The Market Took $1.2 Trillion Off Nvidia While Its Earnings Estimates Barely Changed

One-page summary

1. A Customer Relationship Became a Board Crisis in Fourteen Days

The story the market is pricing began a year before the stock reacted to it. In October 2025 The New York Times reported that the Commerce Department was investigating Megaspeed, a Singapore company that had bought more than $4.8 billion of Nvidia hardware and was suspected of serving Chinese customers from data centers in Malaysia and Indonesia, and Nvidia said its own review had found no reason to believe its products were diverted.^[14] The shares went on to a record close of $239.24 on October 6 of this year. What changed on October 29 was the subject: the Times reported that American officials had found a close personal relationship between Mr. Huang and a former Megaspeed executive, and that United States intelligence believes she works for China’s Ministry of State Security.^[14] China calls the allegation a smear, and Singapore’s Ministry of Home Affairs says it never investigated her for espionage, warned her for giving police inaccurate information about her role, found insufficient evidence of any offence and barred her from returning.^[15]

I set out the record in Exhibit 1 with each row marked by what kind of fact it is, because the valuation depends on keeping an admission, a company filing and a press report citing unnamed people in separate columns. I have not named any of the women. Where the reporting I cite names them, the names add nothing to the investment case.

Exhibit 1 — The record, in order

The filing gives no reason for the departure. The board answered an allegation about national security on a Sunday with a statement of confidence, and it accepted a resignation within hours of a report about a subordinate on a Wednesday; that timing is consistent with a rapid change in the board’s position after the second report, but the public record does not establish the board’s reasoning. Separately, a person familiar with the meeting told CNBC that sentiment turned quickly after that report.^[10] No outlet has reported a vote count, a dissent or an ultimatum, and I have found no source for any claim about how individual directors voted. Three limits on what is known matter for the stock: none of the reporting I cite describes a criminal allegation against Mr. Huang as of November 12, Bloomberg reports that he is not a target of the federal inquiry in New York, and the company has not said whether the Wachtell investigation continues now that he has gone.^[16]

The record is still moving: the Reuters report in the last row arrived during Thursday’s session and rests on two unnamed people. I carry it in the record as an unconfirmed allegation and have not changed a probability on it, and the Housekeeping section says what would make me do so.

2. A Long-Tenured Board Now Controls the Succession

Two people now carry the decisions that one person carried for thirty-three years, and the investment case turns on them and on the founder who left (Exhibit 2).

Exhibit 2 — The founder, the interim chief and the chair

The directors: ten remain, all classed as independent, and four of them joined between 1993 and 1997 (Exhibit 3). Tench Coxe came from Sutter Hill Ventures and Mark Stevens from Sequoia Capital, two of the company’s earliest investors, and each holds more than thirty million shares; Brooke Seawell was finance chief of Synopsys while Mr. Jones ran it. The lead independent director, Stephen Neal, is chairman emeritus of Cooley, the law firm where Nvidia’s general counsel spent two decades before joining the company in 2017, and he drew the lowest support of any director in June at 86.7%.^[3][8] Semafor counts an average tenure of almost fifteen years, about twice the norm for a large company, and describes a board with no current or former chief executive of an S&P 500 company and no director with a national-security background; the description is Semafor’s and I have not rebuilt it.^[13][33] The word “succession” does not appear in the 2026 proxy statement on a full-text search, which does not by itself show that the board had no plan.^[3]

Exhibit 3 — The ten directors

Source: Nvidia 2026 proxy statement (shares at March 23, 2026) and Form 8-K of June 30, 2026 (votes for as a share of votes for and against).^[3][8] Shaded rows joined in the 1990s. Persis Drell resigned from the board on January 20, 2026 and Mr. Huang on November 11, which leaves the ten shown.^[1][4]

The bench: the engineering leadership is long-tenured and the general management is not. Ian Buck, who created the CUDA software platform and runs the hyperscale business, joined in 2004; chief scientist Bill Dally joined in 2009; Jonah Alben, who joined in 1997, leads graphics-processor engineering; Michael Kagan is chief technology officer; and co-founder Chris Malachowsky remains as a fellow.^[10][13] Against that, the operations chief is 71, the head of sales retired over the summer and his successor arrived from Microsoft in late August, and the chief accounting officer started in May.^[4] I see a company that can execute the roadmap it has and has no disclosed successor with comparable enterprise-wide operating experience, which is why the search is my first monitoring priority.

3. The Company Keeps the Roadmap and Loses the Person Who Set It

What Mr. Huang did that a successor does not inherit by appointment was to commit the company, and through it the supply chain, years ahead of demand. Reuters reports that a Taiwan supplier executive expects whoever follows to be less aggressive in securing capacity, and that a Korean chip official’s first question is whether the relationship survives him.^[12] Citi’s analysts wrote that strategy, product vision and customer relationships all ran through one office, and Bernstein and Evercore both point out that there is no second-in-command.^[22] He also carried the company’s case on export policy to two administrations, and that asset has become the reason Congress is writing a bill.

What the company keeps is everything already in motion. Production shipments of the Vera Rubin platform began in August and management expected it to be about a fifth of data-center revenue in the October quarter, and Goldman Sachs reads the order book as twelve to twenty-four months deep with a product roadmap of two to three years.^[6][22] The company employs about 42,000 people, 31,000 of them in research and development, and lost 3.7% of them last fiscal year.^[4]

The employees: the available evidence on employee sentiment is limited, and I use it as an indication of mood and not as a count. Blind, an anonymous workplace forum, polled 232 users who identify as Nvidia employees on November 9 and 10, and 88% said Mr. Huang should remain chief executive, against 60% of professionals at other companies.^[21] The poll closed before the second report, it is self-selected, and no methodology is published. A person inside the company told CNBC that employees were in tears on Thursday.^[10] No petition, open letter or walkout has been reported, which matters in section 6.

Retention: the form of his equity award forfeits unvested units when service ends, the company has no severance agreements with its executives, and its recovery policy applies only after an accounting restatement, so I have identified no disclosed financial arrangement that gives Nvidia leverage over what he does next.^[3] I weigh that more heavily over three years than anything in next week’s results.

