But to whom?

Huang sails off into the sunset.
The Cover Story Wrote Itself
The board announcement came at 7:04 a.m. Eastern on November 12, 2026, and it was four sentences long: Jensen Huang had resigned as president, CEO, and director, effective immediately. Colette Kress would serve as interim CEO while retaining her role as CFO. Harvey C. Jones would become the first board chair in the company’s history. A search for a permanent successor had begun.
No statement came from Huang. Thirty-three years, and not one sentence from the man himself.
By the open, Nvidia was down 12%. By mid-morning it had clawed back to minus six, then bled out again. Roughly $830 billion in market value had evaporated over the week — the largest single-company destruction of value in the history of the AI trade. The stock sat at its lowest level since July 2025, more than a third below its high.
The story you were told is about sex. Two affairs: one with a Chinese businesswoman named Alice Huang (no relation), who US intelligence concluded had been recruited by the Ministry of State Security, and one with Kimberly Powell, an Nvidia vice president who reported directly to him. Huang admitted the first in a companywide email on Sunday, apologized to his family, and pledged to cooperate with the board’s investigation. The second surfaced Wednesday night. The board met immediately aftewards and by Thursday morning he was gone.
That’s the official version of the story. It has the virtue of being mostly true.
It has even more virtue being completely beside the point.
What the Record Actually Shows
Start with the thing nobody has been able to prove.
Alice Huang was a top executive and managing director at Megaspeed International, a Singapore-based cloud provider, an Nvidia Cloud Partner, and a major customer. Megaspeed has been under scrutiny since at least October 2025 over suspicions that it facilitated the diversion of restricted Nvidia chips into China. A US senator asked Commerce to investigate. Singapore police opened a file. Reporters found an office at Collyer Quay with no signs of activity.
And then this: Megaspeed says it underwent on-site inspections by the Commerce Department’s Bureau of Industry and Security, Malaysia’s trade ministry, and Nvidia itself — and that those reviews found no evidence of violations. The New York Times reported that a US trade official who visited a Speedmatrix data center found boxes of unopened Nvidia chips, and that officials did not have conclusive evidence the hardware was being diverted.
That was a year ago. No charges have been filed since. The $7.5 billion figure that circulated came from two anonymous people who know the woman, and the outlet that reported it said plainly it could not verify the number.
Now line up the dates:
| Date | Event |
|---|---|
| October 2025 | Megaspeed investigation opens. Nothing conclusive emerges. |
| November 6, 2026 | The spy allegation drops. |
| November 8 | Huang admits the affair. |
| November 11 | A second affair, with a direct report. |
| November 12 | Resignation. |
The diversion suspicion sat in public view, unresolved, for thirteen months and cost him nothing. What changed in six days was the affair.
So here is the only thing anyone has actually established: the man was unfaithful to his wife and had a relationship with a subordinate. That is a firing offense at any public company. It is not a national security event.
The national security frame wasn’t the crime. It was the accelerant — the thing that turned a personnel matter into a resignation within 96 hours, and the thing that made sure nobody asked the obvious question while the building was on fire.
The Loophole Was Always the Policy
On Wednesday, Axios published an account, sourced to six senior officials, that should have been the story of the week even before the CEO resigned.
According to that reporting, President Trump authorized stronger action against companies suspected of diverting advanced Nvidia chips to China on three separate occasions — in May, in August, and again in November. Each time, the Commerce Department did not follow through.
At one point the matter reached the Situation Room. Treasury Secretary Scott Bessent confronted Commerce Secretary Howard Lutnick directly over the department’s refusal to move — specifically on Megaspeed. Defense Secretary Pete Hegseth and Secretary of State Marco Rubio sided with Bessent. Lutnick reportedly agreed to proceed, then delayed implementation anyway. Trump, per the same account, consulted Elon Musk before directing that the controls move forward. The resistance continued after the meeting.
Commerce disputes key elements of the story, including the reported confrontation.
Notice what the denial covers. It doesn’t dispute that the sales happened. It disputes who said what in the room. That’s the tell. When an agency’s defense is that the argument was less dramatic than reported, the argument happened.
