From DowJo — where this research becomes practice. Train your judgment before you risk your money.
Course Research · Episode 3
NVIDIA Beats Earnings. Why Could the Stock Still Fall?
The evidence-led takeaway
A company reports the past. A price is already a bet on the future.
Ask Jo about this researchJo reads this note before answering. Sign in to ask Jo with this evidence attached — you come straight back here.
In brief
What this research concludes
A company reports the past. A price is already a bet on the future. The only thing a report can move is the distance between them — so the question is never whether the results were good, it is whether they were better than what the price had already assumed.
The judgment skill this hands over
Before judging any result, find the expectation it is being measured against. Name it, date it, and name who published it. If you cannot, you are not measuring anything.
What we investigated
The question, and who it looks at
This is episode 3 of Investing in the AI Era, a 14-episode course. The film tells the story; this note is the research behind it, kept inspectable.
The research question the team set itself: What is the honest evidence that a company can report excellent results and the stock still fall — and what proves the symmetry, that a miss can be met with a rise?
Who and what this looks at
How we tested it
Researched to be refuted, not confirmed
Candidate claims went to an independent pass instructed to refute them against primary sources. Survivors became evidence; casualties became the refused list below.
The method, in the research team's own words
Seven independent research lenses (frontier window, historical case, earnings mechanics, valuation-as-expectations, counter-case, adversarial, attributed ideas) run in parallel; every lens's output sent to an independent fact-checker instructed to REFUTE it against primary sources; two numeric-series builders working only from SEC filings and named price vendors; and a completeness critic over the whole corpus. 17 agents, 1,051 tool calls, zero agent failures. 263 candidate claims, 99 of them returned as non-CONFIRMED. The critic's findings changed the episode: the vivid after-hours sequence that was going to be the spine rests on a single live blog with no named quote source and has been REMOVED (X01); the consensus vendor was standardised (X05); and the definition of 'the reaction' was promoted from an unexamined convention into a scene, because the critic pointed out that the episode's dependent variable was one nobody had justified.
The refutation pass, in numbers
263 proposed164 confirmed47 corrected31 partially unverifiable18 unverifiable3 refuted
Nineteen of the corrections were arithmetic or precision failures caught by recomputation from the filer, including the whole fourteen-quarter return series, which the fact-checker rebuilt from vendor daily data rather than trusting the lens. Three claims were hard-refuted: a superlative about the size of the 27 August move, a claim that NVIDIA changed its reportable segments, and a per-share comparison built on two different accounting bases.
What was screened before the case was chosen · 7 candidates
NVIDIA's fourteen AI-era prints, May 2023 to August 2026
SELECTED as the spine. Beat published revenue consensus in all fourteen; the stock fell the next day in seven. A base rate rather than an anecdote — one company, one measure, one vendor, and no single quarter's circumstances can explain it away.
One dramatic beat-and-fall quarter, told alone
REJECTED as the spine. It is what the title implies and it is the weaker episode: a single case invites the viewer to find a special reason for that day, which is the opposite of what the lesson needs.
The after-hours tape of 26 August 2026 — beat and guide public, stock down 1.3%
CUT AT VERIFICATION (packet X01). The most vivid version of the argument and the thinnest-sourced claim in the research: one live blog naming no quote vendor, and no minute-level after-hours data obtainable by any of seventeen agents.
NVIDIA, May 2025 — guided BELOW consensus and rose 3.3%
SELECTED as the counter-case. Proves the symmetry on the same company, the same measure, and the chart already on screen. No comparability problem, no confound, no second company to introduce.
Netflix, January 2023 — reported 12c against a 45c consensus and rose 8.5%
REJECTED. Fully researched and genuinely striking, but a co-chief-executive announcement arrived in the same release, and using it would have meant plotting an EPS surprise against NVIDIA's revenue surprises on one axis — a comparable-basis violation.
Oracle, September 2025 — missed both lines and rose about 36%
REJECTED. Same axis problem, plus a backlog concentrated in a single counterparty that no source can size reliably, plus a subsequent fall that would have pulled the episode into Episode 12's territory.
NVIDIA, August 2024 — beat revenue and EPS, guided above, and fell 6.4% while the S&P was flat
RETAINED IN THE PACKET, not narrated as a set piece. It is inside the fourteen-print series where it belongs; promoting it to a scene would have rebuilt the anecdote the spine was chosen to avoid.
What the evidence says
27 claims survived refutation
Each carries its source, the date the thing happened and the date it was said — two different facts — and its limits, stated by the research team rather than left for you to discover.
- FactFrontier · Aug 2026Primary sourceF01
NVIDIA reported its second quarter of fiscal 2027 on 26 August 2026, for the quarter ended 26 July 2026. The Form 8-K carrying the results was accepted by the SEC at 16:21:19 Eastern.
Source: NVIDIA Corporation Form 8-K (Item 2.02), SEC EDGAR, accession 0001045810-26-000073 (opens sec.gov)
Event 26 Aug 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Limits: The timestamp is EDGAR's acceptance time, not the wire time.
Research note
Anchors every point-in-time frame in the episode. Nothing in the report could be known before roughly 4:21 pm Eastern. The conference call began at 5:00 pm Eastern the same day.
- FactFrontier · Aug 2026Primary sourceF02
Revenue was $96.2 billion, up 18% sequentially and up 106% from a year earlier.
Source: NVIDIA Q2 FY2027 press release, Exhibit 99.1 to Form 8-K (opens sec.gov)
Event 26 Jul 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Limits: Narrate to one decimal ($96.2bn). The episode never states revenue to the million.
Research note
Filer table value $96,221M. The sequential and year-over-year percentages are the company's own, not a computation. Prior quarter $81,615M; prior year $46,743M.