4. Stable Estimates Are Consistent With Near-Term Supply Constraints

If the company keeps what is already in motion, the estimates should show it. Consensus estimates have barely changed: the fiscal 2028 figure on Yahoo Finance was $15.91 a share thirty days ago and seven days ago and is $15.93 now, on that page’s own trend table, and my pull of October 26 was $15.92.^[23] I interpret that stability as consistent with a near-term outlook limited by supply, supported by committed capacity, customers’ spending plans and the existing product roadmap. It does not establish that earnings are insulated from execution, pricing or regulatory change. Revenue in the July quarter was $96.2 billion with a 75.0% gross margin, the guide for October is $108.0 billion with no data-center compute sold into China, and consensus stands at $109.0 billion for October and $124.9 billion for January.^[5][23] Ms. Kress said in August that fiscal 2028 revenue would grow about 70% and that the figure was limited by supply through at least the end of that year.^[6]

The suppliers say the same thing on their own calls. TSMC’s chief executive told investors on October 15 that demand remains far above what the company can supply, SK hynix said on October 27 that its 2027 high-bandwidth memory was sold with volume and price settled, and Samsung and Micron each said their 2027 output of the same memory was sold out.^[24] Nvidia’s own supply commitments rose to $279 billion in the July quarter from $119 billion, with $92 billion falling in the rest of this fiscal year, $87 billion in fiscal 2028, $88 billion in fiscal 2029 and the remaining $12 billion after that.^[5] Supplier statements, customer budgets and Nvidia’s own purchase commitments are three different kinds of evidence, and none of them is a customer order. Together they may cushion the next several quarters against a leadership disruption, although they do not remove execution or demand risk.

Demand: the four largest buyers spent $208.8 billion on capital expenditure in the third quarter, Alphabet said 2027 spending would increase significantly and Amazon said it would spend meaningfully more, and CoreWeave’s contracted backlog reached $132.5 billion.^[25] I identified no public statement from a customer announcing a change to an order as of November 12.

The counterweights: Alphabet, Amazon and Meta each ran negative free cash flow in the third quarter, and Baird models 2027 spending on AI equipment growing about 40% against Nvidia’s 70%.^[22][25] Memory prices are the other side of sold-out supply, and Ms. Kress flagged in August that gross margin would bottom at 71% to 72% in the January quarter.^[6] Receivables stretched to 60 days of sales from 45, and the company’s commitments and guarantees total about $530 billion, including a residual-value guarantee of up to $105 billion backing OpenAI’s leases at an SB Energy data-center campus in Ohio, which with $3.5 billion of other guarantees makes up the $108.5 billion in Appendix B.^[5] Broadcom is guiding to about $115 billion of AI revenue in its fiscal 2027 and about $230 billion in its fiscal 2028, AMD expects to grow more than 80% in 2027, and both materially outperformed Nvidia this week (Chart 3).^[27]

The buyback: secondary reports put the September 28 increase in the repurchase authorization at $150 billion, taking the unused amount to $235 billion through fiscal 2028, which is 5.9% of today’s market value.^[26] At roughly ten times earnings, repurchases on that scale could materially increase earnings per share, subject to funding costs and other uses of the capital, and I will read Tuesday’s cash-flow statement for whether the pace has risen from the $19.7 billion of the July quarter.

5. Washington Is Now the Larger Variable

The resignation removes the person and leaves the questions about the company. Semafor reports that three Justice Department inquiries touch Nvidia: an antitrust inquiry into the company itself, a national-security investigation of Megaspeed, and an inquiry in New York into servers diverted through a Super Micro channel, with no indication of charges against Nvidia or Mr. Huang.^[13][16] A securities class action filed on November 9 in San Jose covers purchasers from October 9, 2025 to November 8, 2026.^[20] The thread I watch is the reported call from Nvidia’s general counsel to a Commerce official during the 2025 inquiry, because it is the one item in the record that concerns the conduct of the company and not of one man.^[16]

Congress: the chairman of House Foreign Affairs says Mr. Huang has agreed to testify before Thanksgiving.^[17] The AI Overwatch Act, which would require thirty days’ notice to Congress before Commerce licenses chips of the H200 class and would deny licences for the Blackwell class outright, has 56 co-sponsors across the two chambers and sits in the Senate’s defense package, and the Speaker said on Thursday that a consensus is forming to limit exports.^[19] The two senior Democrats on the tax committees opened an inquiry into the H200 approvals the same day, and their party takes both chambers on January 3.^[18]

Commerce: no pause or revocation of an Nvidia export licence had been publicly disclosed as of November 12. A rule that would require licences for Blackwell-class chips sold to local cloud providers in seven more countries, Singapore and Malaysia among them, is reported to be close.^[19] Nvidia’s guidance already assumes no data-center compute revenue from China and H200 shipments were under 1% of data-center revenue in the July quarter, yet customers headquartered in China and Hong Kong accounted for $7.9 billion, or 8.2% of revenue, and the filing does not reconcile the two.^[5] Baird estimates that 10% to 30% of the Southeast Asian channel serves Chinese demand and that an H200 ban could bring a write-down of up to $5 billion.^[22]

I forecast a 5% reduction in fiscal 2028 earnings in the outcome where Washington acts, well below the headline exposure, and the reasoning is the same supply constraint: in a year when demand exceeds available supply, a closed channel moves units to another buyer. That is my inference and not a disclosure. The cost arrives later, as a smaller addressable market in fiscal 2029 and as a lower multiple, and it is the reason the base case carries the largest weight in section 8.

6. I Have Found No Disclosed Bar on Competing, and the Next Scheduled Election Is in June

The question that follows is whether a founder this popular with his employees could take the company back. What follows is my reading of public documents and statutes. It has not been reviewed by counsel and it is not legal advice. I do not see a bid as practical for a holder of 3.4% of a $4.01 trillion company, for three separate reasons: Delaware’s Section 203 restricts business combinations for three years with a holder who acquires 15% or more without the board’s prior approval, subject to the statute’s exceptions; the board has blank-check preferred stock from which it could adopt a rights plan without a stockholder vote; and I know of no financing at this size.^[9] The routes that remain run through the ballot, and the company’s own documents narrow most of them (Exhibit 4).

