Episode 2 · Investing in the AI Era · 9 min film

You may have seen the film. This is the research behind it — every claim, every refusal, kept inspectable.

From DowJo — where this research becomes practice. Train your judgment before you risk your money.

Course Research · Episode 2

The AI Trade That Made Money — and Taught the Wrong Lesson

Which real AI-era investment best teaches that a profitable outcome does not prove good reasoning, and that a bad outcome does not prove bad reasoning?

The evidence-led takeaway

A profitable trade is not evidence that the reasoning was good, and a loss is not evidence that it was bad. Outcome and process are two different measurements, and only one of them repeats.

Make five honest calls on today's market

Ask Jo about this research

Jo reads this note before answering. Sign in to ask Jo with this evidence attached — you come straight back here.

Researched 28 Aug 2026Updated 31 Aug 2026Current22 pieces of evidence13 claims refused20 sources

Behavioral Finance Artificial Intelligence Markets Semiconductors Valuation & Expectations

In brief

What this research concludes

A profitable trade is not evidence that the reasoning was good, and a loss is not evidence that it was bad. Outcome and process are two different measurements, and only one of them repeats.

The judgment skill this hands over

Reconstructing a decision from what was knowable at the time, before letting the result grade it — and writing the record down before the outcome exists.

What we investigated

Which real AI-era investment best teaches that a profitable outcome does not prove good reasoning, and that a bad outcome does not prove bad reasoning?

This is episode 2 of Investing in the AI Era, a 14-episode course. The film tells the story; this note is the research behind it, kept inspectable.

Who and what this looks at

Super Micro ComputerNVIDIAAnnie DukeThinking in BetsSecurities and Exchange CommissionSEC XBRL company facts

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

Six lenses run against primary sources, then adversarial refutation of every candidate claim before it was allowed into narration. Lenses: (1) CASE SELECTION — screen every large AI-era single-name outcome 2023-2026 for the one that best separates decision quality from outcome quality; (2) POINT-IN-TIME RECONSTRUCTION — establish what was publicly knowable at each decision date and nothing later, using public_available_at as the guard; (3) FUNDAMENTAL FORENSICS — rebuild the company's revenue, gross margin, inventory and share-count history from SEC XBRL company-facts rather than from any aggregator; (4) MARKET RECORD — read the split-adjusted end-of-day price record and pin every event date to its close (price levels are described qualitatively in this note); (5) CONTESTED RECORD — where professionals with better access than any outside investor reached OPPOSITE conclusions, record both and narrate neither as settled; (6) FRONTIER — the state of the AI-infrastructure trade as of the compile date, sourced only to company releases and price data. Refutation notes: a commercial data aggregator returned duplicated quarters for the delayed-filing period and would have put four wrong numbers on screen; every fundamental in this packet is therefore taken from SEC XBRL company facts or the company's own results release, and no aggregator figure is reproduced here. A widely repeated 2026 secondary article asserting the company regained Nasdaq compliance in January 2026 was refuted against the filing index (the catch-up filings and the compliance 8-K are dated February 25-26, 2025) and is excluded.

The refutation pass, in numbers

8 proposed6 confirmed0 corrected0 partially unverifiable0 unverifiable2 refuted

Every verdict, claim by claim · 8

CONFIRMED — and the commercial aggregator's version REFUTED: it duplicated the December-2023 quarter into June-2024 and the March-2024 quarter into September-2024. Four numbers would have been wrong on screen.
SMCI quarterly revenue and gross margin, all quarters 2022-2026
SEC XBRL company facts, latest-filed version per period
CONFIRMED — the March 13, 2024 close is the record close; an intraday record printed five days earlier, on March 8, 2024, and is deliberately not used.
March 13, 2024 is the record closing high
Split-adjusted daily price series 2021-2026 (levels not reproduced in this note)
REFUTED — catch-up filings 2025-02-25, compliance 8-K 2025-02-26. The secondary source asserting January 2026 is unreliable.
The company regained Nasdaq compliance in January 2026
EDGAR filing index for CIK 0001375365
CONFIRMED — $17.5 million penalty; former CFO separately penalised; CEO not charged, $2.1 million clawback.
SEC settlement terms, August 2020
SEC Press Release 2020-190, read in full
CONFIRMED verbatim.
Ernst & Young's stated reason for resigning
Form 8-K filed 2024-10-30, Item 4.01, quoting the resignation letter
CONFIRMED verbatim — including the finding that the auditor's conclusions were not supported by the facts the committee uncovered. The record is genuinely contested and the episode presents it as such.
The special committee's findings
Form 8-K filed 2024-12-02, Items 5.02 and 8.01
CONFIRMED — $96.2bn revenue, $89.0bn data centre, 75.0% gross margin, $108.0bn outlook. A secondary source reporting non-GAAP EPS of $2.22 was not used; the GAAP figure of $2.46 comes from the company.
NVIDIA Q2 fiscal 2027 figures
NVIDIA's own results release, August 26, 2026
REFUTED as narratable — the aggregators disagree by more than $100bn and the XBRL series are reported on inconsistent bases across the group. Excluded (X08).
Aggregate hyperscaler 2026 capex figures circulating in secondary media
Attempted reconciliation against SEC XBRL capital-expenditure facts for the four largest spenders