- FactFrontier · Aug 2026Primary sourceF03
Data Center revenue was $89.0 billion, up 117% from a year ago — about 92% of the company's total revenue.
Source: CFO Commentary on Second Quarter Fiscal 2027 Results, Exhibit 99.2 to Form 8-K (opens sec.gov)
Event 26 Jul 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Limits: NVIDIA reclassified one customer between sub-lines during the quarter and recast the prior period, so sub-line splits are not comparable to previously published ones. The episode uses only the total.
Research note
$89,023M of $96,221M = 92.5%. Data Center and Edge Computing are 'market platforms', not GAAP reportable segments — the reportable segments remain Compute & Networking and Graphics. The episode does not decompose this figure; the value chain is Episode 4's subject.
- FactFrontier · Aug 2026Primary sourceF04
GAAP and non-GAAP gross margins were both 75.0%.
Source: NVIDIA Q2 FY2027 press release, Exhibit 99.1 to Form 8-K (opens sec.gov)
Event 26 Jul 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Research note
The high from which the same day's guidance stepped down. Pairs with F07.
- FactFrontier · Aug 2026Primary sourceF05
GAAP diluted earnings per share were $2.46 and non-GAAP diluted earnings per share were $2.22.
Source: NVIDIA Q2 FY2027 press release, Exhibit 99.1 to Form 8-K (opens sec.gov)
Event 26 Jul 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Limits: Never narrate GAAP EPS as 'better' without the equity-gains caveat. The episode avoids the problem by not using it.
Research note
GAAP exceeds non-GAAP this quarter, an inversion of the usual relationship, because GAAP net income of $59,688M includes $7.8 billion of net gains from equity securities that non-GAAP excludes. The episode narrates the non-GAAP figure only, because that is the figure the published consensus was set against.
- FactFrontier · Aug 2026Primary sourceF06
For the third quarter of fiscal 2027 NVIDIA guided revenue to $108.0 billion, plus or minus 2%, and stated that it is not assuming any Data Center compute revenue from China in that outlook.
Source: NVIDIA Q2 FY2027 press release, Outlook section, Exhibit 99.1 to Form 8-K (opens sec.gov)
Event 26 Aug 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Limits: The implied range is roughly $105.8bn to $110.2bn. The episode states the midpoint and the plus-or-minus, never a spuriously precise band.
Research note
The China sentence is the filer's own and is quoted on screen as filed.
- FactFrontier · Aug 2026Primary sourceF07
In the same release NVIDIA guided third-quarter gross margin to 74.0%, plus or minus 50 basis points — a step down from the 75.0% it had just reported.
Source: NVIDIA Q2 FY2027 press release, Outlook section, Exhibit 99.1 to Form 8-K (opens sec.gov)
Event 26 Aug 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Research note
This is the FILED version of the margin story and is what the episode narrates. The lower fiscal-fourth-quarter trough figure exists only in a call transcript and is handled separately as F08.
- FactFrontier · Aug 2026JournalismF08
On the earnings call the same afternoon, management told analysts that gross margins were expected to bottom in the fiscal fourth quarter in the 71 to 72 percent range, citing memory prices that had risen by more than the company had expected.
Source: NVIDIA Q2 FY2027 earnings call transcript (statement attributed to CFO Colette Kress), corroborated across two independent transcript publishers
Event 26 Aug 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Limits: Tier 3. Narrated as what management said on a call, never as a filed figure.
Research note
MATERIAL SOURCING NOTE, and it is on screen: this figure appears in NO NVIDIA filing. The CFO Commentary filed as Exhibit 99.2 does not contain it. Two transcript vendors agreeing is not the filer, and the episode says so. Memory pricing is Episode 5's subject and is not explored here — it appears as one clause inside a guidance sentence.
Show the remaining 19 pieces of evidence
- FactFrontier · Aug 2026JournalismF09
On the same call, management gave a preliminary expectation for fiscal 2028 revenue to grow approximately 70% — the first time the company had given a figure for a full year in advance.
Source: NVIDIA Q2 FY2027 earnings call transcript (statements attributed to CFO Colette Kress and CEO Jensen Huang), corroborated across two independent transcript publishers
Event 26 Aug 2026 · Published 26 Aug 2026 · Publicly knowable 26 Aug 2026
Limits: Tier 3, transcript only, absent from every filing. Narrated as reported speech with the sourcing on screen.
Research note
Huang on the same call: the company had never forecast or guided to a year in advance. Four separate analyst questions on the call concerned this figure. THE EPISODE'S USE OF THIS FACT IS STRICTLY LIMITED: it establishes that a number was published for which no consensus estimate existed, because nobody had been asked to make one. The episode does NOT claim this number moved the share price — see X02.
- FactFrontier · Aug 2026JournalismF10
NVIDIA's shares closed at $209.66 on 26 August 2026, at $227.98 on 27 August — a rise of 8.7% — and at $217.55 on 28 August, a fall of 4.6%.
Source: Daily closing prices, split-adjusted; Yahoo Finance chart API and StockAnalysis.com returning identical values, cross-checked against Nasdaq
Event 28 Aug 2026 · Published 28 Aug 2026 · Publicly knowable 28 Aug 2026
Limits: Narrated to one decimal. The episode never states an intraday or after-hours level — see X01.
Research note
Exact: +8.74% and −4.57%. Two independent vendors return identical closes. No return in this episode's series crosses NVIDIA's 10-for-1 split boundary of 10 June 2024, so every reported return is identical adjusted or unadjusted.
- FactFrontier · Aug 2026JournalismF11
On 28 August 2026 the Nasdaq Composite fell about half a percent, so NVIDIA's fall that day was several times the move in the market around it.
