2.3%, and GAAP PE is 4,281 (no information content). The key cash-quality metric: FCF margin 29.8%, but stock-based compensation is 22.8% of revenue, so FCF less stock comp leaves only +7.0% — far below CRM's 26.1% and PANW's 20.0%.

Valuation and position (⚠️ this passage is more conservative than the market narrative): forward PE 150.7x, PS 35.7x, EV/ARR about 32.3x. 52-week range $85.68–$227.50.

• The pre-market $207.23 is only −8.9% from the 52-week high (note: the −16.8% commonly cited in media is on the $189.18 close basis, which no longer applies after the gap)
• The pre-market price sits at the 85.7th percentile of the 52-week range
• The sell-side mean target is $219.35, only 5.85% above the pre-market price — i.e. the open has already consumed about 60% of the sell-side consensus headroom

Note: CRWD completed a 4-for-1 stock split on 2026-07-02, so historical EPS and share-count bases need care.

Final verdict: priority deep-dive. This is the cleanest set of fundamentals today — the criterion was registered in advance, it was clearly exceeded after the fact, and organic character was confirmed via the cash flow statement. But all the risk is in the price: a forward PE of 150.7x means the market already treats "continued acceleration" as a settled fact, while the company itself added only about $19 million to full-year guidance; at the same time the pre-market price is only 8.9% from the 52-week high and 5.85% from the sell-side mean target. "Cleanest fundamentals" and "chaseable" are two different things — that gap is the thing most worth watching on CRWD today.

---

③ DG — Dollar General | Watch closely | Score 74 | Pre-market +6.7%~+10.2%

Related news: 8-K filed pre-market on 8/27. Q2 (ended 7/31) net sales $11.3 billion (+5.2%), comps +3.5%, of which traffic +2.0% and ticket +1.5%; EPS $2.48 (consensus $2.01, a 23% beat).

Catalyst logic: positive traffic is what makes it a real signal. In an environment of 3.3% core PCE and about $4 gasoline, a retailer turning comps positive can easily be nothing but price increases. DG's +2.0% traffic shows a genuine increase in customer count — consumers are migrating toward discount channels. Together with today's WEN collapse and BBY being sold at highs, that forms a self-consistent K-shaped consumption picture: money is moving down, not disappearing.

Fundamentals: market cap $27.08 billion, PE 17.4, forward PE 16.2, dividend yield 1.92%.

Pre-market and technicals: $135.30, previous close $122.78, +10.20% (08:00 ET, pre-market volume 285,690 shares, notional about $38.7 million). 52-week range $95.11–$158.23, the pre-market price sits at the 64th percentile of the range, still 14.4% below the 52-week high — this is its most important advantage over OKTA/CRWD: it gapped too, but not into all-time-high territory.

⚠️ Data basis disclosure (DG has the widest quote disagreement in this brief; please note): three readings, all against the same previous close of $122.78 —

• stockanalysis movers list (08:00 ET): $135.30 / +10.20% ✅ internally consistent
• investing.com (08:43 ET): $131.00 / +6.69% ✅ internally consistent
• stockanalysis single-stock page (08:04 ET): $132.00 / labeled +9.22% ❌ inconsistent ($132.00 implies +7.51%), discarded

The two internally consistent readings differ by $4.30 (3.3%) in price — a cross-source pre-market quote divergence (different aggregators sample different venues). This brief therefore gives a range rather than a point: pre-market $131.00–$135.30, i.e. +6.7%~+10.2%.

A positioning issue that must be spelled out: the sell-side mean target of $132.31 falls exactly inside that quote range ($131.00 below it, $135.30 above it). In other words, whether DG has already exhausted the sell-side consensus headroom depends on which quote you use — which itself shows it is already pinned to the consensus level. It is still 14.5%~17.2% below the 52-week high, a better position than OKTA/CRWD, but near-term consensus headroom is essentially used up.

⚠️ A cross-sectional reminder: the pre-market prices of today's top four longs (NVDA, CRWD, OKTA, DG) have all consumed more than 60% of the sell-side consensus headroom (CRWD about 60%, OKTA about 73%, DG about 100%). That is a direct consequence of "everything gapped," not an issue specific to any one name — today's core difficulty is not picking the right name, it is picking the right price.

Final verdict: watch closely. The logic is hard (positive traffic), the valuation is not expensive (forward PE 16.2), and there is still room to the 52-week high. The deductions are that it has already gapped 10.2% and the pre-market price is above the sell-side mean target, and that the entire logic rests on "inflation continuing to squeeze consumers" — if Warsh sends a hawkish signal tomorrow, that logic is actually reinforced. DG is one of the few long ideas today that runs in the same direction as the macro risk.

---

4Salesforce | Watch onlyCRMScore 58 | Pre-market +10.5% · downgraded from "watch closely"

This is the name most in need of taking apart today. The market is paying 10% for a beat that does not exist.

The headlines everywhere read "non-GAAP EPS $5.90 vs. consensus $3.27, a beat of 80.38%." That "beat" does not hold up. Everything below is from the original SEC 8-K / EX-99.1:

Item	FY27Q2	Year-ago quarter
Revenue	$11.345 billion (+10.83%)	$10.236 billion
GAAP operating income	$2.331 billion	$2.332 billion (zero growth)
GAAP operating margin	20.5%	22.8%
Net strategic investment gains	$2.613 billion	$6 million
Interest expense	−$473 million	−$67 million
GAAP diluted EPS	$4.29	$1.96
Non-GAAP diluted EPS	$5.90	$2.91
Non-GAAP operating margin	34.1%	34.3% (−20bp y/y)

Table scrolls horizontally

The key: Salesforce's non-GAAP definition does not exclude strategic investment gains and losses. The 8-K reconciliation has only four add-backs: amortization of intangibles $522 million, stock-based compensation $904 million, restructuring and acquisition costs $115 million, and a tax adjustment of −$223 million. The entire $2.613 billion investment gain stays inside non-GAAP net income. The company quantifies it perfectly clearly in an income statement footnote: "Strategic investment gains impacted GAAP diluted EPS by $2.43 and non-GAAP diluted EPS by $2.53."

The control group is decisive: ServiceNow excluded $273 million of strategic investment gains from non-GAAP in the very same quarter (NOW's 8-K reconciliation lists the line explicitly). Same industry, same quarter, opposite treatment.

The real scorecard, restated

	As reported	Ex-investment gains	Consensus / company guidance	True beat
Non-GAAP EPS	$5.90	$3.37	$3.27 (= the top of the company's own May guidance)	+3.1%
GAAP EPS	$4.29	$1.86	Guidance $1.74–1.76	+5.7%
Revenue	$11.345 billion	—	$11.32–11.33 billion	+0.13%

Table scrolls horizontally

Three independent internal-consistency tests, all pointing to the same conclusion

1. Q3 guidance is non-GAAP EPS $3.42–3.44. If Q2 truly earned $5.90 of operating EPS, Q3 would imply −42% sequentially. Whereas the restated Q2 of $3.37 → Q3 guidance of $3.43 connects perfectly.
2. FY27 non-GAAP EPS guidance was raised from $14.06–14.12 to $16.67–16.71, a midpoint raise of $2.60, of which Q2's investment gain alone contributes $2.53. The purely operational raise is just $0.07, i.e. +0.5%.
3. FY28 consensus adjusted EPS is $15.72, 6.0% below FY27's $16.72. A software company showing "negative EPS growth next year" is usually an anomaly signal — and the reason is that this year contains a one-off gain.