What was screened before the case was chosen · 7 candidates

Super Micro Computer (SMCI)
SELECTED — the only candidate where the theme kept being right, the revenue delivered more than the thesis required, and the late shareholder still suffered a deep loss. That dissociation is the episode.
A many-fold rise from January 2023 to the March 2024 record, then a deep drawdown that persisted to August 2026 · Revenue over the same span: FY2023 $7.1B -> FY2026 $39.1B
NVIDIA (NVDA)
REJECTED — process and outcome largely agreed, so it cannot separate them; and Episode 3 owns NVIDIA earnings and expectations.
A many-fold rise to a 2026 record; only a shallow drawdown from it
C3.ai (AI)
REJECTED — collapses into 'do not buy a weak company', which is a different and lesser lesson.
A large 2023 rise, then a deep drawdown
Arm Holdings (ARM)
REJECTED — short listed history, and the interesting part is narrative-versus-royalty economics, not decision quality.
A substantial drawdown from its record
Palantir (PLTR)
REJECTED — the honest question there is valuation and expectations, which belongs to Episodes 3 and 12.
A many-fold rise to a 2025 record; only a shallow drawdown from it
Micron (MU)
REJECTED — memory and HBM is Episode 5's subject; consuming it here would cost the season a whole episode.
A many-fold rise to a 2026 record
Vertiv (VRT)
REJECTED — power and thermal is Episode 6's subject.
A many-fold rise to a 2026 record

What the evidence says

22 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.

FactPrimary sourceF01
At the start of January 2023, Super Micro Computer's shares traded at a small fraction of the level they would reach fourteen months later (prices compared on a basis adjusted for the company's later 10-for-1 stock split).
Source: Publicly reported end-of-day market prices (split-adjusted). Specific price levels and percentage moves are not reproduced in this note. Split ratio and effective date 2024-10-01 confirmed against the company's own corporate actions.
Event 3 Jan 2023 · Published 3 Jan 2023
Limits: Split-adjusted. The price a January 2023 buyer actually typed was ten times the split-adjusted figure.
FactPrimary sourceF02
Super Micro's shares set their highest closing price on record on March 13, 2024, after rising many times over from the start of 2023.
Source: Publicly reported end-of-day market prices (split-adjusted). Specific price levels and percentage moves are not reproduced in this note.
Event 13 Mar 2024 · Published 13 Mar 2024
Research note
An intraday high printed a few days earlier, on 2024-03-08; the episode uses closes only, consistently, because mixing closes and intraday highs inflates a return for free.
FactPrimary sourceF03
By November 14, 2024 Super Micro's shares had given back most of that rise, closing far below their March 13, 2024 record.
Source: Publicly reported end-of-day market prices (split-adjusted). Specific price levels and percentage moves are not reproduced in this note.
Event 14 Nov 2024 · Published 14 Nov 2024
FactFrontier · Aug 2026Primary sourceF04
At the end of August 2026 Super Micro's shares were still far below their March 13, 2024 record close.
Source: Publicly reported end-of-day market prices (split-adjusted). Specific price levels and percentage moves are not reproduced in this note.
Event 28 Aug 2026 · Published 28 Aug 2026
Research note
FALSIFIER: the next trading day. Any refresh of this episode must re-read the price record behind this statement.
FactPrimary sourceF05
Super Micro's net sales were $7,123 million in fiscal 2023, $14,989 million in fiscal 2024 and $21,972 million in fiscal 2025 (fiscal years end June 30).
Source: Super Micro Computer, Forms 10-K, via SEC XBRL company facts (CIK 0001375365) (opens data.sec.gov)
Event 30 Jun 2023 · Published 28 Aug 2025
Research note
Taken from the most recently filed version of each period, so the fiscal 2024 figures are the audited ones filed 2025-02-25, not the pre-delay presentation.
FactFrontier · Aug 2026Primary sourceF06
Super Micro reported net sales of $39.1 billion for fiscal 2026 and gross margin of 10.8%, against $22.0 billion and 11.1% in fiscal 2025.
Source: Super Micro Computer, 'Supermicro Announces Fourth Quarter and Full Fiscal Year 2026 Financial Results', Exhibit 99.1 to Form 8-K (opens sec.gov)
Event 30 Jun 2026 · Published 11 Aug 2026
Limits: The company describes these results as unaudited. The fiscal 2026 Form 10-K had not been filed as of August 28, 2026.
Research note
FALSIFIER: the fiscal 2026 Form 10-K, expected within weeks of the compile date.
FactPrimary sourceF07
Super Micro's gross margin fell for three consecutive fiscal years as revenue rose: 18.0% in fiscal 2023, 13.8% in fiscal 2024, 11.1% in fiscal 2025, 10.8% in fiscal 2026.
Source: SEC XBRL company facts (fiscal 2023-2025) and the company's fiscal 2026 results release
Event 30 Jun 2023 · Published 11 Aug 2026
Research note
Gross profit divided by net sales, computed from the filed figures; fiscal 2023 18.01%, fiscal 2024 13.75%, fiscal 2025 11.06%, fiscal 2026 10.8% as stated by the company.
FactPrimary sourceF08
In the quarter ended December 31, 2025, Super Micro reported net sales of $12,682 million — more than double the same quarter a year earlier — at a gross margin of 6.3%, the lowest quarterly gross margin in the series.
Source: Super Micro Computer, Form 10-Q for the quarter ended December 31, 2025, via SEC XBRL (opens sec.gov)
Event 31 Dec 2025 · Published 6 Feb 2026