Exhibit 4 — The routes back, from the bylaws and the charter

Source: Nvidia Amended and Restated Bylaws (March 2024), Restated Certificate of Incorporation as amended, and the 2026 proxy statement’s deadlines for the 2027 meeting. Other routes would require separate analysis.^[3][9]

The votes: a contested election is decided by plurality and has no fixed threshold, so I illustrate the scale with the June meeting’s turnout. At that meeting 69.6% of the shares voted on non-routine items, and half of that, 34.8% of the shares outstanding, is an illustrative majority of the votes cast.^[8] Mr. Huang held 3.4% at his last filing in September, the other directors and officers held 0.4% in March, and BlackRock, Vanguard, State Street and Geode held about 23.3% between them at their latest filings, although they do not vote as a bloc.^[3][7][29] As an illustration only, if all four voted with the board he would need about 73% of every other share that voted (Chart 4; Appendix C shows the arithmetic). The supermajority thresholds look out of reach on the company’s own evidence: when the board itself asked stockholders to remove them in 2025, the proposal drew 65.7% of the shares outstanding and failed. In June 2026 stockholders passed an advisory proposal for simple-majority voting, 14.6 billion shares for and 2.2 billion against, but changing the charter needs a further stockholder vote, so the thresholds stand at least until the 2027 meeting.^[8]

What a return has taken elsewhere: OpenAI’s board reversed itself in five days in 2023, with nearly the whole staff threatening to leave, a partner offering every one of them a job and a board that had given no specific cause. Gildan’s founder returned in five months in 2024 behind an activist with about 5% in a dispute over strategy. The one founder in my set who forced his way back, at WWE, controlled about 81% of the votes and returned as executive chairman, not chief executive, and founders who tried with stakes between 3.5% and 30% at Men’s Wearhouse, Best Buy, Papa John’s and American Apparel did not succeed.^[31] I see none of those conditions here as of November 12: EDGAR’s index of filings for Nvidia showed no Schedule 13D or dissident proxy material when I checked that day, no top-ten shareholder has spoken publicly that I can find, no employee campaign has been reported, and CNBC reports that Mr. Huang has retained the law firm Paul Weiss for a purpose it could not determine.^[10]

I put the probability that Mr. Huang is chief executive of Nvidia again within twelve months at about 5%, nearly all of it through the first route and conditional on an investigation that finds no substantiated misconduct within its disclosed scope. Section 8 uses the same 5% as a valuation branch and explains why the two are the same number.

March 26 is a potential pressure point: the nomination deadline creates a potential focal point for shareholder pressure; it does not establish that the board is running the search to that timetable. Nominations for the 2027 meeting close on March 26, and a board that reaches that day with an interim chief executive and no named successor leaves a potential opening for a dissident campaign.

Competition: I have not identified a disclosed agreement that categorically bars him from joining or starting a competitor. California generally prohibits contracts that restrain a person from a lawful business, subject to statutory exceptions; confidentiality, trade-secret, equity-award and other contractual obligations may still constrain particular activities, and I have not established the full set of his obligations.^[28] The company has disclosed no separation agreement, and a later filing is where one would appear. When HP’s board removed Mark Hurd in 2010 he was at Oracle thirty-one days later.^[31]

The stake: a separate consideration is the potential disposition of his stake, 811.5 million shares worth about $135 billion, or 6 days of normal trading volume. As I read the reporting rules, public transaction-level visibility into that holding may diminish from February, and Appendix C sets out the reading and its limits.^[37] I do not predict a sale. I note that the founder’s shares are a possible source of supply that the discussion of a comeback has left out.

I see more risk to shareholders in what Mr. Huang builds next and in what he does with his shares than in a contest for the board.

7. Fourteen Earlier Departures Show Why the Succession Deserves the Attention

If the founder does not return, the search for his successor becomes the event that matters, and there is a record of how such searches have gone. I reviewed fourteen chief executives of listed companies who left office between 2005 and 2025 after a publicly reported matter of personal conduct, a convenience sample assembled from press coverage before any return was computed, and twelve had enough price history for a twelve-month comparison (Appendix A lists every case and the rule for including it). The median stock fell 2.6% on the first day across the fourteen, and across the twelve the median twelve-month total return finished 0.6 percentage points behind the applicable benchmark, with a range from 40 points behind to 88 ahead. The departure by itself explained little of that range, and a range that wide makes these cases illustrative evidence and not a statistical basis for the probabilities in section 8.^[31]

The outcomes differed with the succession, although a sample this small and windows this long cannot show that the succession caused the difference. McDonald’s and Texas Instruments named a credible chief executive the same day and finished the year within five points of the market. Intel put its finance chief in the chair, appointed him permanent chief executive after 224 days and trailed the index by 18 points at one year and 47 at three. BP did the same over 127 days and trailed by 40. Boeing used its finance chief as a bridge and hired an outsider in 115 days, and it beat the index by 21 points (Chart 5). Five of these boards made the finance chief interim leader, two made that person permanent, and both were replaced about two years later.^[31]

A franchise can grow while its multiple falls: in the twenty months after Steve Jobs resigned in August 2011, Apple’s trailing earnings rose 75% and its price-to-earnings multiple fell from 14.9 to 8.9. HP’s multiple fell 39% in the year after Mr. Hurd left while its earnings rose.^[31] The reassuring case is TSMC, which was flat for a year after Morris Chang retired in 2018 and then returned 235% in three, and that handover came with eight months of notice and two named successors, the condition Nvidia does not meet today.

No precedent matches this one. None combines a founder of thirty-three years, one of the largest companies in the index, a national-security element and no named successor, and eleven of the fourteen were hired managers. These cases illustrate possible succession outcomes; they do not estimate Nvidia’s expected return or establish an appropriate valuation multiple. I take one caution from them: a strong operating franchise does not guarantee a near-term rerating.

8. Three Cases for What Investors Will Pay After the Handover

Sections 1 to 7 describe what could happen, and this section puts a price on it. I value Nvidia on a bull, a base and a bear case applied to the fiscal 2028 consensus of $15.93 a share, built from five underlying outcomes (Exhibit 5), and I compare the result with a ten-year discounted cash flow. The open questions are discrete events, a search, an investigation and a bill, and cases state them more plainly than a single path.

What the number is: $173, which I call my probability-weighted value or price target, is my expectation of the share price at the target date, February 2028, when Nvidia reports fiscal 2028. It is a price at that date and not a present value, and the upside I quote excludes the dividend, $0.25 a share a quarter since Nvidia raised it from $0.01 with its first-quarter results on May 20, 2026.^[5] The multiples are applied to fiscal 2028 earnings because that is the last year with a published consensus and the last year the supply commitments largely cover. At the target date those will be trailing earnings and the market will be pricing fiscal 2029, for which I have no consensus; if fiscal 2029 earnings were 10% above the fiscal 2028 consensus, or $17.52 a share, and investors paid today’s 10.4 times the forward fiscal year, the price would be about $183, and I have not counted that.

