Tariffs, Free Trade, Export Controls, H20 & Rare Earth Ban | BG2 w/ Bill Gurley & Brad Gerstner

Summary

• Gurley’s core call: “win the AI war” is finite-game thinking applied to an infinite game — no one he talks to believes the US can permanently block China’s AI development, and “there’s no clock that ends… no one that declares a winner.” Policy built on that premise — H20 bans, DeepSeek bans, pressuring ASML and TSMC — will produce “spectacularly problematic policy.”
• Both agree the H20 export ban was self-defeating: Nvidia designed the chip to comply with the prior rule, then got ~$12-15B of China sales banned anyway — a $5B inventory write-off and, per Bernstein, “we basically just handed the China AI market to Huawei,” clearing the decks for monopoly profits and removing CUDA competition while Huawei’s CloudMatrix 384 remains China’s GB200 competitor and Huawei is near the chip frontier.
• Gerstner’s line on where he’d draw it: no GB200s to China is defensible, H20s were a close call gone wrong — Huawei Ascend 920s already surpass the H20, China has 100 nuclear plants under construction at 1/4 US cost, so “if it comes down to power, I guess they’re going to win” (Gurley). The ban also raises the trade deficit by $15B and forfeits $2-3B of US tax receipts — “we just unilaterally disarmed.”
• The tariff regime remains two irreconcilable policies conflated: tactical reshoring of chips/pharma/rare earths (broadly supported) vs. the “nuclear Navarro” external-revenue-service vision. Gerstner still believes Trump “is fundamentally a free trader” and will walk China to 25-30% and rest-of-world to 5-10% ($200B) — but that’s still a 3-4x tariff increase vs. last year.
• Markets haven’t priced the earnings risk: S&P consensus has only slipped from $273 to $265 (15%→12% growth) while CEOs say they’re “on hold” and managers are at “max caution” — United Airlines literally gave dual guides ($7-8 recession / $12 not). Altimeter is in the “bottom third” of normal risk units, leaning back in on big risk-off days; the next leg down, if it comes, “is going to be around growth.”
• Escalation is the underpriced tail: China’s April 5 rare-earth ban (60% mined, 90% refined in China) was called “a kill shot” inside the White House orbit, and if AI is framed as existential, export bans are “firing a shot at somebody’s national security” — Gurley warns any single decision “could go as far as leading to a hot war,” and Apple or Tesla getting kicked out of China is a plausible scenario that must be discounted.
• Timing matters more than terms: Gerstner says if this same uncertainty persists 8-16 weeks, S&P earnings crack; Gurley thinks “months is too long” — reflexivity is already self-reinforcing as CEOs decommit capex and capacity, which is itself inflationary. Gold outperforming the S&P by ~40% YTD tells you where money is hiding.

Deep dive

1. Landing this plane requires winning every hand at once

• Gerstner’s opening frame, borrowed from Gurley’s complex-systems thinking: the administration is simultaneously attempting to transform global trade (negotiations with ~100 countries, an effective China embargo at 140% tariffs plus export controls), pass the tax bill before it doubles, run DOGE and deficit reduction, and manage Ukraine and a near-miss Israeli strike on Iran. “Multiply those discounted rates together… the probability that you’re going to land this plane in a really smooth landing I think is challenging.”
• Gurley’s caveat on why he hates this conversation: in complex multivariable systems “it’s very hard to know which variables are dependent… there may be one you don’t know about that flips and the whole system takes on a different shape” — even accurate macro predictions may be happenstance. But “we’re kind of forced into it” because the destiny of the tech companies they’d rather discuss is caught up in it — e.g., a circulating estimate that Meta loses ~$7B of pure-margin revenue this year from Temu-style Chinese merchant ad pullback.

2. The free-trade steelman: Reagan already litigated this

• The episode plays a ~45-year-old Reagan clip on Smoot Holly as its anchor: tariffs make industries “rely on government protection… stop competing and stop making the innovative management and technological changes they need,” then “inevitably lead to retaliation… markets shrink and collapse… millions of people lose their jobs.” Gerstner: “everything I learned up until this point is congruent with what Reagan said.”
• He cites likely Ricardo Hausmann (the trade-deficit math ignores services, IP, and foreign income from companies like Meta) and likely Niall Ferguson’s hour with likely Bari Weiss, and Gurley calls the counterfactual regret over globalization “fanciful” — comparing today to an imaginary world where America kept manufacturing. The closing version of the argument: comparative advantage “is very mathematical… win-win-win. You start backing that up and you’re going to get lose-lose-lose. I’m certain of it.”
• The likely Milton Friedman retaliation image, kept verbatim: someone shoots a hole in your boat; a retaliatory tariff is “you shooting another hole in the bottom of the boat… it just sinks faster.”