Show the remaining 14 pieces of evidence

FactFrontier · Aug 2026Primary sourceF09
In the quarter ended June 30, 2026, Super Micro reported net sales of $11.1 billion, gross margin of 17.5% and net income of $1,178 million, and guided fiscal 2027 net sales to a range of $65.0 billion to $72.0 billion.
Source: Super Micro Computer, fiscal 2026 fourth-quarter results release (opens sec.gov)
Event 30 Jun 2026 · Published 11 Aug 2026
Limits: Unaudited. Guidance is a forward-looking statement by the company, not a fact about the future.
Research note
FALSIFIER: fiscal 2027 first-quarter results, expected around early November 2026.
FactPrimary sourceF10
Super Micro's inventory rose from $1,446 million at June 30, 2023 to $11,103 million at March 31, 2026.
Source: SEC XBRL company facts, InventoryNet (CIK 0001375365)
Event 30 Jun 2023 · Published 11 May 2026
Limits: Rising inventory in a business whose revenue also rose sharply is not by itself evidence of anything; it is presented as scale, not as a verdict.
FactPrimary sourceF11
On August 25, 2020 the Securities and Exchange Commission charged Super Micro Computer and its former chief financial officer with prematurely recognising revenue and understating expenses over a period of at least three years. Without admitting or denying the findings, the company agreed to a cease-and-desist order and a $17.5 million penalty. The chief executive was not charged with misconduct and reimbursed the company $2.1 million in stock profits under the Sarbanes-Oxley clawback provision.
Source: SEC Press Release 2020-190, 'SEC Charges Super Micro and Former CFO in Connection with Widespread Accounting Violations' (opens sec.gov)
Event 25 Aug 2020 · Published 25 Aug 2020
Research note
This is the load-bearing point-in-time fact of the episode: it was on the public record more than two years BEFORE the January 2023 decision point. It is narrated as something that was available, never as something that predicted what followed.
FactPrimary sourceF12
Super Micro's shares were removed from Nasdaq in 2018 after the company failed to file its financial statements on time, traded over the counter from March 22, 2019, and were relisted on Nasdaq in January 2020.
Source: Super Micro Computer, Form 8-K dated January 9, 2020 and Form 10-K for fiscal 2020 (opens sec.gov)
Event Aug 2018 · Published 9 Jan 2020
FactPrimary sourceF13
At the start of January 2023 the company had reported $6,645 million of revenue and $578 million of net income over the previous four quarters, and the market valued it at a modest multiple of those trailing earnings.
Source: SEC XBRL company facts for the quarters ended March, June, September and December 2022; share count from the Form 10-Q cover page dated January 31, 2023; market value from publicly reported prices
Event 3 Jan 2023 · Published 3 Feb 2023
Limits: The December 2022 quarter was reported on February 2, 2023, so a buyer on January 3 saw the three quarters to September 2022 plus the company's guidance. The multiple is therefore approximate and is stated qualitatively. Market value uses the share count disclosed four weeks later.
FactPrimary sourceF14
On August 27, 2024 a short seller published a report making allegations about the company's accounting, related-party transactions and export controls. The company said the report contained false or inaccurate statements. The shares fell that day.
Source: Super Micro Computer, Form 10-Q for the quarter ended March 31, 2026, risk factors (the company's own description of the report); price move as publicly reported (opens sec.gov)
Event 27 Aug 2024 · Published 11 May 2026 · Publicly knowable 27 Aug 2024
Research note
The short seller is deliberately not named on screen; the company's own filed description is the source, and the episode makes no finding about the report's accuracy.
FactPrimary sourceF15
On August 28, 2024 the company announced it would not file its annual report for fiscal 2024 on time. The shares fell sharply that day — the largest single-day fall of that week.
Source: Super Micro Computer, Form 8-K filed 2024-08-28 and Form NT 10-K filed 2024-08-30; price move as publicly reported (opens sec.gov)
Event 28 Aug 2024 · Published 28 Aug 2024
FactPrimary sourceF16
On October 24, 2024 Ernst & Young resigned as the company's auditor. In its resignation letter, disclosed by the company on October 30, 2024, the firm said it was resigning because of information that had come to its attention which had led it to be 'unwilling to be associated with the financial statements prepared by management'. The shares fell sharply on October 31, 2024.