A fact worth savoring: the largest single item of profit in this Salesforce print comes from a mark-up on its stake in Anthropic — and Anthropic is one of the protagonists of the very "AI will replace SaaS" narrative. Benioff declares "the SaaS-is-dead talk is nonsense" while simultaneously using his stake in an AI company to lift EPS by $2.53.

Three things that matter more, obscured by the $5.90

1. GAAP operating income is flat year over year ($2.331 billion vs. $2.332 billion) while revenue is +10.8% — operating leverage is negative.
2. Organic revenue growth is decelerating: Q2 excluding Informatica's $456 million contribution is +6.38%, while Q1 excluding $444 million was +8.75%. The company's own wording is that organic revenue is expected to "reaccelerate" in the second half — which is an admission that the first half was soft.
3. The $5.90 does not convert into cash at all: GAAP net income $3.526 billion vs. operating cash flow of only $1.269 billion (a 36% conversion rate), because investment gains are non-cash and get backed out of the cash flow statement as-is.

The true composition of the full-year revenue guidance raise (from the 8-K): a raise of $200 million (+0.43%), which the company itself breaks down as organic +$100 million / pending acquisitions +$200 million / FX −$100 million. That means the organic raise is only +0.22%, and the $200 million piece is conditional on the Contentful and Fin acquisitions closing on schedule.

So where is the part of CRM worth watching?

• cRPO is the only clean positive signal: $33.5 billion, +14%, with constant-currency growth accelerating from 13% to 14%, a sequence of 10% → 13% → 14%, rising three quarters running; Q3 guidance is about +14% and explicitly excludes the two acquisitions.
• The valuation starting point genuinely is low: the close corresponds to a TTM PS of 3.83x, the lowest of the nine names in the group (the second lowest, OKTA, is 7.60x). Excluding investment gains, forward PE is about 15.0x (on the close) / 16.6x (on the pre-market price).
• ⚠️ But cRPO has an unverified hole too: the company disclosed Informatica's contribution to revenue ($456 million) but did not disclose its contribution to cRPO. If Informatica brought in roughly $1.3–1.8 billion of cRPO, organic cRPO growth would drop to +9%~10% — and the only clean argument supporting today's move would essentially vanish.

A balance sheet change that must be flagged: net debt rose from $4.9 billion to $27.9 billion (borrowing for a $2.5 billion accelerated buyback program), and quarterly interest expense from $67 million to $473 million; diluted share count fell from 962 million to 821 million (−14.7%). Part of the EPS improvement is financial engineering, not operational improvement. Also, goodwill of $59.25 billion is greater than shareholders' equity of $38.38 billion.

Final verdict: watch only (downgraded from "watch closely"). The cheapness is real and the cRPO acceleration is real, but a large chunk of today's +10.5% is paying for an "80% beat" manufactured by a one-off investment gain. A true beat of +3.1%, an organic revenue guidance raise of +0.22%, and flat GAAP operating income — those three numbers cannot support 10.5%. If you want to track it, the only metric to watch is whether cRPO constant-currency growth can accelerate for a second consecutive quarter, and judgment must wait until the company discloses Informatica's contribution to cRPO.

---

5Okta | Watch closelyOKTAScore 67 | Pre-market +17.7% · the most dangerous position

Related news (SEC 8-K): FY27Q2 revenue $805 million (+11%), subscription revenue $793 million (+12%); non-GAAP diluted EPS $1.05 (the top of the company's May guidance was $0.97); non-GAAP operating margin 28%; FCF $227 million (guidance $155–165 million). cRPO $2.585 billion, +14% (prior quarter +12%, original guidance +11%); RPO $4.858 billion, +17%. The company reported record bookings, with AI security products contributing 30% of bookings.

Full-year guidance	Prior	New	Raise
Revenue	$3.185–3.205 billion	$3.216–3.226 billion	+0.81%
Free cash flow	$855–885 million	$910–930 million	+5.7% (midpoint to midpoint; on the new high end vs. the old midpoint it would be +6.9% — this brief uses the midpoint basis)

Table scrolls horizontally

But three discounts must be applied (this is where this brief diverges from the market narrative)

1. ⚠️ The company guides Q3 cRPO to +11%~12%, below this quarter's actual +14% — but that is more likely a conservative guidance convention than confirmed deceleration (this is a post-QC correction). The key comparison: last quarter the company's original guidance for this quarter's cRPO was also +11%, and it delivered +14%. Same convention, same order of magnitude. This brief therefore downgrades the item from "negative" to "to be verified," and reduces the deduction accordingly. The real criterion is whether next quarter's actual can once again exceed the company's own guidance, not the guidance number itself.
2. Half the EPS growth comes from a change in the tax rate basis. The non-GAAP tax rate changed from 26% to 21% starting in FY27Q1 (8-K footnote: due to the One Big Beautiful Bill Act). Restated at the old tax rate, $1.05 is roughly $0.98, which against $0.91 a year ago is +7.7%, not the reported +15.4%. (Note: this does not affect the reality of this quarter's beat — guidance and actual used the same tax rate.)
3. Margins are declining sequentially: Q3 non-GAAP operating margin guidance of 24–25% and the full-year 26% are both below this quarter's actual 28%; Q3 non-GAAP EPS guidance of $0.92–0.94 is below this quarter's $1.05.

Fundamentals: market cap $23.36 billion, PE 81.7, forward PE 34.3, PS 7.60. The balance sheet is extremely clean: $0 of interest-bearing debt, net cash $2.299 billion (the $350 million convertible maturing in 2026 has been settled in cash). FCF less stock comp is +15.5%, healthy.

Pre-market and technicals: this is the position that requires the most caution today

• Pre-market $158.21, previous close $134.42, +17.7%, pre-market notional about $39.6 million.
• 52-week range $62.66–$157.00 — the pre-market price is already through the 52-week high.
• It opens in all-time-high territory, with no trapped supply or technical reference above.
• The sell-side mean target is $167.09, only 5.61% above the pre-market price — the open has already consumed about 73% of the sell-side target headroom.

Final verdict: watch closely. But the pre-market price sits in the least favorable position of any name today. This quarter is a genuine beat (EPS, revenue, cRPO and FCF all exceeded), and the balance sheet carries zero interest-bearing debt. The main reservations are that Q3 non-GAAP operating margin guidance of 24–25% is below this quarter's 28%, and that roughly half the EPS growth comes from the tax-rate basis change — while the stock has gapped 17.7% in one move to above the 52-week high and sits only 5.61% from the sell-side mean target. How to watch it: holding above $157 (the prior high) on volume counts as a breakout; a quick fade back below $150 is the textbook post-earnings gap fill.

---

⑥ HPQ — HP Inc | Avoid | Pre-market −11.83% | The most information-dense name today

On the surface: revenue $15.7 billion (expected $14.39 billion, a 9.1% beat), non-GAAP EPS $0.83 (expected $0.69, a 20% beat), and it raised full-year adjusted EPS guidance to $3.19–3.29 (from $2.90–3.10).

