Holman W. Jenkins Jr. launders the Nvidia spying scandal as grand strategy.
His “Business World” column in The Wall Street Journal—“Chips, China and Trump,” which appeared in the November 18, 2026 print edition—takes the Nvidia spying scandal, a conspicuously two-handed chip policy, and a White House visibly eating itself, and repackages all three as Cold War doctrine. The operation is scandal management: relabel the conduct as strategy, cartoon the critics, jersey the beneficiary. The column runs fourteen paragraphs; this column walks through it in document order and names the machinery at each station.
One disclosure up front, because the reader will hear it in the register: I wrote copy of this species for the Journal’s op-ed page from 1999 to 2007, and my retained working memory—flagged because it is not independently verifiable—contains versions of the same anxiety-to-alibi operation. I know the drawer this came out of.
So what about Donald Trump’s schizophrenic China strategy, as I called it on Saturday?
Unlike the personal computer or the internet, artificial intelligence is viscerally a national-security issue. Google’s former chairman Eric Schmidt last year warned of a possible pre-emptive attack on U.S. data centers if Beijing found itself falling behind. Keynoting a meeting this month at Bletchley Park, England, AI entrepreneur Nathan Labenz told an audience a “World War II mindset” is overtaking San Francisco—like Los Alamos in the early 1940s. Good.
— Opening, paragraphs 1–2
Frame-engineered relabeling — WSJ Editorial Technique Catalogue §4.1; Bad-Faith Catalog: Frame-engineered relabeling — operates here through a pathology word doing the opening work, with Bandura’s moral justification running directly behind it. The first move is diagnosis: the policy is “schizophrenic,” an illness rather than a decision record. Then, before a reader can hold that thought for two sentences, the column turns the subject into a war and stamps it with a one-word verdict: “Good.” Once the subject is war, everything downstream can borrow wartime necessity without being audited as policy.
Threat inflation — WSJ Editorial Technique Catalogue §4.13 — starts at the front instead of the closer, and the technocratic-credential ledger — WSJ §3.7 — supplies Schmidt and Labenz as atmosphere. The named sources are the article’s receipts for serious concern. They are not receipts for whether the chip policy strengthens the United States. The gear-train is old. Edward Bernays codified the trade on it—Propaganda (1928) systematizes methods he had run in the Creel Committee’s wartime publicity work—but the technique has not changed its basic work: borrow the moral force of the crisis and attach the desired policy to it.
The omitted relation is the actual chip decision: what is sold, what is restricted, and what disclosure or licensing rule governs the difference. The protected beneficiary is an administration that can keep strategic flexibility without settling the terms. The normalized harm is strategic drift presented as resolve. The shell game keeps the camera on China’s menace while the people who benefit from the permanent emergency stay outside the frame. This is the con’s opening bid: define the table as a war table, and every chip on it plays for keeps.
Was Jensen Huang naive about the geopolitics of his chip peddling? Was Mr. Trump blasé due to inattention to the latter’s dalliance with an alleged Chinese spy? I doubt it. For a president who distrusts his own intelligence agencies nearly as much as China’s, Mr. Huang was clearly an instrument of a certain kind of Trumpian outreach to Beijing. When asked about the Nvidia situation recently, he called it “very complex”—a formulation not often heard from his lips.
— Paragraph 3
Interrogative-form assertion — Bad-Faith Catalog: JAQing off (“just asking questions”) — operates here through two questions that perform inquiry while delivering the verdict. The alleged “dalliance” is smuggled in as given; “I doubt it” answers the question before any answer is sought; then “clearly” arrives to certify a conclusion. The catalogue’s detection signals require a presupposition and no genuine pursuit of an answer. Jenkins supplies both in the same breath.
Frame-engineered relabeling — WSJ §4.1; Bad-Faith Catalog: Frame-engineered relabeling — supplies the vocabulary: “chip peddling” makes advanced-chip commerce sound grubby; “dalliance” makes an intelligence allegation sound like an affair; “Trumpian outreach” makes the sale look like a diplomatic project. The article’s evidentiary receipt is one word—alleged. It never states what the contact was, when it happened, with whom, or what Nvidia sold in its orbit. No document in the column connects those sales to espionage. The phrase “very complex” is then converted from uncertainty into evidence of a cover.