Source: Super Micro Computer, Form 8-K filed October 30, 2024, Item 4.01 (opens sec.gov)
Event 24 Oct 2024 · Published 30 Oct 2024
FactPrimary sourceF17
On December 2, 2024 an independent special committee of the board completed a three-month review conducted with outside counsel and a forensic accounting firm. Among its findings: the evidence reviewed did not give rise to any substantial concerns about the integrity of the company's senior management or audit committee; and the committee did not believe the auditor's resignation or the conclusions the auditor reached were supported by the facts it uncovered. The board adopted all of the committee's recommendations, including recruiting a new chief financial officer.
Source: Super Micro Computer, Form 8-K filed December 2, 2024, Items 5.02 and 8.01 (opens sec.gov)
Event 2 Dec 2024 · Published 2 Dec 2024
Research note
This fact and F16 are deliberately narrated together and left unreconciled. Two sets of professionals with far better access than any outside investor reached opposite conclusions about the same company in the same quarter.
FactPrimary sourceF18
On February 25, 2025 the company filed its delayed fiscal 2024 annual report together with two overdue quarterly reports, and announced the following day that it had regained compliance with Nasdaq's listing rules. The shares rose sharply on February 26, 2025.
Source: Super Micro Computer, Forms 10-K and 10-Q filed 2025-02-25 and Form 8-K filed 2025-02-26 (Item 3.01) (opens sec.gov)
Event 25 Feb 2025 · Published 26 Feb 2025
Research note
Refutes a widely repeated secondary claim that compliance was regained in January 2026; the filing index settles it.
FactFrontier · Aug 2026Primary sourceF19
As of March 31, 2026 the company's management concluded that its disclosure controls and procedures were not effective, because of material weaknesses in internal control over financial reporting first identified in the fiscal 2025 annual report. Management stated that the financial statements nonetheless fairly present the company's position in all material respects.
Source: Super Micro Computer, Form 10-Q for the quarter ended March 31, 2026, Item 4 (opens sec.gov)
Event 31 Mar 2026 · Published 11 May 2026
Research note
FALSIFIER: the fiscal 2026 Form 10-K, which will state whether the material weaknesses are remediated.
FactFrontier · Aug 2026Primary sourceF20
The company has disclosed that on March 19, 2026 a federal indictment was unsealed against three individuals employed by or associated with the company at the time, in connection with an alleged conspiracy to commit export-control violations. The company is not a defendant, says it is cooperating with the government's investigation, and states that the three individuals are no longer employed by or associated with it.
Source: Super Micro Computer, Form 10-Q for the quarter ended March 31, 2026, risk factors — the company's own filed description (opens sec.gov)
Event 19 Mar 2026 · Published 11 May 2026 · Publicly knowable 19 Mar 2026
Limits: An indictment is an allegation. No individual is named on screen, no allegation is characterised beyond the company's own filed language, and the episode draws no conclusion from it. It is included only because an honest 2026 account of this company cannot say the file is closed.
Research note
FALSIFIER: resolution of the proceedings.
FactFrontier · Aug 2026Primary sourceF21
On August 26, 2026 NVIDIA reported quarterly revenue of $96.2 billion, up 106% from a year earlier, with data-centre revenue of $89.0 billion, up 117%, gross margin of 75.0%, and guided the following quarter to $108.0 billion plus or minus two percent.
Source: NVIDIA, 'NVIDIA Announces Financial Results for Second Quarter Fiscal 2027' (opens nvidianews.nvidia.com)
Event 26 Jul 2026 · Published 26 Aug 2026
Research note
FALSIFIER: NVIDIA's next quarterly report, expected around late November 2026.
FactFrontier · Aug 2026Primary sourceF22
Measured on closing prices at the end of August 2026, several of the AI-infrastructure winners of the last three years — NVIDIA, AMD, Broadcom, Micron, Vertiv, Arm and Super Micro — were trading below their own record closes, by very different margins; Super Micro was the furthest below.
Source: Publicly reported end-of-day market prices, each name against its own record close since December 2022. Specific levels and percentages are not reproduced in this note.
Event 28 Aug 2026 · Published 28 Aug 2026
Limits: A drawdown from a record close is a statement about price, not about business quality, and says nothing about what happens next.
Research note
FALSIFIER: the next trading day. This item is the episode's designated refreshable block.