A beat + raise, and the stock is −11.83% pre-market. Take it apart and today's whole main line emerges.

Breakdown one: all the revenue growth is price; units are collapsing

Breakdown two: most of the guidance raise is not operational improvement

Stripping out the $0.19, full-year guidance is about $3.00–3.10 versus the original $2.90–3.10 — the high end did not move at all.

Breakdown three: the CFO said the quiet part out loud CFO Karen Parkhill: "Given the impact of commodity-driven price increases, we expect fourth-quarter revenue performance to be below seasonal levels." That is a substantive cut, it just never made it into the guidance table.

Breakdown four: the root cause is memory HP disclosed that RAM already accounts for 35% of its Personal Systems bill of materials (BOM). Price increases are not enough to cover memory inflation, there is further pressure in Q4, and management expects recovery only in FY2027.

Why this matters far beyond HPQ itself: it supplies hard evidence for the "memory supercycle" from the downstream demand end, issued by the injured party. And NVDA said the same thing in its 8-K today — $279 billion of commitments "primarily related to memory purchases," with gross margin falling to 71–72% because of memory costs. One is the PC leader, the other is the most profitable chip company on earth; two entirely independent statutory disclosures point at the same beneficiary. That kind of two-ended cross-confirmation is more reliable than either side's narrative alone.

Fundamentals: market cap $27.91 billion, PE 11.6, forward PE 10.5, dividend yield 3.93%. 52-week range $17.56–$32.19. Down 31% over the past year.

Final verdict: avoid. Cheap plus a high yield will tempt bottom-fishing, but the cause of the margin compression (memory cost) is outside HP's control, and the company itself says recovery does not come until FY2027. In an industry where the cost shock has not yet peaked, a low valuation is not support — it is a trap. A short watch is fine, but it has already gapped −11.83%, so chasing it short is an equally bad position.

---

⑦ BBY — Best Buy | Avoid | Pre-market −5.15% | The same story as HPQ

Also a beat + raise: comps +4.1% (the original outlook was only +1%), adjusted EPS $1.47 (expected $1.38), revenue $9.78 billion (expected $9.59 billion). Full-year guidance: revenue $41.2–42.1 billion → $42.3–42.8 billion; adjusted EPS $6.30–6.60 → $6.70–6.90; comps −1%~+1% → +1.9%~+3.0%. Management attributed it to "strength in computing."

Why it is −5.15% pre-market — three specific reasons

1. The phrase "computing strength" means something entirely different today. HP just disclosed PC unit shipments of −16% against revenue of +18%. If unit demand in the PC industry is falling, a large part of Best Buy's computing sales growth is also ASP inflation driven by memory prices, not consumers buying more computers. The quality of that +4.1% comp is therefore in question.
2. Gross margin includes a $34 million tariff refund (disclosed by the company), the same class of one-off item as HPQ's.
3. The entry price is already stretched. The 8/26 close was $87.44 (+2.52% on the day), while the mean target of 25 analysts is $83.40 — the close is already 4.8% above the sell-side mean target. The 52-week range is $55.10–$91.27, putting the close at the 89th percentile of the range.

Fundamentals: market cap $18.43 billion, PE 16.2, forward PE 13.0, dividend yield 4.39%.

Final verdict: avoid. This is not a bad print, it is a good print at the wrong price, and the good part contains a one-off item plus memory-price inflation. ⚠️ Key reminder: the declines in HPQ and BBY are not evidence of a "consumer recession"; they are evidence of a "memory cost shock." Do not conflate the two, or you will wrongly hit names like DG that genuinely benefit from consumer migration.

---

8美光MUWatch closely | Score 62 | Pre-market +3.97% · Micron

MU has no news of its own today. Its rise comes from two independent pieces of first-hand external evidence — the hardest "indirect catalyst" of the day.

1. NVDA 8-K text: supply commitments rose from $119 billion to $279 billion, "primarily related to memory purchases"; gross margin guidance 75.0% → 74.0%, with management saying Q4 troughs at about 71–72%, attributed to memory costs.
2. HP's results: RAM is 35% of PC bill-of-materials cost, the price pass-through is not working, PC margin 5.2% → 4.6%, and recovery only in FY2027.

Why these two matter: they are not sell-side research — they are two downstream customers, both with an incentive to push prices down, admitting in their own statutory disclosures that the memory makers took the profit. §5① estimates this transfer at about $17 billion of gross profit annualized.

Fundamentals (FY26Q3, ended 2026-05-28)

Quarter	Revenue	YoY	Gross margin	Operating margin
2026-05-28	$41.46 billion	+345.7%	84.6%	80.4%
2026-02-26	$23.86 billion	+196.3%	74.4%	67.6%
2025-11-27	$13.64 billion	+56.7%	56.0%	45.0%
2025-08-28	$11.31 billion	+46.0%	44.7%	32.6%

Table scrolls horizontally

Gross margin rose 39.9pp over four quarters; TTM operating cash flow / net income is 102% (clean cash quality); net cash $19.65 billion.

⚠️ But the valuation position is this brief's most important warning on MU

Metric	Current	Two-year range	Position
PE	21.2	16.0–28.2	Mid-range (looks cheap)
PS	11.7	3.2–11.5	Through the top of the range
PB	10.5	2.1–10.3	Through the top of the range

Table scrolls horizontally

This is the classic "low PE + high PB" cycle-top formation, not "cheap." The PE is low because the denominator uses peak-cycle TTM EPS; PB is what the market is willing to pay for the cycle position, and it has already broken a two-year high. In deeply cyclical industries, a low PE has never been a margin of safety.

Other risks: the financial data is about 3 months stale and the next print is 9/30 — a full month with no catalyst of its own, during which it has to absorb all the macro risk including the Warsh speech alone. The claim that "2027 DRAM/HBM capacity is already sold out" comes from industry hearsay relayed second-hand, not company disclosure; flagged to be verified. The stock is still −25.2% from its 52-week high, meaning the market once paid a higher price for this cycle at some earlier point.

Final verdict: watch closely, but note that this is "theme tracking," not "event trading." The industry logic got strong external corroboration today, but the valuation is already in a cycle-top formation and there is no company data to verify for a month. It is closer to a theme requiring multi-month tracking than an event verifiable the same day.

⚠️ The most important reservation in this section (added after QC): this brief reasons from "both NVDA and HP identify rising memory costs" to "MU benefits," and one link in the middle is not closed — NVDA buys mostly HBM, and global HBM share is led by SK Hynix with Samsung second; this brief did not obtain MU's share of NVDA's memory purchases; what HP is squeezed on is PC DDR5/LPDDR, which is neither the same sub-market nor the same pricing curve as HBM. What the two disclosures jointly confirm is "memory prices are rising industry-wide," and the beneficiary is the memory industry as a whole; "how much MU gets" cannot be derived from these two disclosures, and is flagged to be verified. This is the biggest logical gap in this brief's MU section, and until SK Hynix's and Samsung's current-period data and MU's share are filled in, MU should not be equated with the sole beneficiary of this main line.

---

⑨ WEN — Wendy's | Avoid / short watch | Pre-market −14.71%

Related news: after the close on 8/26, Reuters reported that Nelson Peltz's Trian fund has "no current intention" of making a take-private offer for Wendy's, citing operating performance, valuation multiple and the current strategic direction. Trian holds more than 24% and is the largest shareholder.

