Summary

Kroeber’s core reframe: this is not a Cold War and can’t be won like one. The Soviet Union peaked at ~1% of US trade; China hit 17% plus $600B+ of US corporate investment on the ground — “there’s basically no precedent for it at any point in economic history.” And 140 countries now trade more with China than with the US, so the containment-bloc playbook fails by design: China’s answer is “I will operate so that it is impossible for you to build the bloc that you want.”
China’s industrial policy works as a giant VC fund, not central planning: pick obvious sectors (EVs, solar, semis), subsidize the whole supply chain, tolerate losses for a decade, and let a cutthroat, export-disciplined, foreigner-included domestic market validate the bets. The EV proof: $200-300B in subsidies plus the 2018 decision to let Tesla build a wholly-owned Shanghai Gigafactory, which taught BYD consumer design — by 2022 it competed with Tesla “both on price and on quality.”
The most underappreciated strategic bet is electrification: China’s generating capacity is more than double the US’s, renewables alone equal the entire US grid, and “they add basically a France or UK on average every year.” Kroeber thinks Beijing is “basically right” that cheap, abundant power makes it hard for anyone to compete in manufacturing or AI — the same enabling-infrastructure logic as their 1990s “informatization” bet that used the internet to promote technological development and enhance control.
The bear case on China winning AI is fragmentation, not chips: Kroeber says “with high confidence” Beijing cannot force Huawei, ByteDance, Xiaomi, and Alibaba to pool scarce compute for High-Flyer — “you will not have a national monopoly solution” — and China’s self-enclosed walled garden is at a disadvantage in the global applications layer, where he says most AI value would accrue. Dwarkesh: this “updates me downwards on the probability of China winning in AI.”
China ≠ Japan 1989: cross-shareholding between banks and industry is illegal, so Japan’s one-balance-sheet debt-deflation trap can’t replicate; debt problems (property, LGFVs) are real but “isolated and can be dealt with one by one.” Kroeber puts total debt near 300% of GDP — pushing back on Victor Shih’s 200%-of-GDP government-debt estimate as double counting — and calls it “a tax on growth,” not a collapse trigger; the answer is a demand-side strategy and mild inflation, the late-1990s playbook.
Xi’s pivot from GDP growth as the KPI to technology-at-all-costs is a deliberate trade of growth for tech self-sufficiency, compounded by re-regulating services (Ant’s killed IPO, fintech, telehealth) — the result is deflation and an economy “on the edge of” a deflationary spiral, because hardware fetishism ignores that high-wage economies run on intangible services demand.
The tail risk is ignorance: Trump cut ~100 working-level US-China dialogues to one, the CDC channel that grew out of SARS was gone for COVID, and 300,000 Chinese students in the US face ~1,000 Americans in China. Kroeber’s endgame: “There’s no winning, there’s no losing, there’s just managing” — and panic that BYD ends the US auto industry gets a flat “Let’s get a grip, folks.”

Deep dive

1. The real objection isn’t that China gets rich — it’s how, and under what flag

Kroeber steelmans the hawk case: China aims “to produce all the manufactured goods in the world for everyone” while running ever-growing surpluses on other people’s buying power — a question of whether China gets rich by rules others can live with, not whether it gets rich at all. In welfare terms China’s rise is a positive; politically it collides with a US identity forged in “the morality play of World War II and the Cold War.”
His deeper claim: US elites have never accepted the Chinese system as legitimate — “an incredibly successful authoritarian system where the authority still calls itself communist” breaks the convergence narrative. And the cheap-goods bargain had “flies in the ointment”: the financialized 2000s economy “wound up being pretty bad for the social compact.”
Kroeber’s caveat cuts both ways — much of today’s China policy “is scapegoating of China as a way of diverting attention” from domestic redistribution failures. But with China at 20% of the global economy and a third of global manufacturing, some agreed rules of interaction are a “legitimately difficult problem.”

2. Empirically and conceptually, the Cold War frame fails

The numbers: the Soviet Union never exceeded ~1% of US trade with near-zero investment; China peaked at 17% of US trade (roughly Japan in the late 1980s and early 1990s) with $600B+ of US corporate investment generating sales far larger than US exports to China. “There’s basically no precedent for it at any point in economic history” — so a real Cold War means unwinding all of that, and nobody says how.
The counterfactual that kills regime-change fantasies: wave a wand, delete the CCP, and any successor capable of governing China would still build a strong independent military and — “as every Chinese government since the mid 19th century” — maximize domestic technological self-sufficiency. “It’s a fantasy to think that we have a problem here with just this particular regime.”
A dark footnote from Kroeber’s China trip: the median Chinese voter may be more reactionary than the government — the state spends most of its time tamping down “unfettered nationalist, highly militaristic thinking” online. A fully representative China “could be very, very difficult to deal with.”

