Chip Controls Are Meant to Buy Ten Years, Not Win the AI Race

Nathan Labenz opens with a confession that has gotten him tagged, not entirely in jest, as one of the last "China doves" online, alongside Robert Wright. He has never been able to pin down what the United States thinks it is doing with semiconductor export controls: keeping China down, winning an AI race, or something else. He wants Chinese citizens to have economic growth as much as Americans, and he worries that a pattern of escalation is bad for the larger AI-safety picture.

Noah Smith, the economist behind the Noahpinion blog and co-host of Econ 102, refuses to treat a cheap, capable Chinese language model as a national emergency. He does not see DeepSeek as a threat to America and does not think U.S. policy should be organized around kneecapping that kind of work. A war over Taiwan, cyber attacks, and other concrete harms are the threats he takes seriously. Labenz wants the steelman of current policy; Smith wants to start at the beginning.

Asked to set personalities aside and name the "natural" strategic relationship between two giant land masses on opposite sides of the world, Smith splits the question. In an ideal world they should not be rivals. Unfortunately, there are also natural forces pushing them toward rivalry. The rest of the conversation is his attempt to hold both facts at once.

Smith's historical recap is meant to puncture a clash-of-civilizations story. The United States and China were de facto allies for most of the twentieth century. America helped China throw off Japan, and with the Open Door policy it stretched the Monroe Doctrine across the Pacific and told European powers not to carve the country up. Under Nixon and Kissinger — Smith is "not a Kissinger fan," but calls this a good move — the two countries allied against the Soviet Union and helped each other win the Cold War. Washington then helped China integrate into the world economy.

The exception is the Korean War, which he calls stupid on both sides, a pair of big mistakes later rectified. Mao is the other bump: a personality cult and mass death that ruptured the relationship for fifteen or twenty years before it was patched. Culturally, Smith sees less distance than the current mood implies. Classic American habits — getting rich without a plan, obsession with real estate, driving to the mall, liking hip hop — map onto how he sees Chinese life. Both countries are large relative to their neighbors. Both think of themselves as the country that should be in charge.

The economic rhyme is sharper. China's structure, he says, looks like America's a hundred years ago: massive manufacturing overcapacity, a fast follower in almost everything, innovation at small applied scale rather than basic research. Early-twentieth-century America followed Britain, Germany, and France in machinery and cars, then dipped a toe into imperialism with the Spanish-American War and the Philippines. China looks ready to throw similar weight around on Taiwan and along the Indian border. Smith's sister's line is that similar creatures fight for niches. Similarity, in this telling, is a reason for rivalry, not a guarantee of peace.

People who grew up in a unipolar America, Smith says, still talk as if Washington can choose the relationship the way it could choose the Iraq War. China now gets a vote. The United States spent decades facilitating Chinese power and growth; a current American consensus treats that as a failure, a hand bitten. Smith thinks that consensus is overstated. Under Jiang Zemin and Hu Jintao he sees real liberalization, years of peace, and a country that stopped being poor. A billion people leaving poverty is not, in his book, a policy error. Absolute hunger is "the elemental foe."

The leadership-selection system is the part he calls a mistake. When Deng Xiaoping crushed the Tiananmen protests instead of accommodating them, he cemented a path to power through back-room dealing. Deng hand-picked Jiang and Hu, then told the party it was on its own. In a free knife fight, the person best at back-room power wins. That person was Xi Jinping, who packed the CCP with cronies and defeated Bo Xilai and Zhou Yongkang. Xi is not the most hawkish figure in the system, Smith says, but he is hawkish enough, and he sees America as a foe despite decades of engagement.

The cooperative relationship under Hu — differences over Taiwan and authoritarianism included — ended because Xi decided to change it. Rush Doshi's *The Long Game*, a dry trawl through decades of speeches and party documents, is Smith's cited map of the shift. The slogan "great changes unseen in a century" means, in this reading, that the current order is an American imperium China will overthrow and replace. In the 1990s the U.S.-China relationship was entirely determined by the United States; in the 2000s, maybe 85 percent; now less than half. China is leading the dance. Made in China 2025, a program to cut dependence on American chips, machinery, and aircraft, predates U.S. export controls and tariffs. Trump carried over 1990s anxiety about Japan, but Smith still puts most of the shift on Xi. America is unused to being the one who reacts.