Catalyst logic: this is the worst kind of negative — not bad news, but "the rescue is not coming." Wendy's share price has long embedded a slice of option value for "Trian might take it private." Today that option was voided by the person who knows the company best, and the stated reason for voiding it is that he is bearish on the fundamentals. That hurts more than an earnings miss, because it simultaneously negates the valuation floor and the turnaround story.

Fundamentals: market cap only $1.72 billion, PE 13.7, dividend yield 3.10% (already halved from earlier levels). 2025 revenue $2.18 billion (−3.1%), profit $165 million (−15.07%) — note: these are FY2025 annual report figures, already about 8 months stale. Burger King has already overtaken it as the second-largest burger chain in the US.

The most painful number: the analyst mean target is $7.79, only about 1.0% above the pre-market price of $7.71. Even after a 14.71% collapse, the sell-side consensus offers less than 1% of upside.

Final verdict: avoid, short watch possible. But it has already gapped −14.71%, so the entry for a short is poor; the market cap is only $1.7 billion, so borrow cost and liquidity need to be checked independently. Its main value in this brief is as the evidentiary end of K-shaped consumer divergence.

---

⑩ MRVL — Marvell | Watch only | Score 46 | Pre-market +3.93% | Reports after the close tonight

A data trap must be corrected first: of MRVL's $2.53 billion of TTM GAAP net income, the quarter ended 2025-11-01 contributed $1.90 billion on revenue of just $2.07 billion that quarter (a 92% net margin) — operationally impossible. That was a one-off gain (the specific source has not been verified; flagged to be verified). The TTM PE of 84.7x and PB of 11.8x therefore do not reflect operating earnings power and should not be used directly.

The driver of the +40.5% over the past month has been traced to a first-hand source: MRVL filed an 8-K on 2026-08-19 under Items 1.01 (material agreement) + 3.02 (unregistered sale of equity) + 9.01 (SEC index verified). Media describe the content as a commercial custom-AI-chip agreement with Google, together with the issuance to Google of $12.2 billion of purchase-linked equity warrants. The 8-K item numbers match "material agreement + equity issuance" exactly; the counterparty and the amount come from media and are flagged to be verified.

⚠️ This structure has to be called out separately: customer warrants linked to purchase volume are typically accounted for as part of the consideration, reducing revenue while diluting shareholders. It is the mirror image of NVDA holding customer equity — the same class of "circular" structure. Any demand argument built on it must net that layer out.

Fundamentals: FY27Q1 (ended 2026-05-02, already about 4 months stale) revenue $2.42 billion (+27.6%), gross margin 52.1%, operating margin 14.5%, net debt $1.43 billion. PS 24.7x and PB 11.8x are both at the top of their two-year ranges; forward PE 60.4x.

Tonight's consensus: revenue $2.71 billion (+35.0%), non-GAAP EPS $0.93.

Final verdict: watch only; no directional position ahead of earnings. It is already +3.93% pre-market, and before the results are out, any MRVL fundamental judgment based on May data is unusable. What actually needs checking tonight is where the warrants sit in the income statement — whether they reduce revenue, and how much is recognized this quarter. That determines whether this Google agreement is revenue or cost.

---

6. Negative / Avoid List

---

7. Within-Theme Ranking

Theme 1: Profit Redistribution from Rising Memory Prices (today's hardest logic)

This table carries two pieces of information that matter most.

First, the DELL counterexample actually makes the theme more precise. I originally listed DELL as "same exposure as HPQ, to be verified." After measurement that assumption has to be withdrawn: HPQ −11.83%, DELL +1.10%. The reason is that their revenue mixes differ — DELL's AI server and storage businesses put it on the beneficiary side (selling AI hardware) at the same time, offsetting the injured side of being a memory buyer; whereas HPQ is mostly consumer and commercial PCs, so it only has the injured side. So the short side of Theme 1 is not "everyone who buys memory," it is "those who buy memory without an AI hardware hedge" — specifically PCs and consumer electronics channels (HPQ, BBY), not DELL. DELL reports on 9/1, at which point the hedge can be verified.

Second, the market treats NVDA as a pure winner, but on this chain NVDA is also a tenant. It has covered up the gross margin it is giving away with revenue volume. If revenue growth slows while memory prices do not fall, that cover stops working — which is NVDA's most underrated risk.

Third (added after the supplemental scan): the "landlords" are not just MU — on the storage side, STX at +4.88% and WDC at +4.50% are up more than MU today. This is the theme's most powerful breadth confirmation — if only MU were rising it could be a single-stock event; it takes memory (MU) + hard drives (STX/WDC) + enterprise storage (P), four names moving the same way, to show the market repricing the whole "storage supply side."

⚠️ But the positioning risk here is the most extreme in the brief: STX and WDC are both up more than 200% this year, and over the past 12 months STX is up roughly 600% and WDC roughly 850%, with both in all-time-high territory. The two drew down about 9% in a single day on 8/18. This brief lists them as evidence of the theme's breadth, not as investable entries — after a 6–8x move, the position is not something a pre-market brief should adjudicate.

Theme 2: AI Compute (NVDA Ecosystem) — The Spillover Is Highly Selective

The most important information in this table is the ordering of the moves itself: NVDA +6.07% while AMD/AVGO are up only about 1.2%. If today were a broad "AI theme rally," AMD and AVGO would not be up just 1%. The market is precisely distinguishing "NVDA's supply chain" from "NVDA's competitors." The stronger NVDA is, the weaker the ASIC-substitution narrative. Do not trade today at the coarse granularity of "the semiconductor sector."

Another fact that should not be dodged: over the past 3 months AVGO / VRT / ALAB are −15.7% / −17.5% / −10.8% respectively. The share prices of the supporting chain were already weakening before this NVDA print, and not one of the 10 names in the group is at a 52-week high. That sits in tension with the "the whole AI chain is improving" narrative.

Theme 3: Software / Cybersecurity Repricing — Quality Is Layered

Note on the ranking: names 1–3 have their own earnings, but their quality differs by three notches — the fact most easily buried by "software rallied across the board" today. Names 5–6 owe today's gains entirely to other companies' earnings; such coattail names typically rise on day one of a theme and are the first to give it back on day two when there is no data of their own to carry the baton.

One additional finding: ZS (Zscaler) is −49.5% from its 52-week high while revenue is +25.4% and FCF margin is 30.4% — not very different from CRWD — yet its PS is only 8.68x versus CRWD's 35.7x. That divergence is worth studying on its own, but it needs its 9/3 earnings data to characterize, so this brief draws no conclusion.

Theme 4: K-Shaped Consumer Divergence

Conclusion: comp growth is nearly identical (+3.5% vs. +3.7%), and DLTR's headline EPS beat is far larger (143% vs. 23%), yet the market priced them in exactly opposite directions. The difference lies in only two places: ① DG's growth comes from traffic (a genuine migration in customer count), DLTR's from ticket (pure price pass-through); ② $1.31/share of DLTR's beat is a tariff refund, while DG shows no such item.