3. No grand bargain: China won’t trade AI for batteries

Dwarkesh floats the Silicon Valley deal — cede solar/EVs/batteries to China, keep AI and semiconductors for the US. Kroeber’s rejection is categorical: Chinese elites across the imperial, republican, and communist periods repeatedly drew the same lesson — “We fell behind technologically. Therefore we must catch up technologically.” The current regime just “actualized” it. The mirror test: “Would we accept that? Absolutely not.”

4. The reverse arbitrage: invite Chinese factories in — and why Washington won’t

Kroeber’s most tradeable policy view: the US should be far more open to Chinese direct investment in EVs, green energy, and automation — “if the US is serious about revitalizing its industrial base… it is not going to happen unless you invite in the world’s leading players and have them compete. That is how China industrialized.” Dwarkesh’s kicker: if you believe China “cheated us” via forced tech transfer, running that play in reverse should be irresistible.
Why the Washington consensus blocks it: data and dual-use. “Any manufacturing process today is also a data creation machine,” and where 40 years ago dual-use tech was a sliver, “now basically everything is dual use.” Kroeber’s answer is Chinese-style regulation of inbound investment — data localization, rules of the road — achievable in principle, “very difficult in the current political environment.”

5. BYD × Tesla: the leapfrog that worked

The setup: from the early 1990s China forced 50/50 auto joint ventures expecting national champions to emerge. “Basically this failed. It failed massively” — for 25 years GM, VW, and Toyota supplied all the technology and design while local partners “sat around and clipped coupons.” By 2010-2015 China was no closer to a competitive conventional carmaker.
The pivot: leapfrog to what nobody had cracked — EVs — with $200-300B in subsidies across producers, buyers, and city bus fleets. Yet by 2018-19 BYD was “still not that exciting”: good batteries and software, “terrible at consumer design.” The catalyst was letting Tesla build a wholly-owned foreign auto plant in Shanghai in 2018; its status-symbol cars showed Chinese firms the missing piece. BYD loaded up on German designers and by 2022 competed “both on price and on quality.”
On whether the subsidies were counterfactually necessary, Kroeber reaches for the ad-man’s line: “half of it works, but you don’t know which half.” What mattered was direction and persistence — “we’re just going to keep trying different things until it works.”

6. Why this isn’t the central planning that failed everywhere else

Kroeber’s signature analogy: “Think of China as a giant VC fund that is just willing to lose huge amounts of money for a really long time on the assumption that a few of the bets will pan out.” The strategic-emerging-industries list isn’t clairvoyance — semis, automation, new materials are “on anyone’s list.” Even green energy wasn’t a wild call: John Doerr made the same bet circa 2007, wrong on returns, “totally correct in terms of this being a really big thing.”
The two disciplining mechanisms that separate it from Soviet planning or import substitution: export orientation (“you can’t rig the global economy… you have to compete on raising your game” — what “dooms all import substitution regimes,” and, pointedly, “what I think Donald Trump is trying to create in the United States”) and a cutthroat domestic market open to foreign players, unlike Japan’s protected home turf — “a crucible in which their ideas can be tested.”
The Solyndra contrast: one US loan goes bad and industrial policy is declared dead; China mapped the entire solar supply chain and rode out cycles of losses. “China actually has the right answer to that question. We had the wrong answer.”

7. China is structurally not Japan’s late-1980s trap

The root difference: China is an independent geopolitical actor with 14 land neighbors — North Korea, Russia, Pakistan, India, several nuclear — while demilitarized Japan could outsource security to the US and afford stagnation. “The incentive to get things right is much more existential in China.”
The financial architecture difference: Japan’s cross-shareholdings meant the whole economy sat on “one big balance sheet” resting on land values “completely detached from any form of reality”; when land fell 80%, debt deflation engulfed everything. China studied that and Korea’s late-1990s crisis and made it illegal for industrial companies to own banks and vice versa — the two systems are quarantined, so Japan’s macro trap “is very, very unlikely to occur.” Chinese private industry, long shut out of bank credit, finances itself from retained earnings and carries low leverage.