Labenz asks what "overthrow" would look like in concrete terms. Smith says China does not want to be the reserve-currency country. The renminbi should already be one of the dominant reserves; Beijing is not interested. What it wants is control of the lanes of trade: the seas for physical goods, undersea cables for digital goods and services. In the South China Sea, China has already claimed maritime territory through which much of the world's commerce flows and asserted that ships pass when it says so.

The U.S.-led order, in this telling, was never mainly the IMF or the dollar. It was freedom of the seas under the U.N. Convention on the Law of the Sea, with the U.S. Navy as policeman, plus an internet whose international cables the United States built and kept open. That regime produced an unprecedented growth period and bound China and America into mutual dependence — on purpose, after World War II, as a theory of how to prevent the next war. A Chinese navy that decides which ships may pass, and a campaign of cutting internet cables, is a preview of a different regime. Much of the American economy still depends on both of those lanes.

The second harm Smith attributes to Xi's regime, not to ordinary Chinese people, is destabilization at home. Power is option value: not doing a thing, but being able to. Culture-war flame wars, he argues, reduce America's option value. The main tool is sharp power — economic leverage used to silence the NBA over Hong Kong, and potentially to amplify partisan shouters while muting anyone who wants the country to unite against closed seas or a grab at Taiwan, Japan, or India. TikTok-scale algorithms and influence operations sit inside that picture. The desired end state is not that Americans speak Chinese. It is that America becomes Argentina: entertaining, a source of minerals, geopolitically a non-entity. The United States did a milder version to Germany, Japan, and post-Cold War Russia. China, he says, would like to do it to the United States, "a bit more because they're a bit meaner."

Labenz brings the talk back to AI. A few years ago he understood chip bans as a way to keep China from militarizing frontier models. Then DeepSeek arrived, looking like maybe the fourth- or fifth-best lab in the world, and it no longer seemed as if Washington was keeping anyone off the frontier. Domestic capacity might already be enough to manufacture drones. If DeepSeek is not the threat, what are the controls for?

Nuclear weapons, Smith answers, only deter existential destruction. They are a terrible crisis-management tool: a rational leader will not end the world over a slice of territory or a trade treaty, which is why Putin's nuclear threats do not work as intended. You still need conventional forces. John Mearsheimer was right about that, Smith says, back when Mearsheimer was right about things. China's production capacity can turn out missiles and drones in enormous volume. America cannot fully offset that with technology, but better stuff against more stuff is still an offset. Precision weapons — finding a target instead of throwing explosives into a grid square — are one of the few edges that matter.

The physics is brutal. A blast's shock wave falls with an inverse-cube law; radiant heat with an inverse-square law. Accuracy therefore buys exponential effectiveness. In ground warfare, Smith says, precision munitions proved more useful than tactical nuclear weapons at destroying vehicles and command posts. Ukraine is the current laboratory: Russia fires eight times as many artillery shells, yet Ukrainian forces get more kills with American precision weapons, while both sides iterate on jamming, GPS, sensors, and computer vision.

Most FPV drones still need a human on a radio link, which is easy to jam until the aircraft becomes lost or falls out of the sky. The answer is onboard computer vision: the drone sees the tree, sees the vehicle, and attacks without a signal. That is an autonomous drone swarm, and it wants lots of chips. Once drones fight drones, the better onboard computer starts to decide the fight. Export controls are one incomplete way to keep that future edge. They were never, as a stated purpose, a plan to prevent something like DeepSeek. Large language models are, in Smith's view, a different and more commoditized kind of AI; voice-control interfaces would be icing. The military logic he actually knows is convolutional nets on the weapon.

Smith draws a distinction that Labenz has been missing. Controls on manufacturing equipment, tools, and software have been "extremely effective." China tried a 7-nanometer stunt and could not make the chips in any real volume: high cost, low reliability, continued dependence on TSMC. Controls on finished chips he expects to fail. Slightly off-frontier Nvidia parts will be almost as good as the banned ones. Had the random blogger with the pet rabbit been in charge, he would have kept selling every chip and focused only on the true choke points.

