It is true that the U.S. export controls gave Beijing its pretext. It is also true that Vice Premier Ding Xuexiang has decided to treat the country’s largest AI customers as conscripts in a campaign he has explicitly modelled on the 1960s weapons programme, and that “traitor” is the word he is reported to have used on anyone who demurs. The trouble is that the people most loudly cheering this on, in Beijing and in Washington, are reading from the same playbook of “strategic competition” that has produced the present arrangement, and what follows is about what the playbook actually contains.
It contains, first, a single buyer with no exit. The Chinese state has cut reliance on foreign AI chips from 90 percent in 2021 to under 60 percent in 2025, per Morgan Stanley data cited by The Wall Street Journal, with a stated goal of 25 percent by 2030. The mechanism is a committee under Ding drawing on the country’s best companies and labs to form specialised teams across the chip supply chain — same all-out approach that produced the first atomic bomb, hydrogen bomb and satellite in the 1960s, according to people who have seen the directive. Huawei, the leading domestic developer, presented a plan to Ding last summer and said China could become self-sufficient in some critical areas of AI within three years. Huawei’s deputy chairman Eric Xu put it plainly: “If the U.S. hadn’t forced our country, our company and our industry into a corner, we would never have done something like this.” The corner is the export controls. The conscription is the committee. The word for the conscripts who balk is “traitor.”
It contains, second, a chokepoint the campaign has not closed. Nvidia’s top AI chip carries roughly four times the computing power of Huawei’s best offering. China’s AI computing power sat at about 14 percent of the U.S. in 2025, according to Bernstein estimates. Huawei expects to ship around 1.5 million AI chips this year, doubling its 2025 volume, and aims to produce 750,000 of its latest Ascend 950 chips — alongside a national target of more than 500,000 advanced wafers a month by 2030, up from roughly 30,000 last year. To put that in terms a reader can hold: the chip the campaign is built around is the one whose competitor is four times faster, and the national plan is to ship twice as many of the slower chips this year as last, while building the wafer capacity to eventually catch up. The arithmetic is the strategy.
It contains, third, a smuggler as the unofficial distribution channel. The state initially allowed sales of a downgraded H20 chip, then privately told companies to halt purchases, then required commitments to use more domestic chips, per the Journal. The squeeze boosted domestic demand but produced a supply crunch and a boom in grey-market smuggling, where prices for illegal Nvidia chips have more than doubled over the past six months, according to Chinese distributors. The cheating and the official campaign are not in tension; they are the same campaign. A country that bans its own companies from buying the fastest available chips and calls resisters traitors does not thereby obtain the chips. It obtains, instead, an enforced relationship with smugglers. The message that everyone is supposed to learn is that state direction picks the winner. The lesson most participants actually learn is that state direction creates rents, and rents are extracted by whoever can move product through a customs barrier.
It contains, fourth, a specific twist that bears naming. Moonshot AI, a Beijing startup that had just rushed to capital markets to fund growth, unveiled a model called Kimi K3 this month that closely trails top U.S. systems, then suspended sign-ups for premium services after two days because it lacked sufficient computing capacity. The chip shortage, not the chip programme, is what is keeping a competitive Chinese model off the market. Meituan, a tech company, recently released an AI model it said was comparable to Google’s flagship and trained solely with Chinese chips — the desired outcome of the campaign, and a sign that the chip the campaign can actually produce is sufficient for some classes of model. The two stories together say the campaign is producing working models at smaller scale while denying the compute to scale them. That is not a chip story; it is a planning story.
And it contains, fifth, a specific technology bet called diaohua — fine-carving — that deserves a sentence because it is the engineering fact under all the political noise. U.S. controls cut China off from the most advanced chip-making equipment, so Huawei is using older deep-ultraviolet lithography machines and alternative designs to squeeze performance from older technology. The Journal reviewed some 200 patents published over the past two years by Huawei and its partners showing production workarounds and software, sometimes using AI, to reduce errors and boost yield. Huawei is also bundling hundreds of thousands of chips into massive clusters that function as a single computer, leaning on its networking expertise to mimic cutting-edge silicon, and has outlined a method for stacking circuits to pack more computing power into a single footprint — aiming for chips by 2031 that could outperform any semiconductor currently in production. Several analysts quoted by the Journal expressed caution. Brookings’s Kyle Chan: “It’s trying to catch up with a bullet train.” Bernstein’s Qingyuan Lin: the strategy is “more about finding a way to prevent the gap from widening quickly” than overtaking the U.S. Chris Miller, the Tufts historian: chip manufacturing is “orders and orders of magnitude more complex than almost every other type of manufacturing.” SemiAnalysis’s Ray Wang: matching extreme-ultraviolet lithography might take between a decade and fifty years.