By Thursday afternoon, Senator Jim Banks was on Fox News calling for Lutnick to testify before the Senate Banking Committee:
“It appears Lutnick used loopholes to allow chip sales to China to occur” — in defiance of the president’s own export controls — and “we have a lot of bipartisan questions and concerns about what happened.”
Senator John Kennedy, in usual fashion, said that Lutnick “needs to answer that in front of God and country. If I were he, I’d call a press conference this afternoon and explain what happened and why he agreed to sell ’em chips, and I would stop doing it.”
Senator Tom Cotton wants Chinese memory chipmakers added to the Entity List. Democrats on Banking and Finance are demanding document preservation. And Banks has already written the legislative fix — the AI OVERWATCH Act, which codifies the prohibitions, mandates licenses, and prioritizes American customers and allies. It has bipartisan support, including from Elizabeth Warren, and it is now folded into the Senate’s NDAA framework.
When Jim Banks (R) and Elizabeth Warren (D) are pushing the same bill, you are watching something real happen.
Now Go Back and Read the Megaspeed File
Thirteen months. On-site inspections by the Bureau of Industry and Security, Malaysia’s trade ministry, and Nvidia itself. A company statement in December 2025 asserting that those reviews found no evidence of violations. A trade official who found boxes of unopened chips and no conclusive evidence of diversion. No charges. No enforcement action. No explanation.
We were all invited to read that as an inconclusive, but It wasn’t. It was a delay of any conclusive investigation.
If the enforcement agency does not want to enforce, then an investigation is not a search for truth — it is a holding pattern. You inspect, you note the boxes, you file the memo, you wait. And the wait has a purpose: it protects the revenue while the paperwork accumulates. Thirteen months of “no conclusive evidence” is not the sound of a difficult case. It is the sound of a case nobody wants to close.
A wall doesn’t need a waiver. A toll booth does.
Someone was profiting from the controls, which I have addressed before in “War as a Racket was just the Beginning“.
What This Does to the Removal
The story gets even better. We have to consider that sex scandals and export control violations were not the real reason Huang was removed.
If Commerce was slow-walking enforcement, then the controls were not tightening. They were being held open — by the enforcement agency itself. Beijing had no reason to take out the man who was winning the export control debate, and propping up their domestic alternatives.
The better reading is at home, and it is uglier:
Huang’s removal was collateral damage in a bureaucratic war. The enforcement faction — Bessent, Hegseth, Rubio, Banks, Cotton, and yes, Warren — had spent a year trying to make the Megaspeed case and could not get Commerce to move. Then the affair handed them a body, a hearing, and a headline. They didn’t have to destroy him. They only had to let him become the story that forced the meeting.
That version requires no anonymous source about a honey trap and no speculation about smuggling. It requires only the timeline we already have: a year of nothing, six days of everything, and a senator demanding testimony the same afternoon.
The Real Fight Is Discretion Versus Statute
Which brings us to the part that actually matters, and the reason for the title.
The export control regime was built as a discretionary licensing system. Commerce decides who gets a license, which company gets a waiver, which inspection gets closed quietly, and which case gets referred for prosecution. That discretion is the whole product. It is what makes the regime negotiable, and negotiability is what makes it valuable — to the agency that grants it, to the consultants who broker it, to the lawyers who defend it, and to the companies that depend on it — and yes, for the kickbacks and corruption engendered.
A statute destroys that. The AI OVERWATCH Act doesn’t ask Commerce to be tougher. It removes the discretion — automatic prohibitions, mandatory licenses, defined categories. Once that passes, the loophole stops being a policy instrument and becomes a compliance violation. It’s no longer up to the whims of some bureaucrat.
That is the shift of power nobody is writing about. Not Washington versus Beijing. Congress versus the administrative state — over who gets to decide which choke points are real and which ones are for sale.
The loophole was never a bug in the export control regime. The loophole was the regime. Everything else was the paperwork that made it look like a wall.
The Law
There is one mechanism that explains the chip controls, the rare earth controls, the tariffs, the pharmaceutical shortage nobody has had yet, and the removal of Jensen Huang. It is worth stating once, plainly, before we walk through it five times.