Source: Index closes for 28 August 2026 as reported contemporaneously; Nasdaq Composite 26,402.42, −0.52%
Event 28 Aug 2026 · Published 28 Aug 2026
Limits: Establishes that the move was not simply the market. It does NOT establish a cause — see X02.
Research note
Stated as 'about half a percent'. The episode does not quote index moves to two decimal places from journalism.
- FactPrimary sourceF12
Across all fourteen of NVIDIA's earnings reports of the AI era — from May 2023 to August 2026 — the company beat the published revenue consensus every single time. The stock fell on the next trading day in seven of them.
Source: Revenue from NVIDIA Form 8-K Exhibit 99.1 releases and SEC XBRL company facts (tier 1); consensus from Zacks Investment Research (tier 3); next-day returns computed from split-adjusted closes at two independent vendors
Event 27 Aug 2026 · Published 31 Aug 2026
Limits: 'Beat the published revenue consensus' is measured against ONE vendor throughout (Zacks), because it is the only vendor whose NVIDIA consensus history reaches back all fourteen quarters on a consistent basis. On other vendors' numbers individual surprises move by up to about a point; the direction — a beat in every quarter — does not change.
Research note
THE EPISODE'S SPINE. The seven negative next-day sessions follow the reports of 21 Nov 2023, 28 Aug 2024, 26 Feb 2025, 27 Aug 2025, 19 Nov 2025, 25 Feb 2026 and 20 May 2026. All fourteen returns were independently recomputed from vendor daily data during fact-check and matched. Every one of the fourteen reports was published after the US market close, so the next trading day is the market's first full opportunity to price it.
- FactPrimary sourceF13
NVIDIA's largest revenue beat of the period, about 21% above the consensus estimate in August 2023, was followed by a next-day move of one tenth of one percent.
Event 24 Aug 2023 · Published 31 Aug 2026
Limits: The surprise percentage is vendor-dependent; the episode states it as 'about twenty-one percent' and shows the vendor on screen.
Research note
Surprise +20.9%, next-day return +0.10%. The single most useful observation in the series: the biggest beat produced the smallest reaction.
- FactPrimary sourceF14
Two consecutive quarters — reported in November 2025 and February 2026 — produced an identical revenue surprise of about 4% against the same vendor's consensus, and next-day moves of minus 3.2% and minus 5.5%.
Source: NVIDIA Form 8-K Exhibit 99.1 releases (revenue $57,006M and $68,127M); Zacks consensus $54.7bn and $65.4bn (both +4.14%); next-day returns computed from split-adjusted closes
Event 26 Feb 2026 · Published 31 Aug 2026
Limits: Both surprises round to +4.1% on the vendor used. On a different vendor they would not be identical — which is itself in the episode.
Research note
The cleanest single frame in the packet: the same beat, on the same measure, from the same company, two quarters apart, with materially different results. No narration is required to make the point.
- FactPrimary sourceF15
In May 2025 NVIDIA guided the coming quarter to $45.0 billion, below the $45.9 billion analysts were expecting — and the stock rose 3.3% the next day.
Source: NVIDIA Q1 FY2026 press release, Outlook section (guide $45.0bn); LSEG consensus $45.9bn as reported contemporaneously; next-day return computed from split-adjusted closes
Event 29 May 2025 · Published 31 Aug 2026
Limits: The guidance consensus is vendor-published and this one comes from a different vendor (LSEG) than the revenue series (Zacks). The episode narrates it as 'what analysts were expecting' with the vendor on screen.
Research note
The mirror image, from inside the same company and the same series: the one quarter of the fourteen in which NVIDIA guided BELOW the published consensus, and the stock went up. Exact return +3.25%. This removes the need for a second company to prove the symmetry.
- FactFrontier · Aug 2026JournalismF16
In the second quarter of 2026, 86% of the S&P 500 companies that had reported posted earnings per share above the estimate — against a five-year average of 78% and a ten-year average of 76%.
Source: FactSet Earnings Insight, John Butters, 7 August 2026
Event 7 Aug 2026 · Published 7 Aug 2026
Limits: A later weekly edition of the same report exists and would move the figure slightly; the episode dates the number on screen. FactSet is a vendor, not a primary source.
Research note
The generalisation that stops the episode being one company's anecdote. Verified verbatim from the FactSet report by two independent retrievals.
- FactFrontier · Aug 2026JournalismF17
In the same period, the companies that beat saw their shares rise by an average of only 0.4% in the four days around the release — below the five-year average of 1.0%.
Source: FactSet Earnings Insight, John Butters, 7 August 2026 (window: two days before through two days after the release)
Event 7 Aug 2026 · Published 7 Aug 2026
Limits: Different measurement window from the episode's own series. Said out loud rather than smoothed over.
Research note
0.4%, not 0.6% — the higher figure appeared in a first-pass draft of this packet and was corrected against the source. Note that FactSet's own window is minus-two-to-plus-two days, NOT the next-day close this episode otherwise uses; the episode states the window when it states the number, which is the discipline the whole of S03 is about.
- FactResearchF18
Across more than 230,000 quarterly earnings announcements between 1985 and 2005, roughly 40% produced a share-price move in the opposite direction to the earnings surprise.
Source: Johnson and Zhao, published academic study of contrarian post-announcement returns
Event 31 Dec 2005 · Published 1 Jan 2011
Limits: Sample ends in 2005. The episode does not claim the proportion is the same today.
Research note
Held in every quarter of the sample and was only slightly less common in the most extreme surprises. This is the fact that makes the episode's mechanism general rather than a property of one famous company.