The omitted relation is the commercial and legal chain in which Nvidia sits. The protected beneficiary is continued advanced-chip commerce, with Nvidia the concrete intermediary. The normalized harm is possible national-security exposure converted into proof of a master plan. This is the whitewash: name the scandal, refuse the facts, acquit the suspect by tone.
In fact, Mr. Trump has been practicing a conspicuously two-handed chip policy toward China, both giving and withholding. We’ll see the effect if it now blows up on his administration. We hear about White House screaming matches. On TV in recent days, Treasury Secretary Scott Bessent has obviously been in self-saving mode. The reason is a tsunami caused as much by Mr. Trump’s midterm loss of power as by the Nvidia spying scandal. Not a good sign.
— Paragraph 4
Bandura’s diffusion of responsibility — Bandura: diffusion of responsibility — operates here by spreading one discrete scandal across a crowd of causes: Trump’s midterm loss of power, White House screaming matches, Bessent’s self-saving posture, the alleged spying matter, and the general weather of policy. Once every actor contributes, no actor has to explain what the policy produced. The article’s receipt is the abrupt substitution of a disaster metaphor for a scandal-specific account.
Bandura’s euphemistic labeling — Bandura: euphemistic labeling — calls the operation “conspicuously two-handed,” a frame broad enough to absorb sales, restrictions, and any result. The WSJ §3.4 “of course”/“obviously” marker also does the work evidence would otherwise do: “obviously” carries the diagnosis of Bessent, while “we hear about” supplies no named source. The column calls this “in fact,” which is precisely when the reader should ask for the decision chain.
The omitted relation is who approved the sales, who accepted the risk, and what warning was ignored. The protected beneficiary is the administration’s freedom to keep operating without a settled account of the incident. The normalized harm is instability treated as ordinary political turbulence. This is the blame shuffle: spread the damage across enough players and no player has to repair it. Read the paragraph again and the policy becomes a cover story.
The U.S. has bigger fish to fry. A race is on and not one with any finish line except, hopefully, strategic stability in the AI age.
The U.S. and China find themselves in intricately shared possession of the highest-risk, highest-reward technology humankind perhaps has yet encountered. So far, the practical effect of U.S. policy has boiled down to keeping China behind but not too far behind.
For those who don’t like this result and favor a simpler “U.S. wins, China loses” theory of the AI future, Mr. Huang is a national-security menace. Anthropic’s Dario Amodei in January likened Mr. Huang’s China chip proselytizing to “selling nuclear weapons to North Korea.” In July, I titled a column “Anthropic vs. China”; I could as easily have titled it “Anthropic vs. Mr. Huang.”
— Paragraphs 5–7
Strawman — WSJ §4.6; Bad-Faith Catalog: Strawman — operates here through the simple opponent who wants China to lose. The mechanical check runs inside the excerpt. The column quotes Amodei objecting to China chip proselytizing by comparing it to selling nuclear weapons to North Korea. It does not show that Amodei accepted the larger “U.S. wins, China loses” frame; it parks his sentence inside that frame and lets proximity do work the quotation has not earned.
The article itself supplies the better case one paragraph earlier: the two countries share possession of the technology, and the policy is said to keep China behind without leaving it too far behind. That is not clean victory. It is managed dependence, with danger and commerce in the same circuit. The Amodei comparison supplies heat, not the missing policy analysis.
The omitted relation is interdependence. The protected beneficiary is the commercial position of Nvidia, which can keep selling while policy is called defense. The normalized harm is the belief that dependence itself proves failure. This is the bait-and-switch: reduce the whole game to China winning, then treat anyone who asks about mutual dependence as the enemy. The strawman is the oldest trick on the sheet: lose your opponent’s argument on purpose, then beat the loser.
But maybe dust off a Cold War lesson or two about what makes for strategic stability. Beijing, after all, could always go harder at stealing the core software of U.S. frontier models. It could try blackmailing or assassinating our top technologists, or stuxnetting our labs and data centers.
China hasn’t forgotten that Leonid Brezhnev tried to enlist the U.S. in a pre-emptive war to prevent China from becoming a nuclear power. It hasn’t forgotten that the chips driving America’s AI dominance today are made 90 miles offshore on an island China claims as its own.