Every choke point is a subsidy to the other side’s independence.
A choke point raises the price of dependence. That price is a revenue stream, and revenue streams fund substitutes. But the fatal part isn’t the money — it’s the guaranteed market. The one thing money usually cannot buy in a mature industry is a captive customer base willing to absorb the switching costs of an inferior product. A choke point hands that over for free, along with a political mandate to use it.
Which is why the modern arsenal — export controls, sanctions, tariffs, blockades — consistently produces the opposite of its stated purpose. It doesn’t deny the adversary the thing. It finances their alternative and removes their last reason to keep buying yours.
Everything that follows is that sentence with different nouns.
The Crisis
Jensen Huang enabled the AI revolution. Removing him will cripple Nvidia, at least in the short term. My first thought upon hearing that he was leaving was cui bono. I recalled what started my Substack channel. “Breaking the Spell” written on February 8th laid out Ed Dowd’s forecast for this year which included an acute crisis in China.
We do have signs that an acute crisis has now arrived in China.
The CSI 300 fell 8.5% in October — its worst month since early 2016. Ten consecutive weeks of selling, a record losing streak. The tech-heavy Star50 index is down nearly 40% from its peak. The market entered bear territory in late October, days before the Fifth Plenum.
Beijing responded by becoming the buyer. State funds put roughly $25 billion into the market in October. On November 4, nine national-team ETFs took 76.7 billion yuan — about $10.7 billion — in a single day, the second-largest inflow on record. The PBOC rolled out a $15 billion swap facility, cut the rate on buyback lending to 1% from 1.25%, and the securities regulator ordered index heavyweights to buy back their own stock.
Central Huijin raised $231 billion selling into the first-half rally and has redeployed only $43 billion since July. It has roughly $188 billion left — about 75 sessions of buying at the current pace.
That is not a market. That is a price control on the national story.
And the story is worse than the tape. Rhodium Group estimates China’s real GDP grew 2.5–3.0% in 2025 — roughly half the 5.2% the statistics bureau reported — and projects a significant decline to 1–2.5% in 2026, against the IMF’s 4.5%. Their adjustment implies the level of Chinese GDP is about 11% below official figures, concentrated in investment. Land sales, the revenue engine of every city in the country, fell from 8.7 trillion yuan in 2021 to 4.15 trillion in 2025, then dropped another 28% in the first nine months of 2026. Local government arrears to contractors run around 10 trillion yuan, about 7% of GDP. Fiscal expenditure has fallen about 6% this year — local governments are cutting exactly when they should be spending. Youth unemployment sits at 15.6%, with a record 12.7 million graduates entering a job market that created only 21 million jobs in the last five years
The one thing working is the thing the export controls were supposed to prevent. DeepSeek V4, previewed in April, was built in close collaboration with Huawei, with Ascend chips used in part of the training run. CXMT is in HBM3 mass production and shipping HBM3E samples — at 20–30% yields against SK hynix’s 70%, but shipping. China had two domestic AI chip designers in 2018; it has at least nine now. Domestic vendors took roughly 40% of China’s AI chip market in 2025 — about 1.7 million chips against Nvidia’s 4.5 million worldwide.
And on the day Nvidia lost its CEO and $830 billion in value, Chinese AI chip stocks rallied: Enflame up more than 13% to a record, MetaX up 8.8%, Cambricon up 8%, Tianpu limit-up.
Which brings us to the part that matters. The AI narrative is now load-bearing for Chinese household confidence. It is the last story Beijing has that explains why the pain is temporary. Which means the domestic chip complex is not just an industrial priority — it is the political floor under a deflationary economy.
And that means the American export controls are no longer a trade irritant. They are an attack on the only thing still holding the story together.
Node One: The Chips
Start with the original sin, because it set the template.
The controls were designed to keep China several generations behind. What they actually did was convert a commercial dependency into a national project. Beijing’s own language gives it away: “independent and controllable” AI became an explicit strategic objective in response to US restrictions.
The scoreboard, honestly read:
- Still behind. The best Chinese parts are roughly where Nvidia’s hardware was five years ago. SMIC’s 7nm yields are poor. No EUV. Memory three to four years back.