- FactResearchF19
Ray Ball and Lakshmanan Shivakumar found that an average quarterly earnings announcement is associated with only about one to two percent of a year's total information.
Source: Ball and Shivakumar, 'How Much New Information Is There in Earnings?', Journal of Accounting Research
Event 1 Jan 2008 · Published 1 Jan 2008
Limits: An average across a broad sample. Some announcements plainly carry far more.
Research note
Derived from regressing annual returns on the four quarterly announcement-window returns, giving an R-squared of about 5% to 9%. CHOSEN OVER Ball and Brown's better-known 1968 '85 to 90 per cent' figure, which describes the ANNUAL income report in a 1957-65 sample and would be an anachronism applied to a modern quarterly print. The completeness critic flagged the equivocation between the two; this packet carries only the one that is actually about quarterly announcements.
- FactFrontier · Aug 2026JournalismF20
For NVIDIA's most recent quarter, three different data vendors published three different consensus revenue estimates: about $91.8 billion, about $92.2 billion, and about $92.3 billion.
Source: Zacks Investment Research ($91.8bn); LSEG ($92.2bn); FactSet ($92.3bn) — all measured before the 26 August 2026 release
Event 26 Aug 2026 · Published 26 Aug 2026
Limits: Stated to $0.1bn. A fourth and fifth value were found during research and are not narrated, because two of the five could not be re-sourced independently.
Research note
There is NO primary source for a consensus estimate. It is a poll of analysts, sold by vendors, revised until the print. The disagreement is the teaching point, and the episode shows all three rather than choosing one and calling it 'the' expectation.
- FactPrimary sourceF21
NVIDIA has published three different figures for its own non-GAAP earnings per share in one quarter — $1.05, $1.04 and, restated a year later, $1.01 — all of them correct under the definition in force when each was published.
Event 26 Aug 2026 · Published 26 Aug 2026
Limits: None of the three is wrong. That is the point, and the episode says so.
Research note
$1.05 as originally reported; $1.04 excluding a $180 million inventory release, in the same release; $1.01 after the company began including stock-based compensation in non-GAAP from the first quarter of fiscal 2027 and restated the history. Both filings verified. This is the cleanest available demonstration that 'the number' a beat is measured against depends on a definition the company itself can change.
- FactPrimary sourceF22
Between 25 May 2023 and 28 August 2026, NVIDIA's trailing twelve-month earnings per share rose about forty-one times over, its share price rose about five and a half times, and the multiple the market paid for those earnings fell by about 86%.
Source: Computed from NVIDIA's own net income and diluted share counts in SEC filings, against split-adjusted closing prices from two independent vendors
Event 28 Aug 2026 · Published 31 Aug 2026
Limits: Endpoint-sensitive: both dates are chosen — the first session after NVIDIA's first AI-era report, and the most recent completed session. The episode states both dates out loud so the choice is visible. Trailing multiples are computed from filer earnings, not from any vendor's adjusted figure.
Research note
Exact: price 37.98 to 217.55 = 5.73x; trailing EPS 0.192 to 7.906 = 41.15x; trailing multiple about 197.7 to about 27.5, a fall of 86.1%. The identity holds arithmetically: 41.15 x 0.139 = 5.72. THE EPISODE'S CLOSING ARGUMENT. Independently recomputed during fact-check.
- InterpretationResearchF23
The same valuation can be produced by many different futures: at a thirty percent margin and a nine percent cost of capital, a company worth fifteen times its sales is equally consistent with twenty percent growth for twelve years, thirty percent for eight, or ten percent for thirty-eight.
Source: ILLUSTRATIVE MECHANISM. Computed deterministically in src/data/ai03-evidence.ts from a two-stage free-cash-flow model with stated inputs; re-derived at render time.
Event 31 Aug 2026 · Published 31 Aug 2026
Limits: A model, not an observation. Under the perpetuity convention the continuing value is the majority of the total at every horizon shown, which the frame states.
Research note
NOT A VALUATION OF ANY COMPANY, and labelled ILLUSTRATIVE MECHANISM on the frame. Inputs: operating margin 30%, cash tax 15%, incremental investment 20% of the sales increase, cost of capital 9%, perpetuity continuing value. The three solutions land within about 6% of 15.0x. The purpose is to show that a price under-determines the future it implies.
- InterpretationBook / authorF24
Alfred Rappaport and Michael Mauboussin's method reverses conventional analysis: instead of forecasting cash flows to derive a value and comparing it with the price, it starts from the price and reads out the expectations the price already implies.
Source: Rappaport and Mauboussin, Expectations Investing (Harvard Business School Press 2001; revised edition Columbia Business School Publishing 2021), as described by the authors
Event 1 Jan 2021 · Published 7 Sep 2021
Limits: The authors' own step three ends in a buy, sell or hold decision. The episode takes the reading method and explicitly does not take the decision rule, because this course gives no investment advice.
Research note
Attributed idea, not proof. Book details independently confirmed. No sentence from the book is quoted on screen — the completeness critic noted that every book passage in this research reached it through a secondary source, so the episode attributes the idea and quotes nothing.
- Counter-argumentJournalismF25
Price-implied expectations are the output of a model rather than an observation: reading them requires assuming a cost of capital, a forecast horizon and a competitive fade, so two careful analysts can read the same price and extract different expectations.
Source: Derived from the method's own stated input requirements as published by the authors, and from a CFA Institute review identifying the model's treatment of flexibility as its weak point
Event 1 Jan 2021 · Published 1 Jan 2021
Research note
The episode states this limitation in the same scene that teaches the method, rather than in a footnote.
- FactResearchF26
Studies of thousands of large daily stock-market moves across nineteen national markets required an explicit 'unknown, no explanation' category, and separately recorded how confident the journalist was in whatever explanation was offered.