3. Tactical reshoring vs. nuclear Navarro — the conflation is the problem

• Gerstner restates his two-camps framework: the “fully loaded gun rarely discharged” tactical view vs. the “nuclear Navarro” replace-the-IRS-with-an-external-revenue-service view. Markets down 15% since April 2 “liberation day” reflect not knowing which one is policy. His complaint: “the sequencing of events is confusing — it should not be McKinley-style multi-trillion dollar tariffs if your objective is to reonshore some critical industries.”
• The hypothetical Sacks/Mearsheimer steelman gets aired — a great-power struggle means national security trumps free trade for rare earths, fabs, steel. Gurley’s rebuttal: resilience “can include other countries that are allies,” and consistency matters — people who diversified to Vietnam on the first administration’s instructions are now tariffed there too.
• Gurley pushes likely Thomas Friedman’s op-ed (“I just saw the future and it was not in America”) further than comfort allows: China is ahead in advanced manufacturing and “we need to force them to do JVs here so we can learn from them. No one’s talking that way in the administration because there’s so much vilification.”

4. Accelerate ourselves instead of picking winners

• Gerstner wants an accelerationist agenda — AI funds, tax incentives — alongside any reshoring. Gurley disagrees on the mechanism: after his regulatory-capture speech, “I can’t imagine” US democracy deploying dollars well; the AI boom has already “attracted more capital than I’ve ever seen accumulated in the history of mankind,” and Abilene and Denton, Texas superclusters got built without government.
• The exception is energy: Korea and China deliver fission nuclear at 1/4 the US price, and South Korea being a democracy undercuts the authoritarian excuse. “Figuring out why our governments become friction and mud” beats the government doing the work itself.
• Gurley’s explicitly unverified story relayed from likely Thomas Friedman: when China targets an industry like solar, it reportedly funds ~100 startups and lets them compete down to five — “a lot of fraud and grift and waste” in the losers, but the survivors are “honed by a market force.” Nobody’s mental model of authoritarian China includes the state using market competition to reach efficiency.

5. The H20 ban: punishing your national champion to help Huawei

• The sequence: China’s April 5 rare-earth ban — 60% mined but 90% refined in China, inputs to EVs, phones, and machines, called “a kill shot” by some around the White House — followed by the US ban on all Nvidia H20s to China. Gurley’s read on Nvidia: “how good must it feel… that you become a pawn in these games you had nothing to do with creating,” having designed the H20 specifically to meet the prior export rule.
• Huawei’s CloudMatrix 384 is China’s competitor to the GB200, while SemiAnalysis reportedly put Huawei at or near the chip frontier. A CNBC analyst said, “we basically just handed the China AI market to Huawei anyway,” while Bernstein analysts said the ban “clears the decks for Huawei to have monopoly profits in China” with no CUDA competition. The Boeing/Airbus analogy: reliance on those companies increased China’s incentive to develop its own technology.
• Gerstner’s position, hedged as stated: cutting-edge chips to China is a legitimate line — “I don’t think we should be sending GB200s” — but H20 was “a very close call” resolved the self-defeating way. Ascend 920 volumes are ramping; needing 2-3x the chips and power doesn’t matter when power isn’t China’s constraint. And the irony stack: $15B of H20 sales ≈ $2-3B to the US Treasury, banning them raises the China deficit by $15B, and the $5B write-off was “kind of NBD” for Nvidia’s stock — “we are effectively taxing the Chinese on a nerfed product… and we just unilaterally disarmed.”
• Both flag the demonization risk: rumors of a customer audit of Nvidia (Gurley: “I said a few episodes ago the biggest risk on Nvidia was the government — that turned out to be true”), with Gurley insisting “Nvidia and Jensen are national heroes…” and Gerstner agreeing that “they comply with the exact attributes prescribed by the US government and then after the fact we start demonizing them.”