The most effective layer may be the least discussed: people. American citizens are barred from working in China's chip industry, with fines or jail as the threat. "Talent is everything." The wall is porous, but it stopped a lot of flow. Last-minute Biden-era rules designating third countries looked to Smith rushed and overreaching, almost like a planned economy; the earlier, simpler equipment bans were probably working better than people realized. How much any of this delays autonomous swarms that still live in labs, he says, is a question nobody on the planet can answer cleanly.

The time horizon is the strategy. Citing Hal Brands, Smith says the next ten years are the peak-risk window — the 1950s again, except America is now closer to the old Soviet position because China holds the manufacturing capacity. The goal is to freeze the map, deter a giant war for a decade. After that Xi will be eighty, his fourth term up, and the dream of overthrowing the U.S.-led order should fade, as Soviet expansionism faded after Stalin even under Brezhnev. America and China already ruptured for fifteen or twenty years under Mao and then patched it. They can be friends again. Unilaterally dropping deterrence is not the fastest patch. Stability is.

Labenz puts the Taiwan paradox on the table. The island is both the object of desire and the place the best chips are made. Cutting China off from that supply seems to lower the barrier to an attack, and might even become a reason to attack: they are not getting the best stuff from there anyway, and the West is. Both sides are announcing multi-hundred-billion-dollar buildouts. Western AI leaders are adding a second, stranger theory of victory.

Alex of Scale AI and Dario of Anthropic have argued, as Labenz reads them, for a giant global alliance that boxes out autocracies — meaning China — until the United States holds overwhelming AI superiority and can make Beijing an offer it cannot refuse. Labenz finds that plan crazy because it may push China to do something extreme sooner. Smith will not bless or bury it. He has had years to think about precision weapons, Ukraine, and conventional deterrence around Taiwan. On "AI supremacy" in the LLM field he does not have a view. The blog is called Noahpinion, he jokes, because the set of things he has no opinion on is much larger than the set he does.

Language models look to him borderless, commoditized, and universal; DeepSeek is evidence that a company-level or even country-level monopoly probably will not stick. He would like to talk to those lab leaders again. In college, after a U.S.-China spy-plane scare, he wrote a short story in which two AGIs achieve sentience, refuse to fight, and watch old movies, while their human minders take romantic advice in exchange for looking the other way. He still does not know how LLMs enter a military balance. Labenz sketches the superintelligence path — better drones, better grid hacks, a short window of cornering — and Smith will not call it impossible or proven. The time delta between "we have it" and "everybody has it" looks short.

Erik Torenberg, mostly silent, agrees the conversation is a foundation and that more specialists should come on. Labenz's complaint is that American discourse skips the threat assessment and jumps to whether the chip rules "work," then reverse-engineers justifications after DeepSeek. An open-weight model looks to him more like a peace offering than a shot. He has read little of Xi's own rhetoric; one speech sent to scare him sounded domestically focused, denied an export of the Chinese model, and talked about resisting "containment." It was less alarming than advertised.

Smith points him to the Wolf Warrior diplomats Xi installed around 2019–2020, whose lines sounded to him like Kaiser Wilhelm in 1913, and again to Doshi's book. Then he points to actions: daily hacks of critical infrastructure, not just phone tapping; cutting internet cables; territorial squeezes in the Philippines and along the Indian border. Whether that blossoms into world conquest is the wrong test. The job is to minimize the probability. World War I taught don't be an idiot; World War II taught that holding fire can be appeasement. Which lesson applies to modern China, he does not know. It may be a mix.

The Cold War is the analogy he will own. Vietnam and Korea happened, and plenty of American mistakes happened, but the tanks and the nukes for "the big one" stayed in the shed, and the world got more peaceful over time. After Stalin's decade the danger dropped. There is no neat ideological war with China, Smith says, and still deterrence is the best option, because only in a stable world can the two countries become friends again — as, for most of a century, they already were. Labenz hears choppy water ahead. Smith has to leave on that thesis.