5 further audited series are recorded but not drawn

Super Micro weekly closing price, split-adjusted, June 2022 to August 28, 2026, with the January 3 2023, March 13 2024, November 14 2024 and August 28 2026 closes pinned as exact points
Source: Publicly reported end-of-day market prices (split-adjusted). Specific price levels and percentage moves are not reproduced in this note. · Audit: Series maximum must fall on the 2024-03-13 close; the series must include the 2024-11-14 close; the final point must be the 2026-08-28 close.
Net sales by fiscal year, 2019-2026, USD millions
Source: SEC XBRL company facts for fiscal 2019-2025; company results release for fiscal 2026 (unaudited) · Audit: Must rise in every year from 2022 onward and end above five times the fiscal 2023 value.
Gross profit as a percentage of net sales by fiscal year, 2019-2026
Source: Same as smci_revenue_fy · Audit: Must fall in each of fiscal 2024, 2025 and 2026 from the fiscal 2023 peak of 18.0%.
Quarterly gross margin, September 2022 to June 2026
Source: SEC XBRL company facts; June-quarter values derived as the fiscal-year residual and cross-checked against the company's own stated figures (9.5% for the June 2025 quarter, 17.5% for June 2026) · Audit: Minimum must be the December 2025 quarter at 6.3%; final point must be 17.5%.
Percentage below each name's own record close, at 2026-08-28
Source: Publicly reported end-of-day market prices, each name against its own record close since December 2022. Specific levels and percentages are not reproduced in this note. · Audit: Every value must be negative; Super Micro must be the largest drawdown in the set.

The strongest case against this

The company after the collapse

Carried at full strength, before the conclusion — not as a footnote.

The episode would be dishonest if it stopped at the drawdown, because that would simply replace one outcome-based verdict with another. Since the low, the company has filed its reports, regained compliance, roughly doubled revenue again, restored quarterly gross margin to 17.5% in the June 2026 quarter, and guided fiscal 2027 to $65-72 billion. It also still discloses material weaknesses in internal control and still has unresolved proceedings. Nobody's file is closed — including the sceptic's.

History, under test

The strongest historical case

What the past licenses — and, stated just as plainly, what it does not.

Super Micro Computer, January 2023 to August 2026

The theme was correct, the thesis was correct, and the company delivered more than the thesis required — net sales rose from $7.1 billion to $39.1 billion in three fiscal years — and an investor who bought at the moment the thesis looked most obviously validated is still deep in loss. Meanwhile an investor who bought the same company fourteen months earlier, on identical reasoning, could have multiplied their money many times over. One process, two verdicts. That is the cleanest available proof that an outcome does not grade a decision.

The mechanism: Gross margin. The business sells assembled systems whose most expensive components it buys from others. As the mix shifted toward very large, very price-sensitive orders, revenue scaled and the share of each dollar the company kept did not: 18.0% of net sales in fiscal 2023, 10.8% in fiscal 2026, and 6.3% in a single quarter that was also the largest the company had ever reported.

The symmetry: The same case populates all four boxes of the decision/outcome matrix using real prices and real dates: the January 2023 buyer who did the work (good decision, good outcome), the January 2023 buyer who bought a slogan (bad decision, good outcome), the March 2024 buyer with that identical slogan (bad decision, bad outcome), and the investor who declined in January 2023 on the strength of the public record (defensible decision, bad outcome). The outcome column cannot tell the first two apart.

What remains uncertain

What we still do not know

Stated by the research team, in full, rather than smoothed over.

Whether the recovery in quarterly gross margin to 17.5% in the June 2026 quarter is a change in the economics of the business or a mix effect in one quarter. The company attributes it to customer mix and product architecture; one quarter cannot settle it.
Whether the material weaknesses in internal control disclosed through March 31, 2026 are remediated in the fiscal 2026 annual report.
How the proceedings disclosed by the company in 2026 are resolved.
Whether the drawdown across AI-infrastructure equities through August 2026 is a repricing of expectations, a pause in a continuing build-out, or the start of something larger. The episode explicitly refuses to answer this and hands the question to later episodes.
Whether the January 2023 buyer's reasoning was good is genuinely unknowable from outside. The episode's claim is narrower and defensible: the OUTCOME does not answer it.

What we refused to publish

13 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.