— Paragraphs 8–9
Bandura’s advantageous comparison — Bandura: advantageous comparison — operates here through a grammar trick: set the disputed conduct beside a worse possibility and let the comparison acquit the first. Beijing might steal software; it might blackmail or assassinate technologists; it might launch a Stuxnet-style attack. The checkable things in the column are indefinite—“alleged,” “we hear”—while the invented things are confidently conditional: “could always,” “could try.” The atrocities ask nothing but a shudder and then exit without establishing that the worse thing is happening.
Distortion of consequences — Bandura: distortion of consequences — rides with it, while threat inflation — WSJ §4.13 — supplies the enlarged stakes. The Brezhnev comparison adds historical gravity, not evidence about the present chip policy. The article offers no rates, capacities, or sequence with which a reader could rank the possibilities. The omitted relation is how U.S. sales and restrictions shape Chinese incentives and options. The protected beneficiary is the administration’s freedom to keep the chip policy indefinite. The normalized harm is escalation treated as inevitable rather than as a choice with costs. This is the lesser-devil shuffle: stage a worse possibility beside the allegation and let panic do the acquitting. The con asks no odds.
I’d worry more about the invisible war already playing out between the two countries. With a wink and nod, Americans are assured their cyberwar capabilities are second to none. Let’s hope so. China’s aggressions tend to be well aired: its positioning of malware in U.S. utilities, its stealing of our F-35 plans, its cyber hacking of U.S. government personnel records.
If there’s no cyber steel in the U.S. response, the future could be bleak. Notice how the doom scenarios of serious AI thinkers have shifted in recent weeks. Now they begin with a U.S. adversary stealing a super-advanced model from a U.S. frontier lab that its creators and the U.S. government have decided is too powerful and dangerous for public release.
— Paragraphs 10–11
Expert-consensus / expert-deference framings — Bad-Faith Catalog: Expert-consensus / expert-deference framings — operates here when the column invokes “serious AI thinkers” without naming one, quoting one, or identifying the argument any of them made. The catalogue’s detection signal is exact: the failure is not merely that experts are invoked, but that there is an absence of named specialists in either direction and no engagement with specific arguments. The scenario can be serious. It cannot serve as authority until the alleged thinkers and their reasoning are visible.
Threat inflation — WSJ §4.13 — supplies the wider frame: cyberwar capabilities are “second to none,” China’s aggressions are “well aired,” and without “cyber steel” the future is “bleak.” “Cyber steel” is a vivid phrase. It is also empty unless the article tells us what the steel is and how it would work. The omitted relation is the decision to withhold a model and the criteria for that decision. The protected beneficiary is executive control over frontier-model access. The normalized harm is secrecy and public exclusion presented as the sober response to unnamed expertise. This is the appeal-to-authority scam: say “serious thinkers,” show no thinker, and make the policy look inevitable.
Unreported is whether Mr. Huang has a noncompete. Its terms would be largely unenforceable under California law anyway. One task in coming days will be making sure Mr. Huang remains on Team USA when the world starts dangling funding for his next venture.
— Paragraph 12
Frame-engineered relabeling — WSJ §4.1; Bandura’s euphemistic labeling — operates here when the column asks that Huang remain on a national team. “Team USA” converts a corporate boardroom into a national roster. The live policy question—should a chip company sell frontier compute to a strategic rival during an espionage scandal—becomes a loyalty question: will the star player keep the jersey on?
The article supplies the contradiction and then skips the relation: it says no one has reported whether Huang has a noncompete; it says the terms would be largely unenforceable under California law; and it says the next venture may depend on international trust. Then it supplies the slogan. The omitted relation is Huang’s own stake and duties, including what was sold, under what authority, and what the China business is worth. The protected beneficiary is the state’s ability to demand alignment without stating the legal or material terms. The normalized harm is coerced commercial conduct recast as patriotism. This is the laundering move: call coercion team spirit. It is also the jersey trick: put the corporate interest in the national jersey, and the audit stops at the locker room.