Source: Baker, Bloom, Davis and Sammon, working paper on large daily stock market jumps; methodology retrieved and read directly during fact-check
Event 1 Jan 2022 · Published 1 Jan 2022
Limits: The paper studies market-level jumps, not single-stock moves. The episode's use of it is about the general reliability of same-day explanations, and is worded that way.
Research note
Each jump is classified into seventeen categories, one of which is 'Unknown & No Explanation'. The episode uses this to justify refusing to assign a cause to a single day's move — the discipline is not scepticism for its own sake, it is what the people who study this for a living had to build into their own method.
- FactPrimary sourceF27
NVIDIA's revenue grew from $27.0 billion in fiscal 2023 to $215.9 billion in fiscal 2026 — a compound rate of about 100% a year for three years.
Source: SEC XBRL company facts for CIK 0001045810 (fiscal 2023 $26,974M; fiscal 2026 $215,938M)
Event 25 Jan 2026 · Published 31 Aug 2026
Limits: Fiscal years, not calendar years. NVIDIA's fiscal 2026 ended 25 January 2026.
Research note
Exactly 100.0% nominal; 94.5% after deflating by CPI-U. The episode narrates the nominal figure and uses it for one purpose only: to establish that the company's delivery has been extraordinary, so that the fall in its multiple cannot be read as the business disappointing.
3 further audited series are recorded but not drawn
Fourteen NVIDIA earnings reports, May 2023 to August 2026: reported revenue, the consensus estimate going in, the resulting surprise, and the next-trading-day share-price return.
Source: Revenue: NVIDIA Form 8-K Exhibit 99.1 and SEC XBRL company facts (tier 1). Consensus: Zacks Investment Research (tier 3). Returns: split-adjusted daily closes, Yahoo Finance and StockAnalysis.com, cross-checked against Nasdaq (tier 3).
The arithmetic identity: share price, trailing twelve-month earnings per share, and the resulting multiple, at the first session after each AI-era report.
Source: Net income and diluted share counts from NVIDIA's SEC filings; closing prices split-adjusted from two independent vendors.
ILLUSTRATIVE MECHANISM. Growth-rate and growth-duration pairs that produce the same enterprise value as a multiple of base-year sales.
Source: Computed deterministically from stated inputs; no company data.
The strongest case against this
The one quarter of the fourteen in which NVIDIA guided below the published consensus — May 2025 — and the shares rose 3.3% the next day.
Carried at full strength, before the conclusion — not as a footnote.
The symmetry of the argument has to be shown or the viewer will conclude that beats are simply bad news. Two candidate outside cases were fully researched — Netflix in January 2023, which missed consensus earnings by nearly three quarters and rose, and Oracle in September 2025, which missed on both lines and rose about 36% — and both were cut. Each carried a confound arriving in the same release (a co-chief-executive announcement; a backlog concentrated in one counterparty), and both would have required mixing an earnings surprise with a revenue surprise on one axis, which breaks the comparable-basis rule. NVIDIA's own May 2025 quarter proves the same point on the same company, the same measure and the same chart the episode is already showing.
History, under test
The strongest historical case
What the past licenses — and, stated just as plainly, what it does not.
NVIDIA's fourteen AI-era earnings reports, May 2023 to August 2026
The episode deliberately does not rest on one dramatic quarter. A single beat-and-fall is an anecdote and invites the reader to find a special reason for that day. Fourteen consecutive beats with seven next-day falls is a base rate, computed on one company, one measure and one vendor, and it cannot be explained away by any one quarter's circumstances.
- August 2023: the largest beat of the period, about 21% above consensus, moved the stock one tenth of one percent.
- November 2025 and February 2026: the same 4% beat, four months apart, produced minus 3.2% and minus 5.5%.
- May 2025: guidance BELOW the published consensus, and the stock rose 3.3%.
What remains uncertain
What we still do not know
Stated by the research team, in full, rather than smoothed over.
- Nobody obtained a minute-level after-hours tape for the evening of 26 August 2026. The most vivid version of this episode's argument existed there and could not be sourced (X01).
- The definition of 'the reaction' is a convention, not a measurement. This packet uses the next trading day's close-to-close move throughout, because every one of the fourteen reports was published after the close. FactSet's comparable statistic uses a four-day window and gives a different-looking answer. The episode names the convention on screen rather than hiding behind it.
- There is no primary source for any consensus estimate anywhere in this packet, and five vendors gave five answers for the same quarter.
- Whether the fiscal-2028 growth figure, the margin reset, monetary policy, positioning or something unrecorded moved the price in either direction on 27 or 28 August 2026 is unknown and is stated as unknown.
- The episode teaches a general mechanism largely from one exceptional company. This is answered with two broad-sample facts rather than with a second anecdote, and the limitation is stated on screen.
- Two figures the episode narrates — the fiscal-2028 growth expectation and the fourth-quarter margin trough — exist only in earnings-call transcripts and in no NVIDIA filing.
What we refused to publish
14 claims we would not say — and why
The do-not-narrate list. Some are popular; some are true but unproven; some are simply not this note’s to make. Each refusal is enforced in production, not just recorded.
- UnverifiableX01
“Thirty minutes after the release, with the beat and the above-consensus guidance both public, NVIDIA was down 1.3% in after-hours trading; it was up 4.4% by the time the earnings call ended.”
Verdict: UNVERIFIABLE
Why: This was going to be the episode's opening image and its central proof, and it does not survive its own sourcing standard. It rests on a single live blog that names no quote vendor for the after-hours prices, and no minute-level after-hours tape for 26 August 2026 could be obtained. A claim that load-bearing cannot rest on the thinnest source in the packet — that is the exact inversion the evidence law exists to prevent. The episode was restructured around confirmed closing prices instead. Removing it also removed an implicit causal story, which improved the film.