- But existing. Nine designers instead of two. 1.7 million domestic AI chips shipped in 2025. A frontier-adjacent model trained on domestic silicon. HBM shipping at one-third the yield of the incumbent — which is a business problem, not a physics problem.
The counterargument is that the controls “bought time.” That’s the strongest defense available, and it’s still a terrible trade: they bought time by handing the other side a permanent, subsidized, protected competitor — and they paid for it with the market share that funds your own R&D.
Every test of the bypass validates the bypass. You apply the choke, they reroute, and the reroute is proof the choke no longer works — which is the one result that guarantees you’ll never be able to use it again.
Now apply the law to a person.
Jensen Huang was the most consequential pro-engagement voice in American technology. A confidant of the president’s. The architect of the H20 workaround. The industry’s loudest internal critic of a regime he argued was bleeding his company to build his competitors. He is gone — and within a day, Senator Elizabeth Warren was on record saying his resignation “does not absolve him,” that he and other Nvidia executives “still owe Congress answers about the company’s compliance with U.S. export controls.“
The first political consequence of the departure is a demand for tighter enforcement, from the faction that wanted it strengthened all along.
Ask the cynical question, the one that gets you called a crank: who benefits from removing that man, this week, in that manner? Not Nvidia. Not American consumers. Not the shareholders who watched $830 billion disappear. The beneficiaries are the compliance industry, the security apparatus, and — most of all — Huawei, which just watched its principal obstacle get escorted out of the building by his own board.
You cannot prove a honey trap. You can prove the sequence, the portfolio — chip technology and export policy — and the outcome. The effect was a decapitation. Intelligence services do not spend years cultivating assets in order to make friends.
The sex is the part they let you see.
Node Two: The Rare Earths
Then Beijing did the same thing back, and proved the law from the other direction.
China’s April 2025 licensing regime on seven medium and heavy rare earths — samarium, gadolinium, terbium, dysprosium, lutetium, scandium, yttrium — has never been lifted. Not during the truce, not after it. The October 2025 expansion was suspended for a year, then extended by 24 hours before expiry on November 9, 2026, to January 10, 2027 — the same day the tariff truce ends. Every deadline in the relationship now converges on one date.
The result:
- Exports through September: 43,711 tons, down 9.6% year over year. September was 22.2% below May’s peak.
- Export value, January through August: up 53.5%.
- Yttrium oxide: about $9,144 a ton in China. About $1.15 million a ton in Europe. A 125-fold gap.
- Dysprosium roughly 20x the Chinese domestic price. Terbium 8x. Gallium 12x.
That gap is the margin Chinese producers will never collect — and the subsidy that built the competition:
- Lynas’ average selling price: A$102.1/kg, up from A$54.3/kg a year earlier. Lynas has said publicly that it benefited from China’s controls.
- MP Materials produced 1,018 tons of NdPr oxide in Q3, up 41% year over year, with a $110/kg price floor underneath it.
- Heavy rare earth separation restarted outside China for the first time in thirty years.
- The US and Australia signed an $8.5 billion critical minerals partnership. China’s share of production is forecast to fall to roughly 63% by 2030.
And note the surgical design, because it is the smartest thing Beijing has done in this whole contest: starve the oxide, export the magnet. Foreign magnet makers get a shortage of exactly the input they need; Chinese magnet makers get cheap domestic oxide and export priority. The pain lands downstream, on the customer. That’s not confusion. That’s industrial policy.
But it still proves the law. Every ton withheld financed the competitor that will eventually replace them — and the state collected the leverage while the miner ate the volume loss.
Node Three: The Tariffs
The farce.
On February 20, 2026, the Supreme Court struck down the IEEPA tariffs, 6–3, in Learning Resources v. Trump. Roberts, for the majority:
“Based on two words separated by 16 others in IEEPA — ‘regulate’ and ‘importation’ — the President asserts the independent power to impose tariffs on imports from any country, of any product, at any rate, for any amount of time. Those words cannot bear such weight.”
And: “IEEPA contains no reference to tariffs or duties.”