Three different discounts: the vacancy is temporary, and in every case but the bear case it ends by June 2027, eight months before the target date, so the multiples in Exhibit 5 are not a discount for an empty chair. They are my estimate of what investors pay in February 2028 for fiscal 2028 earnings that have just been reported, under a chief executive who is new or returning, with whatever lasting discount for execution, governance and regulation remains. A multiple on trailing earnings also sits below one on forward earnings for a growing company: the base-case midpoint of 11.5 times trailing earnings is about 10.5 times forward earnings if fiscal 2029 grows 10%, which is where the shares trade today, so the base case assumes no rerating and the bull case a recovery to about 12 times.

Bases and dates: the consensus is the adjusted, non-GAAP series that Yahoo Finance shows by default, for the fiscal year that ends on January 30, 2028, and both of my snapshots, October 26 and November 12, come from that page.^[23] The target date is February 2028, about fifteen months away, which is also the period over which I carry the cash-flow value forward. Nvidia’s multiple on the next twelve months and the three peer multiples come from one provider, StockAnalysis, on one date, November 11, each scaled by the change in its share price since. The historical multiples in Chart 2 are the forward figures a provider reported at each date, from Yahoo Finance and StockAnalysis, and I have not rebuilt them on a single method, so that chart shows direction and its levels are approximate.^[32]

How an outcome is classified: several of these events can happen together, so the cases follow one rule, applied in this order. An outcome is in the bear case if any one of three things happens. The first is a company finding before the target date: a criminal charge against Nvidia, or a settlement, consent decree or monitor arising from the inquiries in section 5 that imposes operating restrictions or costs more than $5 billion, or a licence revoked for sales outside China. The second is no permanent chief executive named by the annual meeting expected about June 24, 2027. The third is two or more of Mr. Buck, Mr. Dally, Mr. Alben and Mr. Kagan leaving within twelve months of November 12, 2026. The three are alternatives and any one is sufficient, and where a company finding coincides with either of the others the finding governs. A return of the founder as chief executive, absent a bear event, is its own outcome inside the bull case whatever happens to export policy. I model it at consensus earnings as a representative assumption, and I treat a return that coincides with a bear event as negligible because a reversal by the board presupposes an investigation that clears the company, so the 5% is both my forecast of a return within twelve months and the weight of that branch. For every other outcome the test is economic: export-policy changes that reduce my modelled fiscal 2028 earnings by 2.5% or more put it in the base case, where I use a 5% reduction as the representative assumption, and a smaller reduction or none, because existing contracts are honoured or the units are resold at the same price, puts it in the bull case even if a bill has passed. Who the permanent chief executive is does not decide the case, and my forecast that it will be an insider is in section 9. Confirming Ms. Kress is therefore not a bear event by definition. It is my judgment, from a record that is entirely in finance and from the two finance chiefs made permanent in section 7, that it would hold the multiple at or below the base-case range, and section 11 treats it as a reason to move weight and not as a classification.

Exhibit 5 — Bull, base and bear values at the February 2028 target date

Fiscal 2028 consensus earnings per share of $15.93, adjusted by my estimate for each outcome, times the multiple range shown. Case midpoints are the probability-weighted midpoints of the outcomes inside them. Probabilities are my judgment. Source: Yahoo Finance consensus, November 12; my estimates.^[23]

The bull case, 33%, $191–231: I expect that in this case the board names a permanent chief executive with a credible record before nominations close on March 26 and keeps Ms. Kress as finance chief, the Wachtell investigation reports in the first half of 2027 with no finding against the company, and repurchases rise toward $40 billion a quarter. The Rubin platform ships on schedule, fiscal 2028 earnings arrive near $16, and the multiple recovers to between 12 and 14 times as the vacancy closes. The 5% I give to the founder’s return sits here.

The base case, 35%, $159–189: Ms. Kress steadies the company and a successor is named by the annual meeting. Congress passes the export provisions, Commerce issues the cloud-provider licensing rule and the H200 inventory is written down, and fiscal 2028 earnings come in about 5% under consensus. The multiple stays between 10.5 and 12.5 times because the franchise is intact and the question of who sets its direction is answered late. Mr. Huang testifies, mounts no contest and starts something new, and the shares spend most of the next year between about $150 and $190.

The bear case, 32%, $108–158: any one of three things happens. No permanent chief executive is named by the annual meeting, or two or more of the four senior engineers leave within twelve months, some of them to the founder, or an inquiry ends in a charge against the company, or in a settlement or monitor that restricts how it operates or costs more than $5 billion. I expect the first evidence of the succession and retention roads to appear in pricing before it appears in units shipped: a gross-margin guide under 71% as the largest customers negotiate fiscal 2029 supply without Mr. Huang across the table. Earnings fall 10% to 15% below consensus and the multiple falls to between 8 and 11 times.

Probability-weighted, the three cases give $173. Every midpoint sits at or below the 13.5 times fiscal 2028 earnings that the market paid two weeks ago, on a different earnings basis: those earnings were then more than a year ahead, and at the target date they will be trailing earnings. That is a judgment: the Apple and HP records in section 7 illustrate that a multiple can stay low while a question of leadership is open, and they do not establish what Nvidia’s should be.

Sensitivity: shifting 10 percentage points of probability from the orderly-handover outcome to the two bear outcomes, split equally between them, lowers the value to $166, and shifting 10 points the other way raises it to $180. The value is less sensitive to the weights than to the multiple, and each turn of the multiple on consensus earnings is worth $16 a share.