This pairing also validates two of this brief's methodological rules: first, "positive traffic is what makes it a real signal"; second, "the more absurd the headline beat, the sooner you should go looking for one-off items" (DLTR's 143% and CRM's 80% are the same species).

---

8. Opening Verification Signals

Pre-market (now to 09:30 ET)

1. Whether NVDA is gap-and-go or gap-fill — today's single most important signal. Pre-market $222.38 (+6.07%, notional about $2.7 billion, the most credible quote of the day), 52-week high $236.54. If it pushes above $230 on volume within the first 30 minutes, the market has accepted the 70% FY2028 outlook; if it falls quickly below $212 (erasing most of the gap), the doubts about "gross margin down to 71–72%" and "circular financing" have won. The latter carries more information.
2. ✅ The 08:30 ET data is out, and it leans hawkish: initial claims 203,000, below the 208,000 expected (stronger than expected); continuing claims 1.778 million, below the prior 1.796 million — note: this brief did not obtain the market consensus for continuing claims, so the phrase "stronger than expected" applies only to initial claims. This is today's first real variable, in the same direction as yesterday's hot core PCE. What to watch: whether the 10Y uses it to break above 4.70% before the open — if so, today's biggest gainers among high-multiple software (CRWD forward PE 150, OKTA 34) take the first hit.
3. ⚠️ Pre-market notional is the most practical column in this brief. NVDA (about $2.7 billion), MU (about $864 million), MRVL (about $248 million) and CRM (about $156 million) have highly executable quotes; whereas VRT (about $15.9 million), ALAB (about $16.2 million), PANW (about $16.6 million), CRDO (about $17.9 million) and S (about $1.3 million) are thin-book drift, and their pre-market percentage moves should not be used as pricing conclusions.
4. ⚠️ Pre-market volume and price accumulate; a single snapshot is not a fact. Several names in this brief have readings at multiple timestamps (e.g. CRM: 08:03 ET $227.24/+10.5% → 08:35 ET $226.27/+10.04%; OKTA: $158.21/+17.7% → $158.51/+17.92%). They do not conflict; they are cumulative results at different moments. Any conclusion based on a single-point pre-market quote must be re-confirmed after the open.

Intraday (first 30 minutes)

5. Sector ETF sympathy: software (IGV) vs. semis (SMH). If today's main-line call is right, IGV should be materially stronger than SMH. If it is the other way round, immediately re-rank with SMH as the reference.
6. The falsification test for the software line lies in the names that did not report: whether PANW and NOW can hold their gains. The three that reported are supposed to rise; whether the two that did not can hold is what decides between "sector repricing" and "three isolated events."
7. Two-sided verification of the memory chain — this is the necessary condition for Theme 1 to hold. MU up AND HPQ/BBY down; both sides must hold simultaneously. If MU falls too, then today is merely "the earnings losers being sold" and there is no structural profit-redistribution story.
8. The relative strength of NVDA vs. MU is a hidden signal. If MU keeps outgaining NVDA, the market is starting to price "gross margin transfer" — which is negative for NVDA and positive for MU, the opposite reading from the surface "NVDA's good news lifts the whole chain."
9. Divergence between the Dow and the Nasdaq. Pre-market, Dow futures are slightly lower while Nasdaq futures are +0.5%. If after the open the Dow turns positive while the Nasdaq weakens, that is a rotation from growth into value, in the same direction as the 10Y at 4.66%.

Options and Sentiment

10. Post-earnings IV crush: CRWD, CRM, OKTA and NVDA have all reported, so implied volatility will collapse quickly after the open. The risk/reward of buying options to go long these names deteriorates sharply after the open — even a correct directional call can lose money on falling IV.
11. VIX around 15.0 is a low-volatility regime. The absolute cost of options protection is at a low, and there is an event tomorrow that cannot be priced in advance. This is an objective description of the current volatility environment and does not constitute any trading recommendation.

Risk Signals (reduce position if they appear)

12. A lone runner: if NVDA rises while MU/VRT/MRVL all turn down, it is passive money buying NVDA, not industry logic.
13. Stock-bond divergence: the 10Y has risen two sessions running to 4.66%. If it keeps rising through 4.70% today while the Nasdaq is still up, that divergence is usually resolved by the equity side.
14. Tomorrow's 10:00 ET Warsh speech: he explicitly does not use traditional forward guidance, so the market cannot price it in advance. That means today's late-session overnight exposure faces an event of unknown variance.

---

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. CRM (downgraded from watch closely to watch only) — the true beat is only +3.1%; the "80% beat" was manufactured by a $2.613 billion mark-up on the Anthropic stake; GAAP operating income is flat year over year; organic revenue growth 8.8% → 6.4%; net debt rose from $4.9 billion to $27.9 billion. The cheapness is real, but a large part of today's +10.5% is paying for a beat that does not exist.
2. Names with too little pre-market notional — S (about $1.3 million), VRT (about $15.9 million), ALAB (about $16.2 million), PANW (about $16.6 million), CRDO (about $17.9 million). These quotes are thin-book drift, not executable prices.
3. Names that have gapped above their 52-week highs — OKTA at a pre-market $158.21 vs. a 52-week high of $157.00, and only 5.6% from the sell-side mean target.
4. Names reporting tonight — MRVL, S, ULTA. MRVL in particular: $12.2 billion of warrants typically reduce revenue in accounting terms, which is a variable of unknown direction.
5. The downstream bearers of the memory cost shock — HPQ, BBY. ⚠️ Note that DELL has been checked and excluded: pre-market +1.10% ($463.82 → $468.90); its AI server business hedges the memory exposure, so it does not belong here (see §7). A low valuation is not support, because the cost pressure is outside their control and HP itself says recovery only comes in FY2027.
6. Names whose own company negated this quarter's data — SMCI: sequential revenue guidance of +30%~+39% against non-GAAP EPS guidance of −35%~−41% sequentially, which is the company telling the market "the 17.5% gross margin is not a run rate."
7. Any heavy overnight position — tomorrow at 10:00 ET Warsh delivers his first Jackson Hole keynote as Chair, and he explicitly does not use forward guidance.

④ The Final One-Line Judgment

On the surface today is a double win of "AI triumph + software revenge," but once the original 8-Ks are laid out, the real structure is a profit-transfer chain: NVDA used 106% revenue growth to cover gross margin being handed down from 75% to 71–72%, HP used 18% revenue growth to cover unit shipments of −16%, and both companies simultaneously point at the memory makers in their own statutory disclosures — so the hardest thing today is not "buy AI" but "buy the rent-collecting end of this chain and avoid the rent-paying end"; meanwhile the three software prints differ by three notches in quality (CRWD genuinely accelerating > OKTA a genuine beat but with next-quarter guidance stepping down > CRM's beat being a one-off investment gain), yet the market gave them almost identical moves, and that mismatch is itself the dividing line between opportunity and trap; and with core PCE at 3.3%, the 10Y at 4.66%, hikes still being priced this year, and Warsh's debut tomorrow morning impossible to price in advance — the right approach is to select names by evidence grade rather than buy by sector, and this brief's evidence-graded ranking puts the group that is "cross-verifiable against first-hand sources and not positioned at an extreme" (DG, NVDA) at the top, explicitly avoids "loud narrative but the numbers do not support it" (CRM) and "the quote is simply not executable" (S/VRT/ALAB/PANW/CRDO), and notes that tomorrow morning's Warsh speech makes overnight exposure an event of unknown variance.