5 Refuted2 Unresolvable4 Unverifiable2 Partly unverifiable

RefutedX01
“The short-seller report caused the collapse in the share price.”
Verdict: REFUTED as a causal claim
Why: The shares fell modestly on the day the report published and sharply the next day, when the company announced its own filing delay. The dates are narratable; the causation is not. Naming the publisher is also avoided: the company's filed description does not name it, and the episode makes no finding about the report.
RefutedX02
“Super Micro committed accounting misconduct in 2024, or the 2024 events proved the allegations.”
Verdict: REFUTED
Why: The special committee found no substantial concerns about management integrity; the company filed its reports and regained compliance; no such finding exists. The word for this allegation and its synonyms are barred from the episode.
RefutedX03
“The 2020 SEC settlement predicted the 2024 filing crisis, or an investor who read it would have avoided the loss.”
Verdict: REFUTED
Why: This is the exact hindsight error the episode teaches against. A base rate is not a forecast. The settlement is narratable as something that was AVAILABLE and unconsidered; it is never narratable as a warning that came true.
UnresolvableX04
“Characterisations of the 2026 indictment's allegations beyond the company's own filed description, or naming any individual.”
Verdict: UNRESOLVABLE — MUST BE STATED AS UNKNOWN
Why: Live criminal proceedings against individuals who are entitled to the presumption of innocence, and immaterial to a lesson about decision process. The company's filed language is the ceiling of what may be said.
UnverifiableX05
“Inclusion in the S&P 500 caused the March 2024 peak.”
Verdict: UNVERIFIABLE
Why: Index inclusion and the peak are close in time. Attributing the price to the flow is an interpretation with no test available inside this episode.
UnverifiableX06
“Gross margin fell because of any specific named customer, contract or product mix decision.”
Verdict: UNVERIFIABLE
Why: The company does not disclose margin attribution at that granularity. The margin series is narratable; the cause is not.

Show the remaining 7 refused claims

UnverifiableX07
“Any figure for how many investors held the stock, what retail investors earned or lost on it, or how large a typical position was.”
Verdict: UNVERIFIABLE
Why: No source exists. The two investors in this episode are explicitly constructed illustrations built on real prices and real dates, and are introduced as such.
Partly unverifiableX08
“Aggregate 2026 hyperscaler capital-expenditure totals of $600-725 billion.”
Verdict: PARTIALLY UNVERIFIABLE
Why: Every figure in that range came from secondary aggregators, several of them low-quality, and they disagree with each other by more than a hundred billion dollars. Not narratable at that precision from primary sources; the frontier block uses company-reported results instead.
UnresolvableX09
“AI capital spending is a bubble, is peaking, or is about to fall.”
Verdict: UNRESOLVABLE — MUST BE STATED AS UNKNOWN
Why: A hypothesis about the future presented as analysis. Episode 12 examines the question properly; Episode 2 must not smuggle a market call into a lesson about reasoning.
RefutedX10
“The company regained Nasdaq compliance in January 2026.”
Verdict: REFUTED
Why: A secondary 2026 article asserts this. The filing index shows the catch-up filings on February 25, 2025 and the compliance announcement on February 26, 2025. Included here because it is exactly the kind of confident secondary claim that ends up in a script.
UnverifiableX11
“Any fiscal 2026 figure described as audited, or any figure taken from the fiscal 2026 Form 10-K.”
Verdict: UNVERIFIABLE
Why: That report had not been filed as of the compile date. The company's own results release is labelled unaudited and the episode must say so where it matters.
RefutedX12
“Quarterly revenue or gross-margin figures for the delayed-filing period as returned by commercial data aggregators.”
Verdict: REFUTED
Why: The aggregator checked returned duplicated values for four quarters spanning the delayed filings, which would have put four wrong numbers on screen. All fundamentals in this episode come from SEC XBRL company facts or the company's own releases.
Partly unverifiableX13
“The 20% single-day fall on August 7, 2024 was caused by the gross-margin figure.”
Verdict: PARTIALLY UNVERIFIABLE
Why: Several things were announced together. The margin figure is narratable; the market's reason for the move is not, and market reaction to earnings is Episode 3's subject in any case.

What to watch next

Dated material, and what would make it stale

NVIDIA reported quarterly revenue of $96.2 billion (+106% year over year), data-centre revenue of $89.0 billion (+117%), and guided the next quarter to $108.0 billion.
26 Aug 2026
Status: reported
What changes: Superseded by NVIDIA's next quarterly report (expected late November 2026).
Invalidated by: The next results release.
Measured at the close at the end of August 2026, NVIDIA, AMD, Broadcom, Micron, Vertiv, Arm and Super Micro were all below their own record closes, by widely different margins; Super Micro was the furthest below.
28 Aug 2026
Status: as at the compile date
What changes: Every number here changes with the next trading day.
Invalidated by: The next close.
Super Micro guided fiscal 2027 net sales to $65.0-72.0 billion, against $39.1 billion in fiscal 2026.
11 Aug 2026
Status: company guidance, unaudited results
What changes: Guidance is revised or confirmed at the fiscal 2027 first-quarter report, expected around early November 2026.
Invalidated by: The next guidance update.

Ideas we borrowed, and tested

Thinkers, taken seriously enough to argue with

Claim → author → evidence → counter-argument → historical test → current relevance. Never doctrine.

Annie Duke

Thinking in Bets (2018)

'Resulting' — judging the quality of a decision by the quality of its outcome, when the two are joined only loosely and through luck. Duke's practical response is to grade the process: what you knew, what you believed, what would have changed your mind.