Discounted cash flow, a present value today: I project revenue for ten years from the fiscal 2027 consensus of $412 billion along a path for each outcome, convert it to free cash flow at a margin between 35% and 42% against 41.9% over the last twelve months, and discount at 11% (Exhibit 6; Appendix B gives every annual input so the figures can be reproduced and reconciles the cash margin to consensus earnings). Debt is a small share of the capital structure, $33 billion against $4.01 trillion of equity, so I approximate the weighted cost of capital with the cost of equity. I use 11% as a normalized assumption and not the rate implied by the reported historical beta: on a 5.1% ten-year Treasury yield and a 4.5% equity premium, 11% corresponds to a beta of 1.3, and the five-year measured beta of 2.25 would give about 15%. This choice materially affects value, and the sensitivity in Exhibit 6 should be read as part of the conclusion. Terminal growth is 3.5% in the two better outcomes and 3.0% in the others, and terminal value is about half of the total.^[5]

Exhibit 6 — Discounted cash flow by outcome and discount rate ($ per share, present value)

Ten-year explicit forecast, fiscal 2028 to 2037; specification in Appendix B. Source: my model; inputs from Nvidia’s Form 10-Q and consensus.^[5][23]

The two approaches answer different questions. The discounted cash flow estimates the present value of the scenario cash flows at my assumed discount rate: about $179 a share today on the same probabilities, 8% above the price. The earnings-multiple framework estimates the price investors may pay in February 2028. To set them side by side I carry the cash-flow value forward fifteen months at the same 11%, which gives about $204 including dividends, or about $203 as a share price after the $1.25 a share of dividends paid in between, against my $173 price target. I use the multiple framework for the price target because I expect a valuation discount to persist. That discount is an additional assumption, not independent confirmation from the cash-flow model, and it may overlap with risks already reflected in the operating forecasts: the probabilities weight the outcomes, the earnings and growth assumptions set what each outcome is worth to the business, the multiple sets what investors pay for it, and I cannot show that the third is free of the first two. Each point of discount rate moves the present value by about $25, more than the gap between my first two outcomes. At today’s price, my reverse DCF requires free cash flow growth of 15% a year for ten years from the $127 billion of the last twelve months at an 11% rate, or 11% at 9%, on a 3% terminal growth rate, net cash of $23.2 billion, 24.1 billion shares and year-end cash flows (Appendix B). That calculation assumes a smooth path from trailing free cash flow and does not incorporate the front-loaded growth in my explicit forecast, so it is a price-implied shorthand and not a comparable operating case. The more relevant figure is that, if fiscal 2028 arrives as in my base outcome with free cash flow of about $267 billion, today’s price requires growth of about 6% a year over the following nine years.^[32]

Relative value: on next-twelve-month earnings Nvidia trades at 13.7 times, against 21 times for Broadcom, 19 for TSMC and 39 for AMD, a discount of about 32% to the first two where it traded level with TSMC in August.^[32] Most of the firms that changed their view this week left their estimates alone. The five downgrading firms whose new targets I can source sit between $180 and $200: Baird and Bernstein at $200, Evercore at $186, Melius at $185 and Craig-Hallum at $180. Goldman Sachs at $220 and Citi at $225 kept their ratings, and Dan Ives values the shares at $225 without Mr. Huang against $300 with him.^[22] My value sits under all of them because I assume a valuation discount persists to the target date.

9. Seven Probabilistic Forecasts I Will Be Scored On

A probability attached to a case is not a forecast until it names an event, a date and a way to score it. The seven below are registered in my ledger with today’s date and this wording (Exhibit 7). Forecast 5 has two parts, and 5b is scored only if 5a resolves yes, which is why Exhibit 7 has eight rows for seven forecasts. Each will be scored yes or no on its outer date at the latest, and I will publish the Brier score, the squared difference between the probability and the outcome, for each one and for the set. A forecast under 50% is a statement that the event is less likely than not, and it is scored the same way.

Exhibit 7 — The forecasts, their definitions and their dates

Where I differ from what is priced, as I read it: the thinly traded prediction markets and much of the sell-side expect an outsider, and one market put about even odds on a name the President suggested.^[38] I expect an insider and an engineer. Most commentary treats the export bill as the risk and its defeat as the good outcome; I think it more likely than not to pass and expect the estimate to survive it. And the discussion of the founder has been about whether he returns, where I give that 5% and put more weight on two things that have had less attention: what he builds, and what happens to the largest individual holding in the company if public visibility into it diminishes.

10. Six Risks, Each With the Number That Answers It

The forecasts say what I expect. These are the six ways I could be wrong, each with the number that bears on it.

11. What Would Change the Rating

My hold band is a twelve-to-eighteen-month probability-weighted value within roughly 10% below and 5% above the market price. At $173 the value is 4.1% above the $166.19 close, so the Hold range is approximately $165 to $192 and the reference price sits inside it. I write this band down here for the first time, and section 12 applies it to my August call. The price sits near the band’s lower edge: a close below about $164.76 would put my value more than 5% above it. I will not change the rating on a price move alone before Tuesday’s report. I re-run the cases on the report, and if my value is then still more than 5% above the close, I will raise the rating in a dated update rather than keep a Hold the band no longer supports. The conditions below are set before Tuesday’s report, and any one of the first six makes a re-underwrite mandatory.

The dates: results and the first call without Mr. Huang on Tuesday, November 17; the company meeting on November 18; his testimony to the House, expected before Thanksgiving on November 26; the company’s Washington conference on December 1 to 3, where the agenda still listed his keynote on December 3; the new Congress on January 3; the nomination window from February 24; and the annual meeting about June 24, 2027. Options priced a move of 6% to 8% on the day of the report as of Thursday morning.^[34]

12. My August View, Scored

On August 26 I published a Hold with a value of about $250 and one test, a revenue guide for the October quarter of roughly $100 billion or more.^[35] The guide was $108 billion and the test passed. The shares closed at $209.66 that day and are 20.7% lower now, while the semiconductor fund fell 9.7% over the same period, so the call lost 11 points to its own sector while the thing I tested came in above my line. My November 3 note promised a re-underwrite by November 11, and this is that note, one day late.

13. My Own Account, and Two Exceptions to My Own Process

I bought my first Nvidia shares years ago and built most of the position in late 2022, and it is the oldest and the largest position in my account. I added to it on November 5 and again in the week of November 9, and I hold a call option that expires in February 2027, all of which my November 11 weekly outlook disclosed.^[35] The shares closed below the exit level in my own written rules on November 10, 11 and 12, and I did not sell.

This note therefore documents two exceptions to my stated process: purchases during the publication blackout, and a decision not to execute the prior exit rule. Neither should be read as compliance with the original policy, and a reader is entitled to weigh this research knowing that I bought on the way down. My reason for holding is the one in section 4: the exit level was written as a stand-in for the estimate breaking, the estimate has not broken, and the multiple fell for a reason the rule did not contemplate. That is an explanation and not an exemption. The revised monitoring framework in section 11 keeps a price-based trigger for mandatory re-underwriting and adds operating, personnel and regulatory tests beside it.