Within hours, Trump announced a 10% global tariff under a different statute, “over and above our normal tariffs already being charged.”
So the legal basis changed. The burden didn’t. Average US tariffs on Chinese goods had been running around 47.5%; the Section 301 and 232 architecture was never touched.
What did change was the money. The government had collected roughly $133 billion by December, against a total liability of about $166 billion. The Court of International Trade ordered universal refunds in April; more than $115 billion had been disbursed by August. Over 330,000 importers are in line.
Of the $166 billion, exactly zero dollars went to consumers. The importer paid the duty, passed part of the cost to the shopper, then collected the refund. The person who ate the markup has no standing and no paperwork. The FTC has now been asked to investigate companies that kept it.
The impact on China’s exports? It didn’t shrink. It just changed address:
| Destination | 2025 change |
|---|---|
| United States | −20% to $419.5B |
| Africa | +25.8% |
| Southeast Asia | +13.4% |
| Latin America | +7.3% |
Those three regions accounted for nearly three-quarters of China’s export growth. Low-value packages to the US fell 30.4% after de minimis closed — while low-value shipments to Southeast Asia rose 32.7% and to the EU 41.8%.
Some of that is real relocation. Much of it is transshipment — goods that get a new certificate of origin in Vietnam or Mexico and arrive in Los Angeles with a different passport. The USCC’s own bulletin conceded the reorientation “likely reflects some level of transshipment, as well as sales of unfinished goods for final assembly.” The tell is always the mirror statistic: when a country’s imports from China and its exports to the US rise in lockstep, it isn’t producing anything. It’s a forwarding mailbox.
The tariff didn’t move production out of China. It moved the paperwork.
Node Four: The Pharma
Now the node where the substitution hasn’t happened — which is why it’s the one still held hostage.
The dependence here is transitive, and that’s what makes it invisible. The US isn’t exposed to China directly. It’s exposed to India, which is exposed to China. Look at the API (Active Pharmaceutical Ingredients) flows:
- China produces roughly 40% of global API output.
- India sources about 70% of its bulk drug imports from China — and 73% of its API imports in H1 FY2026, up from 68% in FY2019. For paracetamol, penicillin, and ibuprofen, China’s share exceeds 90%.
- India supplies about 60% of US solid oral doses.
Go upstream, where the actual exposure lives. KSM’s are Key Starting Materials:
- 41% of key starting materials used in US-approved APIs are sole-sourced from China.
- China is the sole supplier of at least one KSM for 679 APIs — 37% of all APIs analyzed.
- 12% of APIs rely on China for every one of their KSMs.
- 83 of the top 100 generics have no domestic source of key ingredients.
- Of Drug Master Files filed in 2025 — the leading indicator of future supply — 44% came from China, 46% from India, and 3% from the US.
And the cleanest illustration in the entire supply chain: amoxicillin. 94% of its KSMs come from China. Synthesis depends on four KSMs, each produced almost entirely in China. The critical one is 6-aminopenicillanic acid — and 6-APA is also the base for ampicillin, dicloxacillin, nafcillin, oxacillin, and piperacillin.
One chemical. One country. Six antibiotics.
In September 2026, China expanded its export control list for drug precursor chemicals ahead of a Trump–Xi summit. Not a ban. A listing. That’s how leverage works — you don’t pull the trigger, you let the other side see the gun.
And the tariff response can’t even see the problem:
Customs origin maps only the final stage of manufacture. A key starting material made in China, an API made in Ireland, a finished dose formulated in the US — and the imported API gets the EU rate while the supply chain still runs through China.
The tariff reads the certificate. The chemistry doesn’t care.
Node Five: The Nodes Being Seized
Which brings us to the geography, where the law is being demonstrated fastest.
The Gulf. The US–Israel war with Iran opened in March 2026. China took roughly half of its crude, refined fuels, LNG, and LPG from the Middle East in 2025. The US reimposed a blockade of Iran’s shipping and ports on July 13. Iranian crude to China: 785,000 bpd in June, ~823,000 in July, 534,000 in August — and by this week, falling off a cliff. Beijing didn’t fight for Tehran. It switched supplier mix, drew down stockpiles, and accelerated electrification.