- UnresolvableX02
“The stock rose because management gave a fiscal-2028 growth number on the call; it fell two days later because of the gross-margin reset.”
Verdict: UNRESOLVABLE — MUST BE STATED AS UNKNOWN
Why: Both are plausible, both were written that week, and neither can be established from price data. Contemporaneous coverage of the 28 August session also offered a hawkish Jackson Hole speech, higher rate-hike odds, profit-taking after the prior day's gain, and mechanical dealer hedging around a large options expiry. The episode reports the sequence and states explicitly that nobody can settle which sentence did it.
- RefutedX03
“The move was already priced in.”
Verdict: REFUTED AS AN EXPLANATION
Why: As used in market commentary the phrase is compatible with a rise, a fall and no move at all, and is almost always invoked after the outcome is known. It is a restatement of the observation, not a cause of it. The episode uses the phrase exactly once, names it as unfalsifiable, and replaces it with something measurable — a named vendor's estimate, with a date.
- RefutedX04
“NVIDIA's 8.7% gain on 27 August 2026 was its largest one-day post-earnings move in years.”
Verdict: REFUTED
Why: It is the third largest of the last thirteen quarters, behind +16.40% on 22 February 2024 and +9.32% on 23 May 2024. The superlative was in the first-pass research and was killed by recomputation of the full split-adjusted series.
- Partly unverifiableX05
“Analysts expected $92.17 billion. / Analysts expected $104.19 billion for the guide.”
Verdict: PARTIALLY UNVERIFIABLE
Why: Five different consensus revenue values for this one quarter were found across the research, from five vendors, and the two numeric series were initially built on two different ones — which would have put contradictory bars in two charts of the same film. Consensus is now stated to one decimal only, always with its vendor named, and the disagreement between vendors is narrated rather than hidden. Five-significant-figure consensus is banned: no vendor publishes a freeze timestamp and the vendors differ by up to half a billion dollars on the same quarter. The Q3 guidance consensus of $104.19bn specifically failed re-sourcing and is not narrated at all.
- UnverifiableX06
“The buy-side bar, or whisper number, was above published consensus; traders needed $110 billion from the guide.”
Verdict: UNVERIFIABLE
Why: Attractive, widely repeated, and unsupported. The desk notes quoted in press coverage could not be retrieved; one of the two figures turned out to describe a bank's own forecast rather than any measured buy-side hurdle. The academic literature is genuinely split on whether a measurable non-sell-side bar exists at all. Treated as folklore and cut, despite being the tidiest possible explanation of the episode's own question.
Show the remaining 8 refused claims
- Partly unverifiableX07
“The options market priced an implied move of about 5.5% (or 5.9%, or 7.0%), and the realised move exceeded it.”
Verdict: PARTIALLY UNVERIFIABLE
Why: Three published estimates, three methods, three measurement dates, and one of them quotes a reference share price that matches no pre-print session — i.e. a live quote refreshed after publication. The 'expected move' is also a vendor convention (85% of the at-the-money straddle) that embeds a risk premium rather than being a forecast. No implied move appears anywhere in this episode.
- UnverifiableX08
“NVIDIA is the largest constituent of the S&P 500 at roughly 7 to 8 percent of the index, the highest single-stock concentration in decades.”
Verdict: UNVERIFIABLE
Why: The cited source contains no weight, no market capitalisation and no such statement. Not re-sourced, so not narrated.
- UnresolvableX09
“NVIDIA has beaten its own guidance for thirteen consecutive quarters.”
Verdict: UNRESOLVABLE — MUST BE STATED AS UNKNOWN
Why: Three different streak counts — thirteen, fourteen and sixteen — were produced by three different agents from the same filings, depending on where the count starts. The underlying quarter-by-quarter beats are all verified; the streak headline is not, so no streak is narrated.
- Partly unverifiableX10
“Post-earnings-announcement drift means prices continue to move in the direction of the surprise for months afterwards.”
Verdict: PARTIALLY UNVERIFIABLE
Why: Three incompatible positions in the literature: the classic finding that drift follows the surprise, a finding that contrarian reactions show little drift, and a finding that drift has been absent for large stocks since about 2006. The episode is about the announcement day and does not need the drift literature; asserting any one of the three would be picking a side the evidence does not settle.
- RefutedX11
“Buy, sell, hold, accumulate, avoid, overvalued, undervalued, cheap, expensive, a good entry point, one to watch, due for a pullback, or any statement that NVIDIA's price is or is not justified.”
Verdict: REFUTED BY POLICY
Why: This course teaches how to read expectations; it does not advise on securities. The distinction in US law between impersonal financial publication and personalised advice is exactly the line this episode stays behind, and a script of this shape drifts across it through a small set of recurring sentence patterns rather than through one obvious sentence. Banned by pattern in the script's excludedClaimTerms.
- RefutedX12
“The market was right / the market was wrong about this quarter.”
Verdict: REFUTED BY POLICY
Why: Both readings are available and both are cheap. The episode's whole discipline is that it does not need a view on whether the price is correct, and taking one would smuggle in Episode 12's job.
- UnverifiableX13
“NVIDIA's closing price on 31 August 2026.”
Verdict: UNVERIFIABLE
Why: Four different values were returned by different agents for the same session, three of which claimed two-vendor agreement, while the US market was still open at the time of research. The episode is dated 31 August and reports through the 28 August close.
- RefutedX14
“Where a dollar of AI spending travels, which layer of the value chain has pricing power, why memory prices are rising, and whether any of this is a bubble.”