Evidence:: This episode's case supplies an empirical instance rather than proof: the identical stated reasoning produced a many-fold gain from one date and a deep loss from another, about a year apart.
Counter-argument:: Over enough decisions, outcomes do carry information about process — a sample of one tells you nothing, but a sample of five hundred tells you a great deal. The habit is not 'ignore outcomes'; it is 'do not let a single outcome be the whole verdict'.
Historical test:: The distinction long predates the phrase, in poker, in insurance underwriting and in medicine, where a correct treatment with a bad result is a recognised and separately-classified event.
Current relevance:: A three-year bull market in one theme has manufactured an unusually large number of people whose only evidence about their own judgment is a profit.

Nassim Nicholas Taleb

Fooled by Randomness (2001)

Alternative histories — a realised outcome is one draw from a distribution of worlds that could have happened, and judging a decision by the single world you got is a category error. Related: survivorship, where only the winners are still around to be interviewed.

Evidence:: The counterfactual in this case is unusually cheap to state: the same company, the same reasoning, a different entry date.
Counter-argument:: Taken too far this becomes unfalsifiable — any loss can be excused as an unlucky draw and any win dismissed as a lucky one. It is a discipline for grading process, not a way to avoid ever being wrong.
Historical test:: The 1990s produced an entire generation of celebrated managers whose records did not survive the following decade.
Current relevance:: Almost every AI-era track record on public display is measured from a starting point inside the same upswing.

Michael Mauboussin

The Success Equation (2012)

The luck-skill continuum: activities sit on a spectrum, and the more luck contributes to an outcome, the larger the sample needed before results say anything about ability.

Evidence:: Used only as a framing for why one trade cannot settle the question; no quantitative claim is made from it.
Counter-argument:: Placement on the continuum is itself contested and hard to measure.
Current relevance:: Directly explains why the shortest AI-era track records are the most confident ones.

Evidence & sources

20 sources, by tier

Tier 1 is primary and authoritative — filings, regulators, official statistics. Journalism and books are attributed ingredients, never proof by reputation.

Primary sourcePublicly reported end-of-day market prices (split-adjusted). Specific price levels and percentage moves are not reproduced in this note. Split ratio and effective date 2024-10-01 confirmed against the company's own corporate actions.supports F01
Primary sourcePublicly reported end-of-day market prices (split-adjusted). Specific price levels and percentage moves are not reproduced in this note.supports F02, F03, F04
Primary sourceSuper Micro Computer, Forms 10-K, via SEC XBRL company facts (CIK 0001375365) (opens data.sec.gov)supports F05
Primary sourceSuper Micro Computer, 'Supermicro Announces Fourth Quarter and Full Fiscal Year 2026 Financial Results', Exhibit 99.1 to Form 8-K (opens sec.gov)supports F06
Primary sourceSEC XBRL company facts (fiscal 2023-2025) and the company's fiscal 2026 results releasesupports F07
Primary sourceSuper Micro Computer, Form 10-Q for the quarter ended December 31, 2025, via SEC XBRL (opens sec.gov)supports F08
Primary sourceSuper Micro Computer, fiscal 2026 fourth-quarter results release (opens sec.gov)supports F09
Primary sourceSEC XBRL company facts, InventoryNet (CIK 0001375365)supports F10
Primary sourceSEC Press Release 2020-190, 'SEC Charges Super Micro and Former CFO in Connection with Widespread Accounting Violations' (opens sec.gov)supports F11
Primary sourceSuper Micro Computer, Form 8-K dated January 9, 2020 and Form 10-K for fiscal 2020 (opens sec.gov)supports F12
Primary sourceSEC XBRL company facts for the quarters ended March, June, September and December 2022; share count from the Form 10-Q cover page dated January 31, 2023; market value from publicly reported pricessupports F13
Primary sourceSuper Micro Computer, Form 10-Q for the quarter ended March 31, 2026, risk factors (the company's own description of the report); price move as publicly reported (opens sec.gov)supports F14
Primary sourceSuper Micro Computer, Form 8-K filed 2024-08-28 and Form NT 10-K filed 2024-08-30; price move as publicly reported (opens sec.gov)supports F15
Primary sourceSuper Micro Computer, Form 8-K filed October 30, 2024, Item 4.01 (opens sec.gov)supports F16
Primary sourceSuper Micro Computer, Form 8-K filed December 2, 2024, Items 5.02 and 8.01 (opens sec.gov)supports F17
Primary sourceSuper Micro Computer, Forms 10-K and 10-Q filed 2025-02-25 and Form 8-K filed 2025-02-26 (Item 3.01) (opens sec.gov)supports F18
Primary sourceSuper Micro Computer, Form 10-Q for the quarter ended March 31, 2026, Item 4 (opens sec.gov)supports F19
Primary sourceSuper Micro Computer, Form 10-Q for the quarter ended March 31, 2026, risk factors — the company's own filed description (opens sec.gov)supports F20
Primary sourceNVIDIA, 'NVIDIA Announces Financial Results for Second Quarter Fiscal 2027' (opens nvidianews.nvidia.com)supports F21
Primary sourcePublicly reported end-of-day market prices, each name against its own record close since December 2022. Specific levels and percentages are not reproduced in this note.supports F22

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. The fact-check verdicts recorded in its evidence packet are shown here instead.