How the holding relates to the rating: a Hold rating says my probability-weighted value is close to the price, and keeping shares I already own is the action that rating describes. Adding is a different decision. My purchases on November 5 and in the week of November 9 were made before this work was finished and on a higher estimate of value than the one I publish today, so they are evidence of my conviction and of my bias, and they are not evidence for the rating. The two exceptions should also lower the weight a reader gives to my price discipline, because I did not follow it, which is why section 11 states its triggers as observable numbers that can be checked without relying on me.

I intend to keep the position, and I may add to it. Any purchase would be a decision about the size of a position I have held for years, it would be disclosed, and it should not be read as a rating above Hold. My policy bars new orders in a covered name until five calendar days after a publication, so I will place none before November 18, the day after the results.

Appendix A: The Fourteen Departures

Exhibit A lists every case behind section 7 so the study can be reproduced. These precedents illustrate possible succession outcomes; they do not estimate Nvidia’s expected return or establish an appropriate valuation multiple.

Inclusion: chief executives of listed companies who left office between 2005 and 2025 after a publicly reported matter of personal conduct, a relationship, a harassment finding or personal behaviour, identified from press coverage. It is a convenience sample of well-documented cases and not a census, and it was assembled before the returns were computed. Papa John’s is excluded because John Schnatter had stepped down as chief executive in December 2017 and left only the chairmanship in July 2018; including it would move the first-day median to −2.7% and the twelve-month median to −4.2 points. Windows: the first-day move runs from the close before the first session on which the news could trade to that session’s close; the twelve-month figure runs from the same base close to the close twelve months later. Returns: total returns from daily adjusted closing prices, which include dividends, in the currency of the listing shown. Benchmark: the S&P 500 through the SPY fund on the same dates, except Nestlé, measured in Swiss francs against the Swiss Market Index; BP is the New York listing in dollars. Exclusions: American Apparel and CBS have no usable adjusted series after the first day, through delisting and merger, so the twelve-month statistics use twelve cases and the first-day statistics use fourteen.

Exhibit A — Chief executives who left over personal conduct, 2005 to 2025

Twelve-month figures are percentage points of total return against the benchmark. The American Apparel and CBS first-day figures are from press reports; successor dates are from the companies’ announcements as reported at the time. Median first-day move, fourteen cases: −2.6%. Median twelve-month relative return, twelve cases: −0.6 points. Source: Yahoo Finance adjusted closes; company announcements; my computation.^[31]

Appendix B: The Discounted Cash Flow, Specified

Exhibit B gives every annual input behind Exhibit 6. Free cash flow is revenue times the margin shown, where the margin is defined as Nvidia defines free cash flow: net cash from operating activities less purchases of property, equipment and intangible assets and principal payments on property and equipment. Stock-based compensation is not deducted, and I hold the share count at 24.1 billion on the assumption that repurchases offset the dilution from awards, which is the pattern of the last two years and may not hold. Timing: year-end discounting, with fiscal 2028 as year one discounted a full year. Cash generated in the remaining quarter of fiscal 2027 is not counted, which roughly offsets the eleven weeks of discounting this convention omits. Terminal value: year-ten free cash flow grown once at the terminal rate and divided by the discount rate less the terminal rate, discounted ten years. Bridge to equity: enterprise value plus cash of $22,443 million and marketable debt securities of $34,143 million, less debt of $33.4 billion, a net $23.2 billion, divided by 24.1 billion shares. Equity stakes in customers and suppliers of $94 billion at carrying value are left out, which understates value by about $4 a share, and guarantees with a maximum exposure of $108.5 billion are not deducted because nothing is accrued against them.^[5]

From earnings to cash: the consensus implies adjusted net income of about $384 billion in fiscal 2028, 55% of revenue, and my orderly outcome converts the same revenue to free cash flow at 41%, or about $284 billion, roughly three quarters of that earnings figure. The gap is the cash that growth absorbs and that adjusted earnings do not show. Receivables ran at 60 days of sales in July against 45 a quarter earlier, so each $100 billion of added revenue ties up about $16 billion; inventory and prepayments under the $279 billion of supply commitments are paid before the revenue arrives; and capital expenditure and principal payments took $2.7 billion in the July quarter, when free cash flow fell to 22% of revenue. I treat the 41.9% of the last twelve months as the ceiling and not that quarter as the norm. Two items sit outside the bridge. Stock-based compensation is excluded from adjusted earnings and is not deducted from free cash flow, so both measures leave out the same cost, which I handle by holding the share count flat. Guarantees of up to $108.5 billion are in neither figure because nothing is accrued against them. The cash-flow value is therefore lower than a simple capitalisation of consensus earnings mainly because of cash conversion and the 11% discount rate; the first year of growth in the orderly outcome is the consensus’s own.^[5][23]

Carrying the value forward: the target-date figure in section 8 compounds today’s value at 11% for fifteen months and deducts five quarterly dividends of $0.25 a share, without reinvesting them. Cash generated and shares repurchased before the target date are already inside the forecast cash flows and the flat share count, and they are not added again.

Exhibit B — Revenue growth by fiscal year, margin and terminal assumptions

Base revenue is the fiscal 2027 consensus of $412.06 billion. Reverse calculation: the growth rate quoted in section 8 is the single rate at which the $127.0 billion of free cash flow of the last twelve months must compound for ten years, with a 3% terminal rate and the same bridge, to equal the market’s enterprise value. The second figure in section 8 fixes fiscal 2028 free cash flow at the base outcome’s $267 billion and solves for the single growth rate of the following nine years on the same terms. Source: my model.

Appendix C: The Founder’s Shares and the Voting Arithmetic

This appendix holds the mechanics behind section 6. It is my reading of public rules and filings, it has not been reviewed by counsel, and it is not legal advice.