6. Export controls are external regulation — and “internet arbitrage is undefeated”

• Gurley’s structural argument: domestic policy regularly “spectacularly fails” — produces the opposite of its intent — and export control is regulation imposed on a world we don’t control the pieces of. “Borders are porous, money finds a home.” Exhibit A: the “spectacular failure of Russian sanctions.” The hawks’ logic baffles him: “the Biden rules didn’t work, so we’re going to redouble them and see if they work again?”
• The scope test: restricting a fighter jet to one deep ally is one thing; “if you’re going to sell something in 120 countries and tell one country they can’t have it, there’s no effing way that’s going to work” — especially for goods with digital attributes. Next escalations he’s hearing: more restrictions on ASML (Dutch) and fining TSMC (Taiwanese) for building Huawei chips — grabbing “pieces of chessboard that we don’t even own.” “I won’t be surprised at all if we wake up one day and one of these companies tells us no.”
• Gurley’s geopolitical corollary: if China reaches frontier AI regardless — which even hawks concede — the winning move is the rest of the world running on US compute and US models, so “the number one thing you would not do is anger the rest of the world with a simultaneous tariff assault.” Instead the EU, Japan, Korea, Vietnam, and the Middle East are all negotiating with China, whose tone Gerstner notes is “more statesmanlike than us.”

7. There is no winning the AI war — it’s an infinite game

• Gurley on record: “almost everyone in the AI space with a microphone says the US has to win the AI war. I don’t know what that means. And if I guess as to what it means, I don’t think it’s possible.” Via Finite and Infinite Games by likely James Carse: the vitriol borrows finite-game thinking, but AI, like the stock market, has no clock and no declared winner. Even the biggest hawks admit they can’t prohibit China’s long-term AI development — and the gap is arguably shrinking (Eric Schmidt said so this week).
• His historical rhyme: the space race was justified by a belief that whoever reached space first would control a global military mesh — “that never played out,” and the AGI-controls-the-globe thesis reads the same. Plus the hypocrisy check: “it wasn’t that long ago that OpenAI borrowed or copied the innovation at Google and DeepSeek… ideas spread like the wind.” And the Godfather: “don’t hate your enemies, it clouds your judgment.”
• The slippery slope both are hearing: a DeepSeek ban preventing US hyperscalers from integrating it. Gerstner’s reaction: “oh, I guess Europe and South America will run on DeepSeek” — currently “the best performing, most open” model on his quadrant — and a ban just accelerates DeepSeek-on-Huawei optimization. Their metaphors for the whole posture: Gerstner’s swimmer looking sideways instead of ahead, and Gurley’s football analogy — “we’re trying to intentionally inhibit the other player… you’re trying to deflate the football,” which signals you’re scared and stops you training your own team.
• The shared north star they land on: tactical reshoring yes, but “shift the focus radically on accelerating our own race” — the KPI is whether American AI companies win the way they won the internet, not whether China is slowed.

8. Positioning: down 10% isn’t irrational, and earnings haven’t repriced

• The tape: S&P -10% YTD, NASDAQ -20% peak-to-trough, VIX >30, gold outperforming the S&P by ~40%, Apple -21%, Nvidia -25%, Google -20%, Tesla -40% — a whiplash from likely Stanley Druckenmiller’s “biggest shift to a pro-business administration in my 50 years” to this in weeks. Gerstner adds the tail nobody models: Apple or Tesla “getting kicked out” of China is plausible and must be discounted.
• Gerstner’s earnings math is the tradeable tension: consensus S&P EPS has barely moved, $273 → $265 (15% → 12% growth), while CEOs say “we’re on hold, we’re tightening our belts” — you can’t wipe two months of decision-making and take only a 3% cut. United’s dual guide ($7-8 recession vs. $12 not) shows companies literally can’t forecast.
• Altimeter’s book: bottom third of normal risk units, leaning back in on big risk-off days, on the thesis that Trump is fundamentally a free trader — base case rest-of-world tariffs settle at 5-10% (~$200B), China at ~25-30% (still 3-4x last year), export bans traded away. The condition: “if we’re still having this conversation in 8, 12, 16 weeks… a much bigger headwind to S&P earnings” — and the next move down is growth.
• Gurley’s skeptical close on the timeline: the India “terms of reference” breakthrough (JD Vance/Modi) still means months of hammering out, and “months is too long — the reflexivity already starts… you decommit capex, you raise prices when you decommit, the fear of it becomes the reality.”