---

⚠️ Risk disclaimer: this list is pre-market information organization 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 occur; automatically generated content may contain stale information or factual errors. Rely on company disclosures and SEC filings, and do not use this directly as a basis for trading.

---

Data Basis and Items Not Obtained (Disclosure to Clients)

• Pre-market prices and percentage moves: readings taken 08:00–08:08 ET, source stockanalysis.com, cross-verified against the CNBC quote API (10/10 agreement), baseline the 2026-08-26 close. Pre-market trading is cumulative, so the same name will read differently at different moments (this brief labels two readings each for CRM and DG; they do not conflict).
• "Pre-market notional" = pre-market shares traded × pre-market price, computed by this brief and used to judge quote executability. Anything below about $20 million has been flagged "thin" in the body, and its percentage move should not be used as a pricing conclusion.
• ⚠️ The baseline used for "distance from the 52-week high": this brief uses the pre-market price basis in three places — NVDA (−6.0%), CRWD (−8.9%), DG (−14.5%) — as explicitly labeled in the body; everywhere else (AVGO −28.2%, NOW −35.4%, MU −25.2%, ZS −49.5%, CRM −23.6%) uses the 8/26 closing price basis. On a day like today when everything gapped, the two bases can differ by 8 percentage points (CRWD: −16.8% on the close basis vs. −8.9% on the pre-market basis), so please note which one is being cited.
• Sell-side mean targets are used only as a yardstick for "where the market consensus sits"; they are not this brief's price targets, and this brief does not provide price targets.
• Financial data: the key figures for NVDA, CRM, OKTA, CRWD and SMCI are all taken from the original SEC 8-K / EX-99.1; HP, BBY and DG are taken from company press releases and 8-Ks.
• NVDA's FY2028 "about 70%": spoken by the CFO on the 2026-08-26 call; it is not in the press release, not in the 8-K, and is not formal guidance. A full-text search of the 8-K (113,510 characters) confirms the Outlook section contains only the three Q3 items plus the FY27 tax rate.
• NVDA's FY27 Q4 gross margin of 71–72%: media paraphrase of the call, not in the 8-K; the 8-K only goes as far as Q3's 74.0%.
• Consensus-basis warning: judging NVDA's guidance beat/miss uses the pre-print consensus ($103.8–104.2 billion). The $108.67 billion currently displayed on terminals is a post-print upward revision, and using it produces the opposite conclusion.
• Informatica's contribution to CRM's cRPO was not disclosed — the company disclosed only the revenue contribution ($456 million), not the cRPO contribution. This brief therefore cannot confirm whether the cRPO acceleration is organic, as flagged in the body.
• The counterparty of the MRVL–Google agreement and the $12.2 billion warrant amount: the 8-K item numbers (1.01/3.02) have been verified, but the specific terms were not read in the original; flagged to be verified.
• The nature of MRVL's $1.90 billion one-off gain in FY26Q3: not verified, so this brief explicitly does not use its TTM PE/PB.
• P (Everpure)'s pre-market percentage move is not given: two figures within the data source ($102.81 → $111.07 back-computes to +8.03% vs. the page's labeled +1.99%) are inconsistent and cannot be reconciled, so only the price and reading time are given. The data source now shows the company as Everpure, Inc. (NYSE: P), i.e. the former Pure Storage has changed its name and ticker.
• Names for which no reliable pre-market quote was obtained: BURL, ULTA. This brief draws no directional conclusion on these two. (DELL has been checked: $463.82 → $468.90, +1.10%, which overturned this brief's original assumption that "DELL is in the same bucket as HPQ"; see §7.)
• "2027 DRAM/HBM capacity already sold out," VRT backlog >$15 billion, AVGO's $162 billion backlog: all are industry hearsay or stale cumulative figures rather than current company disclosure; this brief flags them to be verified and has not used them for any conclusion.
• The 8/26 closing levels of the three major indices: sources disagree (the S&P 500 has two versions, 7,707.16 and 7,677.24), so this brief uses only the percentage moves on which all sources agree (S&P −0.02%, Dow −0.21%, Nasdaq −0.08%) and does not publish closing levels for the Dow or the Nasdaq.
• ⚠️ The pre-market movers list covers only the top 20 (data source pagination failed); names ranked 21 and below were not obtained. The strength ranking in §0② and §3 holds only within that sample. This brief ran one supplemental scan with a second data source (investing.com), adding STX, WDC, DLTR, NDSN, FANG, MRNA, MSI, MCK and others, but it still does not constitute a full-market census.
• ⚠️ The biggest open question in the NVDA section (previously logged internally only, now moved here): when the CFO said FY2028 growth of about 70%, she qualified it as a "supply-constrained outlook." That phrase has two opposite readings — it could be "the floor once supply constrains us," or "the ceiling that only holds if supply arrives on plan." This brief did not obtain the official call transcript and cannot confirm which. That difference directly determines whether the 70% is conservative or aggressive, so please do not treat it as a one-way positive before it is confirmed. Likewise, both "Q4 gross margin troughing at 71–72%" and "attributed to memory costs" come from media paraphrase, not the original 8-K.
• Today's 08:30 ET economic data was updated with actual figures at 08:35 ET: initial claims 203,000 (expected 208,000, prior 207,000), continuing claims 1.778 million (prior 1.796 million), source the US Department of Labor. This brief did not obtain actual figures for the goods trade balance and wholesale inventories, which remain expectations and are labeled as such in the body.
• DG's pre-market move is internally inconsistent within the data source, and this brief has adopted the self-consistent version: the single-stock page gives "$132.00 / +9.22%," but back-computing from the previous close it itself lists, $122.78, yields only +7.51% — the triple is not self-consistent; the movers-list page's "$135.30 / +10.20%" reconciles exactly with the previous close. This brief uses the latter and discards the $132.00.

Data Collection and Quality Log (Internal)

Data collection failures in this run

• yfinance fully rate-limited: two sub-agents made 19/19 calls that all returned YFRateLimitError; fast_info threw outright, and the query2 chart endpoint returned 429 under concurrency. This brief used no yfinance figures. Substitute channels: SEC EDGAR full text (data.sec.gov + Archives) + stockanalysis's __data.json (40/40 successful) + CNBC restQuote cross-verification.
• Main process got 403 from SEC EDGAR: the main process fetching sec.gov/Archives/... directly got 403 both times (NVDA q2fy27pr.htm, CRM q2fy27 exhibit 99.1). But two sub-agents succeeded using different methods — meaning the 403 is intermittent/UA-related, not a site-wide block. This was the most important failure recovery of the run: CRM's GAAP → non-GAAP reconciliation was stuck behind the main process's 403 and had to be left blank at drafting time; only after the sub-agent retrieved it did we reach the "true beat +3.1%" conclusion.
• CNBC site-wide 403: every article page returned 403, so it could not be used as a first-hand source.
• Bloomberg 403, marketwatch refused at the tool layer, globenewswire 60s timeout (switched to nvidianews.nvidia.com successfully), yahoo finance 429.
• Some stockanalysis paths 404: /markets/, /markets/indexes/, /markets/futures/ and /quote/index/VIX/ all 404; single-stock pages and /markets/premarket/{gainers,losers}/ work normally. Indices and VIX were switched to tradingeconomics.
• stockanalysis pagination broken: ?p=2 does not work, so the pre-market movers list beyond rank 20 was not covered.