Getting the 70- and 80-year-olds of the U.S. House and Senate to understand any of the above might be the bigger challenge. With plenty of Trump-related scandals to investigate, Democrats might want to think through carefully how they approach the Nvidia espionage scandal. That’s something they didn’t do when making Ukraine the center of their failed 2019 Trump impeachment. The same goes for those Trump minions now deciding whether to engage in a game of all-out blame-ducking for what was, after all, a very Trumpian policy of good cop, bad cop on China chips. There’s a bigger picture that needs to be kept in mind.
America is cursed with having to navigate perhaps the most challenging technological transition in history with a notably aging, sclerotic political leadership class. An embarrassing example this week is the obsessively meme-chasing Sen. Elizabeth Warren, for whom the only thing to know about the Nvidia espionage scandal is that Mr. Huang qualifies as a billionaire. What we wouldn’t give for a president like 43-year-old JFK or 56-year-old Richard Nixon. Both at least brought a cool temperament and hardheaded realism to the job. Both, of course, were veterans of the world war.
— Paragraphs 13–14
Whataboutism — WSJ §4.17; Bad-Faith Catalog: Whataboutism — operates here when the article turns the Nvidia matter into a test of congressional conduct during the 2019 impeachment and then into a test of the age of Congress. The passage contains no evidence about the alleged espionage or the chip decision. It supplies partisan blame, a senator whom Jenkins pre-labels “meme-chasing,” and presidential age and military credentials.
Ad hominem — Bad-Faith Catalog: Ad hominem, with its circumstantial and abusive features — supplies the displacement. Remove the senator and the sentence about her motive leaves nothing about the scandal; the argument was never in the paragraph. The WSJ §4.10 common-sense/elite pivot then makes age the measure of realism, while the WSJ §4.18 identity-as-credibility move makes veterans and a cool temperament substitute for an argument about the policy now being made. The WSJ §3.5 closing-line cadence supplies the send-off: the old presidents get the final words, and the reader gets no audit of the present one.
The omitted relation is case-specific evidence. The protected beneficiary is the administration and the officials who shaped the policy. The normalized harm is the diversion of oversight into a personality contest. This is the blame-flip: make the scandal about the accusers, make oversight about age, and let nobody answer for the conduct. The hit arrives dressed as a sunset.
So here is what the column actually amounts to, taken together.
A scandal is named three times across the piece—“the Nvidia spying scandal,” “the Nvidia espionage scandal,” “the Nvidia scandal”—and never given the facts needed to test it. An incoherent policy is renamed doctrine in terms so loose that no fact could embarrass it; that is not a description of strategy but a strategy for surviving facts. The critics are sorted into two filing cabinets, the naive and the elderly. The beneficiary is issued a national jersey, and the jersey hides the money question: what was sold, under what license, what the China business is worth, and to whom. Fear does the persuasive work while responsibility stays outside the room. The reader is left with alarms but no audit, a doctrine but no decision, and a threat so enlarged that policy can hide inside it.
Concede the kernel, because the kernel is real: AI and advanced chips are national-security issues; a two-handed policy can be deliberate; strategic stability matters. The indictment is not that such a policy is impossible. It is that Jenkins supplies no record distinguishing calibration from improvisation—no terms, no licenses, no facts of the contact—and renames the improvisation doctrine precisely so the record never gets requested. The policy’s central mechanism is ambiguity, and ambiguity is doing the work that strategy claims.
Ask who the frame protects. The column answers through its silences: continued access to a contested China market, and an administration whose conduct record improves every time “policy” replaces “scandal.” The burden belongs to the column that made the categorical claim. State the facts or retract the frame. Once scandal is relabeled strategy, nobody is obliged to say what was sold, to whom, or who was contacted; the next arrangement stamped “very complex” inherits the same pass.
I will disclose the temperature once, as I do in every column: this species of copy is bitter work to read from the outside, and the reader need not credit the temperature—the excerpts above are checkable word by word. The reader is not a fool for finding the frame persuasive; the frame was built to work. This is the shell game: hold the threat in one hand and the sale in the other, sell both as prudence, and call the absence of accountability a strategy. Strip the labels. Name the money. Ask what fact could possibly count against the story. If the answer is nothing, it isn’t analysis. It’s the brochure.
— Phukher Tarlson