Venezuela. Maduro was seized in January. Washington now controls the proceeds from Venezuelan oil sales — which were servicing Chinese debt. Roughly a tenth of Venezuela’s $150 billion foreign debt is Chinese, with over $10 billion outstanding. CNPC and Sinopec hold entitlements to 4.4 billion barrels, the most of any foreign country. The August–September deal handed fields holding some 65 billion barrels of proven reserves to a new entity under Washington’s tutelage.
Africa. China controls over 72% of Congolese copper and cobalt mines. The American answer is the Lobito Corridor — a rail route to Angola’s Atlantic port, backed by a $550 million DFC loan and a financing capacity expanded past $140 billion, plus a December 2025 partnership giving US firms a right of first offer on designated DRC deposits.
One detail worth sitting with: China rebuilt the Benguela Railway from 2006 to 2014 for $1.83 billion. The first copper and cobalt shipment on the corridor moved under Chinese auspices. By this year, comparatively little ore has actually left Congo by rail to the US — road transport still rules. The verdict from the trade press: China looks like it might benefit yet again from U.S. policy in Africa.
Washington can seize nodes. It cannot seize the network. Each seizure teaches Beijing the same lesson: build redundancy, build substitutes, build the system that doesn’t need the node. That’s not a side effect. It’s the dominant response of the last three years.
⚔️ The Inversion
The standard line is that interdependence makes war irrational. That’s half right, and dangerously incomplete. What interdependence actually creates is mutual hostage-taking — and hostages get used.
Rare earths against advanced chips. Hormuz against Malacca. Venezuelan crude against Chinese loans. Congolese cobalt against Chinese refining. Pharmaceutical precursors against whatever the other side holds that month.
Every node is bilateral now. So each side’s de-risking is the other side’s provocation, and the escalation has no natural brake — because both sides are rationally responding to the other’s rational response.
Which brings us to the analogy everyone reaches for and almost everyone gets backwards: imperial Japan, 1941.
Japan was a resource-dependent rising power facing what it called the ABCD encirclement, capped by the July 1941 oil embargo. It imported roughly 80% of its oil from the United States. Its leadership concluded the window was closing — American rearmament was ramping, and every month of waiting made the reckoning worse. So it traded a slow decline for a fast gamble.
Where the analogy holds for China: import dependence, chokepoints at Malacca and Hormuz, a technology blockade aimed at the industrial core, a genuine sense of encirclement, a leadership whose legitimacy rests on restored status — and the same psychological mechanism. Sanctions compress time horizons. A state that believes it is being strangled stops discounting the future.
Where it breaks: China is nothing like as resource-poor as Japan was. Nuclear weapons change the calculus of a first strike entirely. And most importantly — Japan’s window was closing. China believes its window is opening. Industrial dominance, shipbuilding, EVs, batteries, and now chips. A power that thinks time is on its side doesn’t gamble. It waits.
Which inverts the analogy. If anyone is behaving like the power whose window is closing, it’s Washington — escalating controls, seizing nodes, building corridors, blockading ports, all on a compressed calendar. That’s not the behavior of a confident hegemon. It’s the behavior of one doing arithmetic on a deadline.
The Shift
So what actually shifts?
Not capability. China is still behind on the hardware frontier and will be for years. Not the balance of GDP — the gap is opening, not closing, and the official numbers are the only thing holding the narrative together.
What shifts is the ability to set terms.
For thirty years, one side set the terms because it held every choke point and the other side had no alternative. That era is over, and both governments ended it — one by applying the choke points so aggressively that it funded the alternative, the other by doing the same thing back.
The bill lands in the same place every time. The Chinese rare earth miner eats the volume loss while the state holds the switch. The American importer collects the refund while the consumer eats the markup. The Nvidia shareholder absorbs $830 billion while the compliance industry bills the hours. The domestic producer pays for the strategy; the middlemen collect.
And underneath all of it, the thing that actually runs the world isn’t a policy, a tariff, or a blockade. It’s a certificate of origin that nobody reads closely — a piece of paper asserting that a container of magnets is Vietnamese, that an API is Irish, that a chip is Malaysian, that a medicine is European.