⚠️ Two errors of the same kind that I made myself this run (one caught by a sub-agent, one caught by back-computation in self-check)

Error 1 (NVDA previous close): on the main process's first NVDA fetch, WebFetch returned "222.59 +12.93 (6.17%)" but reported the previous close as $213.05. I used $213.05 directly while drafting, without back-computing. In fact $222.59 − $12.93 = $209.66 is the previous close; $222.59/$213.05 is only +4.48%, which does not match +6.17% — the draft was already internally contradictory the moment it was written. After the sub-agent cross-verified with CNBC restQuote, it produced $209.66 / $222.38 / +6.07%, which is self-consistent.

Error 2 (DG's percentage move): the draft said "$122.78 → $132.00, +9.22%." Before finalizing I ran the whole-table back-computation script and found $132.00/$122.78 is only +7.51%. Looking back: the single-stock page's triple is not self-consistent, while the movers-list page's "$135.30 / +10.20%" reconciles exactly with the previous close. Switched to the latter and disclosed the discrepancy to clients. This error would have shipped had I not run the script, and it would have made DG's "good position" conclusion look better than reality (the pre-market price has in fact already moved past the sell-side mean target of $132.31).

Error 3 (DELL, caught on the spot during the supplemental check): the single-stock page gave "previous close $451.50, pre-market $468.90 +$5.08 (+1.10%)." $468.90 − $5.08 = $463.82 is the previous close; $468.90/$451.50 is +3.85%, which does not match +1.10%. Third time, same field, same site.

Error 4 (the baseline for "distance from the 52-week high," caught by the same back-computation script): §5② said "the pre-market $207.23 is −16.8% from the 52-week high." In fact −16.8% is computed from the closing price of $189.18; using the pre-market price of $207.23 gives −8.9%. On a gap day those two bases differ by 8 percentage points, and it happens to be exactly the number that decides "whether CRWD is still chaseable" — read as −16.8% it looks like there is plenty of room; read as −8.9% you realize it is already close to the prior high. The same check also found: CRWD's pre-market price is only 5.85% from the sell-side mean target (previously not written at all), OKTA's "consumed 76% of the target headroom" should be 73%, and NVDA's 6.4% should be 6.0%. All corrected, and the client-visible basis section now states one by one which places use the pre-market price and which use the close.

Common root cause and the practice now locked in: errors 1–3 all occurred on the same data source's single-stock page previous close field, while the movers-list page was self-consistent every time — this is no longer sporadic; the field is untrustworthy; error 4 is downstream of the same root cause: mixing the previous-close and pre-market-price baselines. For any positional metric — distance from a high, distance from a target, range percentile — the price used must be stated explicitly.

• Criterion: for any "price + change + percent change" triple, any two imply the third, and it must be back-computed on the spot.
• Locked into the process: run the whole-table back-computation script once before finalizing (20 names checked one by one this run; only DG failed), row by row, not by sampling.
• Data source preference: within the same site, the pre-market movers page > the single-stock page; the single-stock page's previous close field was wrong twice this run.

Instances of search summaries giving wrong numbers (all refuted with first-hand sources)

1. CRWD's three numbers: the summary gave "net new ARR $255.8 million (+32%), total ARR $5.51 billion, full-year ARR guidance $6.53–6.56 billion." Compared against the original 8-K, all three are FY27Q1 (end of April) figures, matching the prior quarter's 8-K verbatim. The actual Q2 figures are $332.8 million / $5.84 billion / $6.603–6.612 billion. Taking the summary at face value would have turned a 16% beat and acceleration into a below-guidance deceleration — the exact opposite conclusion. What exposed it was the $284–286 million guidance range logged in yesterday's report: $255.8 million is below the low end of guidance, which contradicts "the stock is +9.4%." A verification point written down yesterday saved us from an inverted error today.
2. NVDA's pre-market move: stockmarketwatch gave "+7.2%/$224.69," Benzinga gave "+7.32%"; measured it is +6.07~6.17%.
3. AIXI / DAIC: Benzinga's summary said AIXI +89.7% and DAIC +94.7%; stockanalysis measured AIXI at −26.95% and DAIC at +15.15%. AIXI's direction is inverted. Both are low-notional penny stocks and were not used.
4. The 8/26 closing levels: several summaries treated a Zacks-style "Stock Market News for Aug 26" (actually an 8/25 close recap) as the 8/26 close. 7,677.24 → 7,707.16 back-computes to +0.39%, irreconcilable with −0.02%, so publishing levels was abandoned.
5. July CPI: one summary claimed "CPI +3.4%, core down to 2.5%," which is incompatible with the verified core PCE of 3.3% (core PCE is usually below core CPI). Not used; PCE only throughout.
6. NVDA's Q3 consensus: terminals today show $108.67 billion, which is a post-print upward revision; the pre-print bar was $103.8–104.2 billion. This is the most insidious trap of all — getting it wrong turns "the entire guidance range is above consensus" into "guidance missed."

The five most valuable cross-verifications this run

1. CRM's $5.90: $5.90/$3.27−1 = 80.4%, matching the "80.38%" the media reported exactly, proving that is indeed how they computed it. The company's own footnote quantifies "investment gains impacted non-GAAP diluted EPS by $2.53," and $5.90−$2.53=$3.37 vs. consensus $3.27 = +3.1%. Three independent tests (Q3 guidance of $3.42–3.44 connecting, $2.53 of the $2.60 full-year raise being the gain, FY28 consensus 6% below FY27) all point to the same conclusion. The control group, ServiceNow, excluded $273 million of investment gains in the same period — same industry, same quarter, opposite treatment; that single fact is stronger than any derivation.
2. The NVDA ↔ HP two-way identification: NVDA's 8-K says "$279 billion of commitments primarily for memory purchases" + gross margin falling to 71–72% because of memory; HP says "RAM is 35% of BOM, the pass-through is not working, recovery only in FY2027." Two independent companies, two statutory disclosures, pointing at the same beneficiary. This is the hardest chain of reasoning in the brief, and it rewrites NVDA from "pure winner" into "also a rent payer."
3. HPQ's $0.19 tariff refund: of the full-year EPS guidance raise (midpoint +$0.24, $3.00 → $3.24), $0.19 comes from the tariff refund (Q3 $0.11 + Q4 $0.08), about 79%, and stripping it out leaves the high end completely unmoved. The draft used the low-end-to-low-end +$0.29 (66%); QC flagged the basis inconsistency — every other guidance figure in this brief uses midpoints, so this one should be unified to midpoints too. That restatement amount is in the press release details, not in the guidance table. It is the same species as HD's $685 million tariff refund and ANF's "beating guidance is not beating last year" in memory.
4. BBY's $34 million tariff refund: same day, same species as HPQ's. Two unrelated companies dressing up results with tariff refunds on the same day suggests this is a systematic basis issue this earnings season, worth checking proactively in the next several briefs.
5. Back-computing CRWD's acquisition cash flow: six-month acquisition cash outflow of $881 million = the Q1 standalone figure → Q2 standalone is 0 → the $332.8 million is purely organic. Using the cash flow statement to prove "organic" is more reliable than trusting the company's adjectives.