SMCI quarterly revenue and gross margin, all quarters 2022-2026
Verdict: CONFIRMED — and the commercial aggregator's version REFUTED: it duplicated the December-2023 quarter into June-2024 and the March-2024 quarter into September-2024. Four numbers would have been wrong on screen.
SEC XBRL company facts, latest-filed version per period
March 13, 2024 is the record closing high
Verdict: CONFIRMED — the March 13, 2024 close is the record close; an intraday record printed five days earlier, on March 8, 2024, and is deliberately not used.
Split-adjusted daily price series 2021-2026 (levels not reproduced in this note)
The company regained Nasdaq compliance in January 2026
Verdict: REFUTED — catch-up filings 2025-02-25, compliance 8-K 2025-02-26. The secondary source asserting January 2026 is unreliable.
EDGAR filing index for CIK 0001375365
SEC settlement terms, August 2020
Verdict: CONFIRMED — $17.5 million penalty; former CFO separately penalised; CEO not charged, $2.1 million clawback.
SEC Press Release 2020-190, read in full
Ernst & Young's stated reason for resigning
Verdict: CONFIRMED verbatim.
Form 8-K filed 2024-10-30, Item 4.01, quoting the resignation letter
The special committee's findings
Verdict: CONFIRMED verbatim — including the finding that the auditor's conclusions were not supported by the facts the committee uncovered. The record is genuinely contested and the episode presents it as such.
Form 8-K filed 2024-12-02, Items 5.02 and 8.01
NVIDIA Q2 fiscal 2027 figures
Verdict: CONFIRMED — $96.2bn revenue, $89.0bn data centre, 75.0% gross margin, $108.0bn outlook. A secondary source reporting non-GAAP EPS of $2.22 was not used; the GAAP figure of $2.46 comes from the company.
NVIDIA's own results release, August 26, 2026
Aggregate hyperscaler 2026 capex figures circulating in secondary media
Verdict: REFUTED as narratable — the aggregators disagree by more than $100bn and the XBRL series are reported on inconsistent bases across the group. Excluded (X08).
Attempted reconciliation against SEC XBRL capital-expenditure facts for the four largest spenders

Community

Talk it through

Discuss this research

Members only · in 📡 Intelligence Watch

Open in Community

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.

Join the discussion with a free account Already train here? Sign in See the room

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.

Practice this judgment

Verdict Replay

The engine's own description is 'same scenario, blind re-answer, sealed old verdict revealed after commit' — the learner measures the distance between a past judgment and a…

Open

Practice this judgment

Practice journal

Where a practice decision is recorded and later carries a reviewed / needs-review state — the app's existing home for the four lines this episode asks the learner to write.

Open

Train it

Make five honest calls on today's market

Commit before the outcome exists, then see what happened. Practice is where this research gets measured.

Train 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.

Learn the concepts underneath

Process vs. outcome Margins & unit economics Drawdown anatomy Narrative vs. data All of the Academy

Community

Join the discussion

Challenge the argument, bring evidence, or ask what would change our mind — in the open, with other members.

Mentor

Ask 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 attached

Provenance — 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 case-selection; 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 2af5e640a169…	—
Evidence packet	course/research/AI02-evidence-packet.json	sha256 27d27cebbb2b…	compiled 2026-08-28 · rev 1
Approved script	course/scripts/AI02-outcome-is-not-process.json	sha256 c88ebcf51d0b…	v v5 — visual director pass
Episode manifest	course/manifest/AI02.manifest.json	sha256 434ab2e53462…	—
Approval record	renders/approved/AI02-FINAL.json	sha256 71540628aacc…	compiled 2026-08-31 · v 2e8a9b7
Poster still	qa/stills-AI02/S00-055.png	sha256 5a9d462dec51…	—

Approved master (AI02-FINAL-elevenlabs.mp4): sha256 8f7f8a91cf103478d3cc492d75b9bcb2633c71235feaa256174702d3bff20c7e

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.

What this research concludes

Which real AI-era investment best teaches that a profitable outcome does not prove good reasoning, and that a bad outcome does not prove bad reasoning?

Researched to be refuted, not confirmed

22 claims survived refutation

The company after the collapse

The strongest historical case

What we still do not know

13 claims we would not say — and why

Dated material, and what would make it stale

Thinkers, taken seriously enough to argue with

20 sources, by tier

The research desk

Talk it through

Take it further

That looks interesting too

NVIDIA Beats Earnings. Why Could the Stock Still Fall?

Everyone Knows AI Is the Future. So What Do You Actually Invest In?

Can AI Become a Better Investor Than You?