The paper says one thing. The chemistry says another. And the distance between them is where the entire shift of power is happening.
Huang Will Be Fine — But Will Nvidia?
The Verdict Came in Nine Minutes
The stock opened down 9% and clawed most of it back before lunch.
That’s the whole story, and it’s the part nobody expected. A founder-CEO with thirty-three years, no named successor, and a scandal gets removed before the bell — and the market shrugs. Sell-side analysts had been circulating estimates of a 20–30% valuation hit from a forced exit. The market delivered 9% and a morning.
Which means the market has already answered the first question, and the answer is unflattering to everyone who spent the last week writing about how indispensable he was.
Today the market said he wasn’t.
Nvidia’s value was never the man. It was the installed base, the software moat, and the fact that leaving CUDA costs more than staying. That’s a distributed system, no longer dependent on one genius.
So the decapitation was cheap. Sit with that, because it’s the most important number of the week. If removing the most consequential CEO in American technology costs 9% and a morning, then the price of doing it is low enough that the question stops being who would do this and becomes who wouldn’t.
You don’t need a conspiracy when the operation is free.
Thirty-Three Years, and Not One Sentence
He’ll be fine. $147 billion buys a lot of insulation (we call that F U money), and he was never the type to need the office for validation.
The tragedy is different, and it’s worth saying plainly because it’s the thing that makes this a removal rather than a resignation.
Read the press release again. The board thanks him in the third person — “The Board and the company thank Jensen for his immeasurable contributions” — while Kress and Jones both get direct quotations. The founder, the man who built the company from a graphics card outfit into the most important industrial firm on earth, is the only person in the announcement who doesn’t get to speak.
Not a farewell. Not a sentence. Not even a “we wish him well.”
That’s not a resignation letter. That’s a closed-casket funeral with a press release attached.
And what he actually did is underrated even by the people who lionize him. He didn’t just make the chips. He bet the company on a market that didn’t exist yet, and he built the software layer that made the hardware impossible to leave. The chip was never the moat. The ecosystem was. That’s a twenty-year act of conviction almost nobody in American business has pulled off.
The bitter structure of it: once AI became a strategic asset, the founder stopped being an entrepreneur and became a node. And nodes get managed. The success that made him indispensable made him removable, because a company that important cannot be allowed to have a CEO with independent judgment about geopolitics.
The board didn’t remove a leader. It removed a variable.
The Shelf Was Never Empty
Here is where the standard objection arrives: he had it coming.
He did. That’s the whole point. It’s called leverage.
Everyone at that altitude has something on the shelf. The only variable is the date on the can. You don’t need to allege an operation to notice that the incentives manufacture the file automatically:
- A CEO with a relationship with a direct report has created a permanent, board-activatable liability. Not a scandal — a switch. It sits there, inert, for years.
- The relationship with Alice Huang was spotted in public in June 2024 and again in May 2025. The intelligence community reportedly briefed the president on it last year. Nothing happened. Because nothing needed to happen yet.
- What changed in November 2026 was not new information. It was new necessity — a market in crisis, a chip program that needed protecting, and a CEO whose independent judgment had become an obstacle.
Old information, new timing. That’s the tell. When the file gets used, the question is never “is it true.” It’s “why now.” And the answer is almost never in the file.
The board is the part that should get more scrutiny than it will. Boards always know. They knew about the subordinate, they knew about the Chinese executive, and they did nothing for years — until a newspaper report made the cost of not acting higher than the cost of acting. That’s not governance. That’s inventory management.
And the mechanism is self-concealing, which is why it keeps working: the removal always looks like accountability.
Accountability is what we call it when the file gets used on someone we don’t like. The shelf is never empty — the only question is the date on the can.
Or Maybe He Read the Numbers
Now the theory that actually keeps me up, and the one that makes the timing legible.
He is sixty-three. He has $147 billion. He has nothing left to prove and a company whose valuation depends on a story about demand that he, more than anyone alive, knows the texture of. The AI bubble has arguably popped and financial analysts are starting to notice all the shady accounting tricks behind it.