The voting illustration: the 34.8% in Chart 4 is half of the 69.6% of shares that voted on non-routine items at the June 2026 meeting. If the four index managers, with about 23.3%, all voted with the board, the shares left to decide a contest would be 69.6 less 23.3 less Mr. Huang’s own 3.4, or 42.9 points, and he would need 34.8 less 3.4, or 31.4 of them, which is about 73%. The managers do not vote as a bloc and pass some votes through to clients, turnout in a contest would differ from June’s, and most of Mr. Huang’s shares sit in trusts where he shares voting power with his wife, so the figure shows scale and nothing more. A dissident would also have to solicit the holders of 67% of the voting power under the universal proxy rule.^{[3][7][8][29][30]}

Visibility into the stake: as I read the rules, a former officer and director reports a trade only if it falls within six months of an opposite trade made while in office, the resale conditions that apply to an affiliate fall away three months after that status ends, and a holder below 5% files no ownership schedule.^[37] Public transaction-level visibility into about $135 billion of stock may therefore diminish from February, once three months have passed since his departure; the timing and extent depend on his continuing affiliate status, his transaction history and the reporting obligations that apply to him. Departure alone does not end disclosure: a trade inside the six-month window is still reported, a Form 144 notice is still required for sales above the rule’s thresholds while he remains an affiliate, and any agreement with the company would be filed by the company. Two founders in my record sold quickly: Steve Wynn sold his entire 11.8% of Wynn Resorts within about forty-five days of resigning, and Travis Kalanick sold 29% of his Uber holding in the tender that settled his dispute with the board.^[31]

A negotiated purchase: the reported unused repurchase authorization of $235 billion is larger than Mr. Huang’s whole holding, so a negotiated purchase of his shares with a standstill is arithmetically possible.^[26] I raise it only because a later Form 8-K or the annual report is where any such agreement would appear. An authorization of that size does not make the structure practical, I have found no indication that one is being discussed, and I draw no conclusion on whether any agreement could restrain competition under California law, which is a question for specialist counsel.

Housekeeping

Verification. On November 12 I opened and read the following myself: the Form 8-K reporting the resignation and its press-release exhibit, the board’s statements of November 8 and November 10, Mr. Huang’s message of November 8 as the company posted it, the House committees’ release of November 5, and Reuters’ report of November 12 on a second employee. After the close I re-read the consensus page and the closing quote on Yahoo Finance. The check for a Schedule 13D or dissident proxy material was made against EDGAR’s index of Nvidia’s filings on November 12; it covers filings made on the company’s record and is not a search of every holder. The proxy statement, annual and quarterly reports, bylaws, charter and Form 4 were read for this note from the SEC’s site. I could not inspect the other newspaper reports on which the allegations rest, which are behind paywalls, or the text of the export bill and of the senators’ letters, and this note is not a source-by-source clearance of that reporting.

A late report, not confirmed. At 2:27 PM ET on November 12 Reuters reported, citing two people familiar with the matter, that Mr. Huang had an intimate relationship with a second employee. Reuters presents her as a different person from both women already reported, the former Megaspeed executive and the vice president in the Times report, so the report, if accurate, describes a third relationship and the second with an employee. Reuters did not name her and could not reach her, Nvidia declined to comment, and Mr. Huang and a representative of the board did not reply. Nothing in the report has been confirmed, and I have not changed a rating, a value or a probability on it. If it is corroborated I would revisit two numbers: the 5% I give the founder’s return, and the 12% I give a finding against the company, because a pattern of relationships with employees would widen what the board’s investigation and any regulator would have to examine.^[40]

Scope and sourcing. Company facts are taken from Nvidia’s own filings and releases, read directly: the Form 8-K and release of November 12, the board statements of November 8 and 10, the 2026 proxy statement, the fiscal 2026 Form 10-K, the Form 10-Q for the July quarter, the bylaws and charter, and Mr. Huang’s Form 4 of September 18. The reporting of The New York Times, The Wall Street Journal, Bloomberg, the Financial Times and The Information sits behind paywalls and was read through the named outlets that reported on it; every such item is marked as reported or alleged in Exhibit 1. Call remarks are from a secondary transcript, and the remark on January-quarter gross margin rests on a single reading of it. The repurchase authorization is from secondary reports of a September 28 release I have not read.

Law. Sections 6 and 9 and Appendix C describe corporate, securities and employment law as I read the public documents and statutes cited. I am not a lawyer, the readings have not been reviewed by counsel, and nothing in them is legal advice.

Forecasts. The seven forecasts in section 9 and the conditions in section 11 are entered in my ledger with their wording and dates before publication, and will be scored as section 9 describes.

Names. I have not named the women in the reporting. An investment note does not need them.

A correction to my own files. My internal working papers described a backlog above $2 trillion as Nvidia’s. Ms. Kress was describing the cloud industry’s contracted backlog on the August call, and Nvidia publishes no backlog figure. Nothing I published carried the error, and the files are corrected.

Trading window. My policy restricts trading in a covered name from ten calendar days before a publication to five calendar days after it, counted from the day a note is actually published. A note published November 12 carries a window of November 2 through November 17; had this note appeared on November 11, as my November 3 note said it would, the window would have run from November 1 through November 16, and the trades below fall inside either. The purchases of shares on November 5 and in the week of November 9, and of the call option, fell inside that window. They were trades inside the window, logged as such in my trade log; they were not protective orders at approved levels. No new order will be placed in the name before November 18.