8. The debt fight: Kroeber vs Victor Shih

On LGFVs, Kroeber defends the original model: capitalizing future land appreciation to fund infrastructure was central-government-sponsored and initially conservative — early-2000s land-value projections that “seemed ridiculous and astronomical” proved “way, way too low.” The Frankenstein’s monster came post-2008, when commercial banks with weak underwriting handed local governments “free money.”
The numbers dispute: Shih pegs total government debt near 200% of GDP; Kroeber “respectfully disagrees,” citing a hard-to-resolve double counting problem in chained local debt, and anchors on total-economy debt around 300% of GDP — normal for developed economies, “extremely high” for a country with Brazil-level per capita income. His verdict: no financial collapse (contained system, local-currency debt), but “a tax on growth.”
The exit path he sees: demand. More domestic demand → pricing power → profits and a little inflation eroding real debt — “essentially how they got out of their last debt problem in the late 1990s.” That would “probably solve the debt problem over a decade or two.”

9. The efficiency fallacy — and the arithmetic of 1.4 billion people

Dwarkesh’s challenge: if they’re this competent, why is China still at a fifth of US per-capita income? Kroeber flips it: growing 10% a year for 20 straight years — “the fastest growth rate recorded by any economy in the history of mankind” — while economists trained on the developed-US norm commit an “efficiency fallacy.” The wasteful wall of money was the mechanism: maximal-efficiency lending rules would have meant less growth and possibly more inequality.
His book’s thesis, quoted back by Dwarkesh: for a country of China’s scale, what matters is not efficiency of resource use but effectiveness of achieving outcomes. And scale is the answer to the income gap: Taiwan and Korea moved 20-45M people up in a generation; China urbanized ~20M a year — “building the equivalent of New York plus Philadelphia plus San Francisco every single year for 20 years.” Moving 1.4B people to US incomes “just takes a really, really long time.”
The hedge that matters now: Kroeber sees no evidence growth was below potential from 2000-2020, but “a lot of evidence that it is below potential now” — China is sacrificing household catch-up for industrial-policy technology goals.

10. Xi’s supply-side fetish and the missing demand machine

Xi’s pivot: the old KPI was simply GDP — “growth will take care of itself” — but from 2015-16, and hard after Trump’s trade war and Biden’s export controls, the focus became technology self-sufficiency “at all costs.” Kroeber notes the irony that CCP leaders share Silicon Valley’s “technological fetishism”: the sincere belief that semis, robots, and green energy are the productivity engine from which all growth emanates.
His structural critique: the people making chips and robots are “a tiny fraction of the totality of the population,” and East Asian industrial models are “very materialist” — convinced physical stuff magically spills over. “The evidence that we have is that that actually doesn’t work that much.” Rich economies run on intangible services demand, and “I don’t see in China how they set up the linkage” between the high-tech core and the other 90% of the economy.

11. Re-regulating services: a real-life experiment in deflation

The mechanism, walked through case by case: the Ant Financial IPO killed in November 2020 (“China should build a financial system that resembled the United States in 2005… actually that’s a really dumb idea”), internet platforms barred from fintech, telehealth, and broader social media; finance re-constricted into state banks; healthcare privatization reined in. Kroeber concedes the share-price crashes overshot the revenue hit — “Alibaba, Tencent, they’re still great companies.”
Dwarkesh’s pushback lands: these crackdowns were about political control, not freeing resources for SMIC. Kroeber agrees — “It’s a political choice” — and names the consequence: bet everything on hardware, and “you get a persistent shortage of aggregate demand… you get deflation,” households save more, and China ends up “on the edge of” a deflationary spiral. Beijing’s recent demand rhetoric is real, but after 45 years perfecting a supply machine, “they have basically devoted no time… to figuring out how to create a demand machine.”
Kroeber’s honest deflation of the master-plan myth: “the amount of intelligent design behind that was actually quite low” — a lot of it “was essentially a random walk” where enabled things interacted, exactly like Dwarkesh’s GPUs-came-from-video-games account of AI. Credit goes to a few big directional calls, stuck with, plus willingness to cut losers fast.