Verdict: REFUTED BY SCOPE
Why: Episode 4 owns the value chain, Episode 5 owns memory and the bottleneck, Episode 12 owns the capital cycle. Memory appears here as one clause inside a guidance sentence and is not explained. Data Center revenue appears as a single total and is never decomposed.
What to watch next
Dated material, and what would make it stale
NVIDIA's guidance of $108.0 billion for the quarter ending October 2026, plus or minus 2%, explicitly assuming no Data Center compute revenue from China.
26 Aug 2026Status: current
What changes: The whole frontier scene is replaced when the next quarter is reported.
Invalidated by: NVIDIA's Q3 fiscal 2027 results, expected in November 2026. Also invalidated early by any change in US or Chinese policy that puts China Data Center revenue back into the outlook.
Gross margin guided to 74.0% for the coming quarter, down from the 75.0% just reported.
26 Aug 2026Status: current
What changes: Superseded by the next quarter's reported margin and the next guide.
Invalidated by: Q3 fiscal 2027 results, November 2026.
A preliminary expectation, given on the earnings call, for fiscal 2028 revenue to grow approximately 70% — the first full-year figure the company has given in advance, and from the moment it was said, a number with an expectation attached to it.
26 Aug 2026Status: current, transcript-sourced only — absent from every NVIDIA filing
What changes: This is the episode's own live experiment. The next time NVIDIA reports, this number will have a consensus around it, and the thing that had no bar will have one.
Invalidated by: Q3 fiscal 2027 results, November 2026; or any revision of the figure by the company.
Ideas we borrowed, and tested
Thinkers, taken seriously enough to argue with
Claim → author → evidence → counter-argument → historical test → current relevance. Never doctrine.
Alfred Rappaport and Michael Mauboussin
Expectations Investing (2001; revised edition 2021)
Read the expectations embedded in a share price before judging anything about the company: start from the price and solve backwards for the performance it implies.
- Evidence:
- The method's inputs are explicit and public — sales growth, operating margin, incremental investment, cash tax rate, cost of capital, and the length of the forecast period — which is what makes it checkable rather than rhetorical.
- Counter-argument:
- The output is a model result, not an observation. It requires a cost of capital and a horizon that nobody can observe, so two careful analysts can read the same price and get different answers; and a discounted-cash-flow frame can understate the value of flexibility for a business facing wide uncertainty.
- Historical test:
- Applied here to a real series: NVIDIA's trailing multiple fell about 86% over three years while its earnings rose about forty-one times. The expectations embedded in the 2023 price were, in the event, largely met — and the multiple came down anyway.
- Current relevance:
- It converts an unfalsifiable phrase — 'it was priced in' — into an arithmetic statement that can be written down, dated, and later found to be wrong.
Ray Ball and Lakshmanan Shivakumar
How Much New Information Is There in Earnings? (Journal of Accounting Research, 2008)
The quarterly earnings announcement is a small part of the year's information, not the main event.
- Evidence:
- Regressing annual returns on the four quarterly announcement windows yields an R-squared of roughly 5% to 9% — about one to two percent of a year's information per announcement.
- Counter-argument:
- An average across a broad sample; individual announcements plainly carry far more, and this one moved a company's value by tens of billions of dollars in a session.
- Historical test:
- Consistent with NVIDIA's own record: the largest beat of the AI era moved the stock a tenth of a percent.
- Current relevance:
- Explains, without any appeal to irrationality, how a spectacular report can be nearly uninformative.
Evidence & sources
24 sources, by tier
Tier 1 is primary and authoritative — filings, regulators, official statistics. Journalism and books are attributed ingredients, never proof by reputation.
- Primary sourceNVIDIA Corporation Form 8-K (Item 2.02), SEC EDGAR, accession 0001045810-26-000073 (opens sec.gov)supports F01
- Primary sourceNVIDIA Q2 FY2027 press release, Exhibit 99.1 to Form 8-K (opens sec.gov)supports F02, F04, F05
- Primary sourceCFO Commentary on Second Quarter Fiscal 2027 Results, Exhibit 99.2 to Form 8-K (opens sec.gov)supports F03
- Primary sourceNVIDIA Q2 FY2027 press release, Outlook section, Exhibit 99.1 to Form 8-K (opens sec.gov)supports F06, F07
- JournalismNVIDIA Q2 FY2027 earnings call transcript (statement attributed to CFO Colette Kress), corroborated across two independent transcript publisherssupports F08
- JournalismNVIDIA Q2 FY2027 earnings call transcript (statements attributed to CFO Colette Kress and CEO Jensen Huang), corroborated across two independent transcript publisherssupports F09
- JournalismDaily closing prices, split-adjusted; Yahoo Finance chart API and StockAnalysis.com returning identical values, cross-checked against Nasdaqsupports F10
- JournalismIndex closes for 28 August 2026 as reported contemporaneously; Nasdaq Composite 26,402.42, −0.52%supports F11
- Primary sourceRevenue from NVIDIA Form 8-K Exhibit 99.1 releases and SEC XBRL company facts (tier 1); consensus from Zacks Investment Research (tier 3); next-day returns computed from split-adjusted closes at two independent vendorssupports F12
- Primary sourceNVIDIA Form 8-K Exhibit 99.1 (revenue $13,507M, quarter ended 30 July 2023, reported 23 August 2023); Zacks consensus $11.2bn; next-day return computed from split-adjusted closes (opens sec.gov)supports F13