And earnings are next week.
Nobody resigns eight days before an earnings call they expect to be good.
Look at what the exit gives him. No farewell quote. No transition period. No advisory role. No “spend more time with family.” He is out of the building before the quarter prints, before the guide, before anyone can ask him under oath what the backlog is made of and who is paying for it. A resignation is the one move that lets a man never have to answer that question.
And the structure he built is the thing worth asking about. Somewhere around 2024, Nvidia stopped being purely a chip vendor and became its customers’ investor, landlord, and guarantor — equity stakes in the very companies placing the orders, backstops on their capacity, deals underwritten by money that originated on Nvidia’s own side of the table. That’s the circular architecture the private credit market financed, and it’s exactly the channel that cracks first when the collateral stops appreciating.
Whether that exposure sits on Nvidia’s balance sheet or merely in the architecture it built is an accounting question. The exposure is the same.
So here is the strongest version of the theory, and I think it’s right:
He didn’t retire to escape the debt hole. He retired into it. The affair was the door, and he walked through it — because the alternative was standing in front of that call in a week with $830 billion already gone and nothing to say that wouldn’t start a slide.
The two stories aren’t in competition. He was pushed and relieved. That’s what a man looks like when the pretext arrives and he decides not to fight it.
The one thing that complicates it: a voluntary exit is choreographed. The founder gets a sentence, a standing ovation, a “we thank him and wish him well.” This one has a board thanking him in the third person and a CFO doing two jobs. That’s not a retirement. That’s a termination with the paperwork filed in the correct order.
Which means the honest version is: he didn’t choose the moment, but he didn’t resist it either. And a man who doesn’t resist his own removal, eight days before the numbers, is a man who has already read them.
He didn’t leave because of the affair. He left because the affair arrived before the earnings call — and he was the only person in the building who knew which one he’d rather explain.
Now the Harder Question: Will Nvidia?
Here’s what the recovery actually tells us, and it cuts both ways.
The bullish reading: the backlog is contractual, the revenue is locked, the moat is the install base, and the market correctly priced the man at zero. If that’s right, the stock is cheap and the panic was noise.
The bearish reading: the reflex to buy bad news is what a bubble does. A stock that recovers on day one of a CEO removal tells you nothing about the backlog — it tells you there were buyers. And the buyers may be the same people who need it to be true.
The distinction that matters isn’t the quarter. It’s the guide.
Watch five things, in this order:
- The guidance, not the print. The quarter is already booked. The guide is the first statement the company makes without him in the room.
- The composition of the backlog. Contracted and funded is a different animal from “intent” and “framework agreement.” If the disclosure stays vague, that’s the answer.
- Customer concentration — and who funds the customers. If the biggest buyers are entities Nvidia has invested in, guaranteed, or backstopped, the revenue quality is a financing question wearing a sales costume.
- Whether Kress is made permanent. A finance chief running a technology company is what happens when the company becomes a balance sheet. That’s not a succession plan, it’s a signal.
- What the board’s investigation actually produces. If it closes quietly with nothing about the China business, the national security framing was the pretext — and everyone involved knew it.
And there’s a detail that should get more attention than it will: Nvidia has never had to think about succession. Thirty-three years, one CEO, and his two adult children both work at the company. He told Bloomberg they’d have to “go earn it.” Now the question is settled by a board that has never done this before, appointing a first-ever chair and an interim CEO who is also the CFO, into an earnings print, in a market that just lost $830 billion.
That’s not a company in transition. That’s a company improvising in public.
The Close
He’ll be fine. The fortune survives, the legacy survives, and in five years he’ll be vindicated in writing by people who won’t mention that they fired him for being right about the export controls.
What doesn’t survive is his reputation. Thirty-three years of building the future, and the record will remember the girlfriend.
But the real question was never about him. It’s about the thing he built and whether it was ever a chip company at all — or a financing structure that happened to manufacture silicon, whose founder knew the difference and got out while the getting was good.
The market says the man was replaceable. Next week we find out whether the backlog is.
He’ll be fine. The company is the open question — and the only person who already knows the answer just walked out eight days before the call.

Leave a Reply