12. The two enabling bets: the panopticon and the grid

Informatization (xìnxīhuà): while 1990s Western discourse called the internet “the death knell of authoritarian regimes,” Beijing bet it would both accelerate technology and enhance control — solving what has “bedeviled every Chinese administration” since the second century B.C.: knowing what people far away are thinking. “Now we have a panopticon… They won that bet.”
Electrification is “the comparable move… severely underappreciated”: from prosaic coal shortages, the logic compounded — EVs and a 20x-Japan high-speed-rail network need massive power; imports are insecure, coal is filthy, so renewables, because “you don’t have to import the sun or the wind, it’s just there.” Result: capacity more than double the US, renewables alone equal to the entire US grid, electricity at ~30% of Chinese final energy versus 15-20% typical elsewhere.
The strategic conclusion Kroeber endorses: cheap abundant electricity — plus leadership in ultra-high-voltage distribution — makes it “very difficult for anyone anywhere else in the world to compete with their manufacturing, with their AI, with their whatever.” “I think they’re basically right about that.” Getting a few all-purpose enabling technologies right beats picking winners sector by sector.

13. How China doesn’t win AI: walled gardens and no chip pooling

Dwarkesh’s bull case is blunt: talent (DeepSeek/High-Flyer near frontier), SMIC eventually producing H100-equivalents within 5-10 years, and China’s superpower — scaling — applied to power-hungry deployment. Kroeber’s counter, flagged as beyond his core competence: per Amar Bhidé’s The Venturesome Economy, value accrues in consumer-driven applications, not the LLM substrate — and China’s self-enclosed walled garden faces Western operators with “the entire rest of the world to play with.”
Against Victor Shih’s view that Beijing might throttle AI as destabilizing, Kroeber’s hunch is the opposite: they’ll rerun the internet bet — “it could be out of our control, but… we’ll probably be able to figure out how to control it because our control systems are so good.” Though DeepSeek — “a quant hedge fund guy in Hangzhou” outside the official plan — reportedly caused “a certain amount of disquiet in Beijing.”
The exchange that moves the host: could Beijing force Huawei, ByteDance, Xiaomi, and Alibaba to hand scarce chips to one lab? “We can say with high confidence that that’s not going to happen… You will not have a national monopoly solution. That almost never occurs in China.” Dwarkesh, on record: “that updates me downwards on the probability of China winning in AI” — unless the domestic chip constraint dissolves fast.

14. The communication collapse — and why containment can’t work

There’s no red telephone for AI — “It’s like saying we’re going to have a red telephone for the Industrial Revolution,” Dwarkesh offers; Kroeber agrees the nuclear balance was “a two or three dimensional problem” solvable by Schelling-style game theory, while AI is diffuse and more-than-three-dimensional. Meanwhile the plumbing is gone: Trump cut ~100 working-level dialogues to one (trade); the CDC-embedded channel that grew out of SARS was gone for COVID — “you can probably trace many, many, many deaths” to that gap.
The asymmetry Kroeber now concedes after years of resenting it: Chinese experts’ claim that “we understand America much better than you understand China” — “for years and years I basically thought this was BS… Now I think it’s just completely true.” 300,000 Chinese students in the US versus a struggle to reach 1,000 Americans in China; consulates closed (Houston/Chengdu), journalists expelled — “a massive own goal.” His EMBA classes proved the mechanism: cohorts with even superficial China exposure resisted “cartoon stereotypes”; post-COVID cohorts with none were “way more negative… dare I say, cartoonish.”
Against the extreme narratives — total domination or total collapse, both “a product of ignorance”: US manufacturing employment share is “a straight line from 1946 to today… you cannot find NAFTA, you cannot find China” — just technological progress plus a failed domestic social contract. On BYD ending the US auto industry in five years: “Let’s get a grip, folks.” And the containment strategy is structurally dead: 140 countries trade more with China than with the US, so China’s counter isn’t a rival bloc — “I will operate so that it is impossible for you to build the bloc that you want.”
The close, and the through-line: China since 1985 is “a very turbulent and very deep ocean” — capitalism “with a Leninist political carapace” atop centuries of stasis, while the US actually changed more socially. The best plausible deal from current talks: a permission structure for Chinese investment in US industry over “a year or two or three,” Chinese movement on domestic demand as zombie-company ROI deteriorates, and the acceptance that “there is no end point to this… There’s no winning, there’s no losing, there’s just managing.”

Mercedes-Benz’s Wang Xin on German Autos’ Tech Turning Point Tony Robbins | All-In Live from Miami