- Primary sourceNVIDIA Form 8-K Exhibit 99.1 releases (revenue $57,006M and $68,127M); Zacks consensus $54.7bn and $65.4bn (both +4.14%); next-day returns computed from split-adjusted closessupports F14
- Primary sourceNVIDIA Q1 FY2026 press release, Outlook section (guide $45.0bn); LSEG consensus $45.9bn as reported contemporaneously; next-day return computed from split-adjusted closessupports F15
- JournalismFactSet Earnings Insight, John Butters, 7 August 2026supports F16
- JournalismFactSet Earnings Insight, John Butters, 7 August 2026 (window: two days before through two days after the release)supports F17
- ResearchJohnson and Zhao, published academic study of contrarian post-announcement returnssupports F18
- ResearchBall and Shivakumar, 'How Much New Information Is There in Earnings?', Journal of Accounting Researchsupports F19
- JournalismZacks Investment Research ($91.8bn); LSEG ($92.2bn); FactSet ($92.3bn) — all measured before the 26 August 2026 releasesupports F20
- Primary sourceNVIDIA Q2 FY2026 press release (2025-08-27) and Q2 FY2027 press release (2026-08-26), both Exhibit 99.1 to Form 8-K (opens sec.gov)supports F21
- Primary sourceComputed from NVIDIA's own net income and diluted share counts in SEC filings, against split-adjusted closing prices from two independent vendorssupports F22
- ResearchILLUSTRATIVE MECHANISM. Computed deterministically in src/data/ai03-evidence.ts from a two-stage free-cash-flow model with stated inputs; re-derived at render time.supports F23
- Book / authorRappaport and Mauboussin, Expectations Investing (Harvard Business School Press 2001; revised edition Columbia Business School Publishing 2021), as described by the authorssupports F24
- JournalismDerived from the method's own stated input requirements as published by the authors, and from a CFA Institute review identifying the model's treatment of flexibility as its weak pointsupports F25
- ResearchBaker, Bloom, Davis and Sammon, working paper on large daily stock market jumps; methodology retrieved and read directly during fact-checksupports F26
- Primary sourceSEC XBRL company facts for CIK 0001045810 (fiscal 2023 $26,974M; fiscal 2026 $215,938M)supports F27
Complete evidence depth
The research desk
Everything the research evaluated before it was distilled — including what it rejected, and why.
This note predates DowJo’s research-preservation standard, so the pre-collapse shortlist was not kept. Everything the film was allowed to say — and refused to say — is above.
Community
Talk it through
Discuss this research
Members only · in 📡 Intelligence Watch
Challenge the argument, bring evidence, or ask what would change our mind. This note is discussed in 📡 Intelligence Watch, a DowJo room where members compare reads. Filings, news events, and what they actually mean — the discussion room for the intelligence feed.
Read and reply, then keep following the story — you come straight back to this note.
Education-first ground rules apply in every room: mechanisms and evidence, never calls.
Now train it
Take it further
Reading is where judgment starts. Practice is where DowJo measures it — and remembers what to train next.
Train it
Can you spot what the market already priced in?
Five reps on today's tape. Commit a call before the market grades it.
Train itTrain it
Think you can spot it before the reveal?
Train it on a chart: an unseen setup, revealed bar by bar, graded on your process.
Sign in to train this setupLearn the concepts underneath
Join the discussion
Challenge the argument, bring evidence, or ask what would change our mind — in the open, with other members.
MentorAsk Jo what could invalidate this thesis.
Jo reads this note before answering. Sign in to ask, with this evidence attached — you come straight back here.
Ask Jo with this evidence attachedProvenance — where this note comes from
This page is a deterministic projection of canonical research artifacts. It adds presentation and discovery; it never adds, removes or softens a finding. Question taken from the title; thesis from the script. Projected 21 Sep 2026 by dowjo-research-projector v1.1.0 from origin commit c83c8797ab79.
| Artifact | Path (origin-relative) | Identity | Version |
|---|---|---|---|
| Episode registry | course/episodes.json | sha256 40022b5ffeaa… | — |
| Evidence packet | course/research/AI03-evidence-packet.json | sha256 2c6aa4903434… | compiled 2026-08-31 · rev 1 |
| Approved script | course/scripts/AI03-results-are-not-expectations.json | sha256 47e95aaaa6be… | v 1 |
| Episode manifest | course/manifest/AI03.manifest.json | sha256 8020a521a748… | — |
| Approval record | renders/approved/AI03-FINAL.json | sha256 e3d9e9347f4e… | compiled 2026-08-31 · v 4f6c9bb |
| Poster still | qa/stills-AI03/S00-055.png | sha256 e6226fee0f41… | — |
Approved master (AI03-FINAL-elevenlabs.mp4): sha256 9993a560484295e6921d7e4d73ec9c9266cf43ddcd39e936eb156c5297106d7c
Commercial disclosure: none. No affiliate relationship, review copy, sponsorship or publisher relationship applies to this note.
For education only. Not financial advice. No buy, sell or hold recommendations, ever.
Related research
That looks interesting too
Investing in the AI Era · Episode 2
The AI Trade That Made Money — and Taught the Wrong Lesson
A profitable trade is not evidence that the reasoning was good, and a loss is not evidence that it was bad.
Shares Behavioral Finance, Artificial Intelligence, Markets, Semiconductors, Valuation & Expectations
Investing in the AI Era · Episode 12
The AI Bubble Everyone Knows Is Coming
A bubble is not a feeling everyone shares.
Shares Artificial Intelligence, Valuation & Expectations, Filings & Disclosure, Markets
Investing in the AI Era · Episode 14
The Market Becomes Your Curriculum
The thesis of the theses: under every film sat the same two questions, asked until they no longer needed the subject — compared with what, and as of when?
Shares Artificial Intelligence, Markets, Valuation & Expectations