Sources

1. NVIDIA Corporation, “NVIDIA Announces CEO Transition,” November 12, 2026, and Form 8-K (Items 5.02 and 9.01) filed the same day. Nvidia newsroom · SEC EDGAR
2. NVIDIA, “Statement From NVIDIA Board of Directors” and “A Message to Our Company,” November 8, 2026; “Statement From the Independent Members of the NVIDIA Board,” November 10, 2026. Board statement · Message · Independent members
3. NVIDIA 2026 proxy statement (DEF 14A), filed May 12, 2026: beneficial ownership, director biographies, employment terms, deadlines for the 2027 meeting. SEC EDGAR
4. NVIDIA Form 10-K for fiscal 2026, filed February 25, 2026: executive biographies, employees, turnover, export history; Form 8-K of July 2, 2026 on the sales-chief change. 10-K · 8-K · Form 8-K on Persis Drell’s resignation, January 2026
5. NVIDIA Form 10-Q for the quarter ended July 26, 2026, and the second-quarter release and CFO commentary of August 26, 2026: results, guidance, supply commitments, guarantees, receivables, revenue by customer headquarters. 10-Q · Release · First-quarter release of May 20, 2026, raising the quarterly dividend to $0.25 from $0.01
6. NVIDIA second-quarter fiscal 2027 earnings call, August 26, 2026, from the published transcript at StockAnalysis, a secondary source. Transcript
7. Jen-Hsun Huang, Form 4 filed September 18, 2026. SEC EDGAR
8. NVIDIA Forms 8-K (Item 5.07) reporting annual-meeting votes, filed June 30, 2026 and July 1, 2025. 2026 · 2025
9. NVIDIA Amended and Restated Bylaws (March 12, 2024) and Restated Certificate of Incorporation (December 9, 2021, as amended). Bylaws · Certificate
10. CNBC, live coverage of November 12, 2026, and “Nvidia stock slides after NYT report,” November 11, 2026; CNBC on the Kress and Jones staff messages. Live · Nov 11 · Messages
11. NBC News, “Nvidia CEO Jensen Huang resigns,” November 12, 2026, including the company spokesperson’s statement. NBC News
12. Reuters, on the board meeting (via Yahoo Finance, November 11–12) and on the Asian supply chain (via SRN, November 12). Board · Supply chain
13. Semafor, November 9, 11 and 12, 2026: the board’s composition; the Justice Department inquiries. Nov 9 · Nov 12
14. TechCrunch timeline, November 12, 2026, and Foreign Policy, November 9, 2026, on The New York Times reports of October 9, 2025 and October 29, 2026 and the responses to them. TechCrunch · Foreign Policy · CNBC, October 10, 2025
15. Singapore Ministry of Home Affairs, statement of October 31, 2026, and its November 12 statement as reported by The Straits Times. MHA · Straits Times
16. Bloomberg, via Yahoo Finance, November 10, 2026, on Nvidia’s contact with Commerce; Bloomberg via FA-Mag, November 12, 2026. Nov 10 · Nov 12
17. House Select Committee on China, release of November 5, 2026; TechCrunch, November 12, 2026, on the testimony. Committee · TechCrunch
18. Reuters, via HuffPost, November 12, 2026, on the congressional inquiries; Reuters via MarketScreener, November 11, 2026, on the senators’ records letter. Nov 12 · Nov 11
19. MLQ.ai, November 12, 2026, summarising the export bills and the cloud-provider licensing rule; Roll Call, January 21, 2026, on the committee vote. MLQ.ai · Roll Call
20. Bloomberg, via BNN Bloomberg, November 10, 2026, on Kistenmacher v. NVIDIA, No. 5:26-cv-13448 (N.D. Cal.). BNN Bloomberg
21. Blind, poll published November 10, 2026; Investing.com, November 11, 2026, for the sample size. Blind · Investing.com
22. CNBC analyst round-up, November 12, 2026; InvestingLive on Bernstein; Investing.com on Goldman Sachs; CNBC on Baird, November 11. Secondary reports of sell-side notes, cross-checked across outlets. CNBC · Bernstein · Goldman · Baird · MLQ.ai round-up · Craig-Hallum, via TheFly · Seoul Economic Daily on Aletheia and Mizuho
23. Yahoo Finance analyst estimates for NVDA, read November 12, 2026 at 1:58 PM ET and again after the 4:00 PM ET close, including the page’s earnings-trend table; my pull of October 26, 2026. The series is the page’s default, which it labels normalized, a non-GAAP basis. Consensus figures, labelled. Yahoo Finance
24. TSMC third-quarter 2026 earnings call, October 15, 2026; SK hynix third-quarter call, October 27, 2026; reports of Samsung (October 29) and Micron (October 16) on 2027 high-bandwidth memory. TSMC · SK hynix · Samsung · Micron
25. Third-quarter 2026 reports: Microsoft call transcript (October 28), Alphabet call (October 28), CNBC on Amazon (October 29), Meta release (October 21), CNBC on CoreWeave (November 10). Microsoft · Alphabet · Amazon · Meta · CoreWeave
26. StockAnalysis, September 28, 2026, on the repurchase authorization; a secondary report. StockAnalysis
27. Broadcom third-quarter fiscal 2026 call as summarised by Lighthouse Canton (September 3, 2026); AMD third-quarter release (November 3, 2026). Broadcom recap · AMD
28. California Business and Professions Code section 16600; California Civil Code section 3426.1. Sections 16600–16601 · Section 3426.1
29. Yahoo Finance, NVDA holders page, read November 12, 2026 (13F-derived holdings at June 30 and September 30, 2026). Yahoo Finance
30. SEC Rule 14a-19 (universal proxy); ISS United States proxy voting guidelines effective February 1, 2026. Rule 14a-19 · ISS
31. My precedent study: returns computed from Yahoo Finance daily adjusted closes against the S&P 500 through the SPY fund, and against the Swiss Market Index for Nestlé, from the close before each announcement; multiples computed from trailing reported earnings in SEC filings. The cases, dates and method are in Appendix A. Academic context: Cline, Walkling and Yore, Journal of Financial Economics (2018); Larcker and Tayan, Stanford (2016). Appointment base rates from Spencer Stuart’s 2025 study of chief-executive transitions. Gildan as reported by CNBC, May 2024. Larcker and Tayan · Cline et al. · Spencer Stuart
32. Yahoo Finance daily prices and volumes for NVDA, SMH, AMD, AVGO, TSM and SPY through the November 12, 2026 close; StockAnalysis statistics pages of November 11, 2026 for forward multiples, scaled to the November 12 close; United States Treasury ten-year yield from Yahoo Finance. My arithmetic throughout.
33. Fortune, November 11 and 12, 2026, on the board and on Ms. Kress. Board · Kress
34. Saxo options brief, November 12, 2026, on the implied move for the report. Saxo
35. My own notes: August 26, 2026; September 8, 2026; November 3, 2026; November 11, 2026.
36. The Information, reported November 10, 2026 (headline read; the article is paywalled), as summarised by 24/7 Wall St. via Yahoo Finance. Yahoo Finance
37. Exchange Act Rule 16a-2(b) on transactions after an officer or director leaves; the SEC’s summary of Rule 144 on resales by former affiliates. My reading, not legal advice. Rule 16a-2 · Rule 144
38. CBS News, November 12, 2026, on the candidates and the prediction markets; TechCrunch, November 10, 2026, on Mr. Huang’s October remarks about an internal successor. CBS News · TechCrunch
39. Jon Peddie Research, add-in board report for the second quarter of 2026 (September 9, 2026), as reported by Tom’s Hardware, September 11, 2026: Nvidia at about 90% of desktop discrete graphics-card shipments. Jon Peddie Research · Tom’s Hardware
40. Reuters, “Former Nvidia CEO Huang had a relationship with a second employee, sources say,” November 12, 2026, 2:27 PM ET, read on Reuters’ site. Reuters