A major premise error corrected during this run's self-check I initially assumed, from general knowledge, that "the Fed is in a cutting cycle" and "Powell hosts Jackson Hole." In reality: the Chair is Kevin Warsh (sworn in 2026-05-22), the policy rate of 3.50–3.75% has been on hold for a year, the market is pricing 1–2 hikes this year, and Jackson Hole is 8/27–29 with Warsh's keynote at 8/28 10:00 ET. Had I not checked, the entire macro section would have been wrong while remaining internally self-consistent and incapable of exposing itself — the same class of status-type factual error as "Changxin Technology is already listed" in memory. Status-type facts (who holds the office, what the rate is, which day the meeting falls on) must be checked every single time.

Three new findings worth writing into long-term memory

1. Consensus itself gets revised upward after a print, so judging "last night's guidance" against "current consensus" produces an inverted conclusion. The NVDA case: $104.2 billion pre-print → $108.67 billion on terminals today, against guidance of $108 billion. Beat/miss judgments must lock onto the pre-release bar.
2. Whether non-GAAP excludes strategic investment gains and losses is a company option, and peers in the same industry can do the opposite. CRM does not exclude (the gain goes into non-GAAP in full); NOW excluded it in the same period. When a beat looks abnormally large, the first move is to count the add-back lines in the reconciliation table and read what they are — not to go looking for an explanation.
3. The "price + change + percent change" triple must be back-computed on the spot. I got caught on NVDA's previous close this very run (see above).

risk-auditor QC results (14 red flags, all addressed)

The 3 that changed conclusions (most valuable)

1. I judged OKTA's cRPO wrong and deducted points on that basis. I treated "the company guides Q3 cRPO to +11%12%, below this quarter's +14%" as a deceleration negative and deducted 13 points. QC pointed out: last quarter the company's original guidance for this quarter was also +11%, and it delivered +14% — the same conservative convention. In the same table I both wrote down last quarter's original guidance of +11% and treated this quarter's +1112% as a negative — two pieces of my own evidence contradicting each other without my noticing. The deduction has been withdrawn, the total went 64 → 67, and the item was downgraded from "negative" to "to be verified."
2. I used the wrong basis for HPQ's tariff refund share. I wrote "of the $0.29 full-year guidance raise, $0.19 comes from the refund, or 66%" — but $0.29 is low end to low end; every other guidance figure in this brief uses midpoints, and the midpoint raise is $3.00 → $3.24 = +$0.24, making the share 79%. Using the wrong basis actually understated the severity of the problem.
3. I did not measure NVDA's valuation with the same ruler as CRM's. I spent an entire section criticizing CRM for leaving a $2.613 billion investment gain inside non-GAAP, yet when saying "NVDA is at two-year lows on all five valuation metrics" I was using the GAAP basis, which includes $7.771 billion of non-cash equity investment gains. A note has been added.

The 4 that tightened the reasoning 4. One link in the causal chain of the top main line is not closed: NVDA buys mostly HBM (share led by SK Hynix), while HP is squeezed on PC DRAM — different sub-markets. What the two disclosures confirm is "memory prices are rising industry-wide," not "MU collects the rent." SK Hynix did not appear anywhere in the original text. Qualifiers have been added in four places: §2, §5⑧, §7 and §9②. 5. The $17 billion annualized figure was self-contradictory: the $123 billion base has no disclosed source, and it annualizes a "gross margin trough quarter" ×4 — exactly the single-quarter extrapolation I criticize SMCI for in §6. Changed to give only the static estimate of $4.3 billion per quarter, not annualized. 6. PANW was used as evidence and declared unusable at the same time: I used "NOW/PANW rose without reporting" to argue sector repricing, while the same brief says PANW's notional of $16.6 million is below my self-imposed $20 million threshold. It has been explicitly downgraded in §9②. 7. "Software was the strongest thing pre-market today" was a universal claim built on a truncated sample (covering only the top 20 of the list), and that gap was originally written only in the internal block. Changed to a qualified statement, and the coverage limit moved to the client-visible basis page.

7 basis and fact items: the direction of WEN's target price was reversed ($7.79 is above the pre-market $7.71; I wrote "below"), WEN's 8-month-stale results were not labeled, "prior" was used as "expected" for continuing claims, OKTA's FCF raise of +6.9% mixed the high end with the midpoint (the midpoint basis is +5.7%), CRWD's distance-from-52-week-high baseline was mixed, Everpure's renaming had not been confirmed against a first-hand source yet it was still scored and labeled (now downgraded to "to be verified · not scored"), and the DELL verification conclusion had not been carried into the conclusions section. All fixed.

What QC could not refute: the six columns of the scoring table sum to the stated totals; NVDA's GAAP > non-GAAP cause, the $279 billion year-by-year subtotals, DSO of 45/60 days, OCF / net income of 40%/86%, and CRM's $5.90−$2.53=$3.37 → +3.1% plus the three consistency tests all pass back-computation.

The single most memorable thing from this QC round: all three "conclusion-changing" red flags were "two pieces of my own evidence contradicting each other," not missing external information. The OKTA one is especially textbook — last quarter's +11% guidance was written in my own news table, in direct conflict with the "deceleration" judgment I gave. Before filing next time, run a dedicated "internal consistency check": when the same number appears multiple times in the text, has it been used to support opposite conclusions?

Not covered / known gaps

• Ranks 21–50 of the pre-market movers list were not obtained (pagination broken), so genuine mid-cap catalysts may have been missed.
• DELL unverified → verified and the original assumption overturned: DELL is +1.10% pre-market ($463.82 → $468.90), up rather than down, because the AI server business hedges the memory cost exposure. The plan to write it up as "the second name on Theme 1's short side" was wrong and has been rewritten as a counterexample. The lesson is of the same family as [[no-news-found-is-not-no-news]]: an item marked "unverified" must not carry its original assumption into the body; once verified, the conclusion may flip entirely.
• ULTA and BURL pre-market quotes were not obtained.
• The biopharma FDA / M&A line was scanned once: the FDA and M&A events retrieved (Regeneron garetosmab approval, Moderna/Merck phase III, Lantheus/Curium, Repligen/BioLife, Zymeworks/Theravance) all occurred on or before 8/19, outside this brief's 8/26 16:00 → 8/27 08:10 window. No major biopharma catalyst inside the window was found — that is a conclusion reached after checking, not a "could not find one."
• The 8/26 after-hours earnings were not exhaustively covered; seven companies were covered: NVDA/CRM/CRWD/OKTA/HPQ/NTNX/P.
• NVDA's official call transcript was not obtained, so both the "70%" and the "71–72% gross margin" rely on media paraphrase. In particular, Kress's "supply-constrained outlook" has two opposite readings (the figure is a floor after supply constrains it / the figure depends on supply arriving on plan), and which one could not be confirmed; this is the biggest open question in this brief's NVDA section.
• NVDA's 10-Q (accession 0001045810-26-000075) has been located but not read: its customer concentration, related-party disclosures, and the share of the $279 billion that is non-cancellable are the only route to confirming or refuting "circular financing" — news coverage cannot do it.