One could quote those analysts and stop. One could also notice what the campaign’s own architects are admitting in the receipts. The atomic-bomb analogy is flattering and wrong. The bomb programme of the 1960s worked because the design was already known, the fissile material was a single engineering problem, and the bottleneck was organisational will. A modern leading-edge fab is two dozen specialised processes, a thousand pieces of equipment, and a supply chain that takes a decade and tens of billions of dollars to build — as Miller’s remark is doing the work of saying without saying. The people who put the campaign together know this, which is why “traitor” is in the briefing. It is the word for the moment the planning committee runs out of engineering options and starts running on political obedience.
Who benefits is the work of cui bono and the part the official framing most carefully obscures. The beneficiaries, in order: Huawei, which gets a captive domestic market and the legitimacy that comes with being designated the national champion at exactly the moment Eric Xu is on the record saying the company would never have built this without the corner it was forced into; the smuggler-distributor class that the squeeze creates; the senior engineers inside state institutes who become indispensable for the duration; the senior Party officials whose careers ride on the campaign’s success and whose careers also ride on the campaign’s silence; and, eventually, the Politburo Standing Committee, which gets a credible story about technological sovereignty to tell at the next plenum. The costs fall, in reverse order: on the Chinese AI startups whose models work but cannot get the compute to scale; on the Chinese academic users who lose access to the fastest hardware the world actually produces; on the smugglers who will, at some point, be the ones the campaign decides to punish for having been useful; and on the Chinese public that pays for it through the ~$48 billion semiconductor fund raised in 2024 and the salary-cap break for equipment maker Naura Technology to attract talent.
The U.S. side of this is not innocent, and the symmetric-application discipline applies here. The export controls that gave Beijing its pretext are themselves the product of a chokepoint play — using control of the lithography-and-fab supply chain to slow a competitor, on the theory that controlling the top technology means the rest of the world depends on Washington’s favour, as the Journal itself noted. What has been said here since this argument started writing about chokepoints is a framework for understanding why: competition, regulation, self-help interoperability, and labour power are the four forces that hold back the enshittification of any market, and if you remove even one of them by state fiat, the market it was holding back enshittifies. The evidence is already on the table. The smuggler’s price-doubling — illegal Nvidia chips costing twice what they cost six months ago — is the competition force being removed by state fiat. The Ding committee’s “traitor” warning to companies that prefer faster hardware is the regulation force being removed by the absence of a market check that would let buyers discriminate on engineering merit. The Moonshot K3 suspension, a working model throttled not by technical failure but by a planning committee’s allocation decision, is the proof that the removal is real and its consequences are already arriving. The four-forces diagnosis cuts both ways. The U.S. is, here, removing the competition force for itself, in the hope that the gap widens rather than closes. The campaign is, on its side, attempting to remove the regulation force for itself, by building a state-directed internal market that does not have to compete with the world’s fastest hardware. Each side is using a chokepoint, and each side’s chokepoint has the same shape: a small number of firms or committees deciding, on behalf of a much larger population, what the population is allowed to use.
The four-forces check also says what would actually work to keep the gap from closing as fast as the campaign wants, and what would actually work to let the gap close as fast as the technology allows, and neither of those is what is on offer. What would keep the gap open is the boring thing: keep exporting the best chips to commercial customers, let the market discriminate, and let Chinese AI startups choose on engineering merit. The campaign exists because someone in Beijing decided that was intolerable. What would let the gap close is the boring thing on the other side: sell the best equipment to whoever will pay for it under nonproliferation terms, let Chinese fabs build on the open commercial supply chain, and let Huawei and SMIC compete on the same equipment everyone else competes on. The export controls exist because someone in Washington decided that was intolerable. Neither side can choose the boring alternative, because each side’s political coalition depends on the chokepoint: the U.S. defence-industrial base and the Chinese Party-state’s technology-sovereignty narrative both require the controls to continue. The four-forces diagnosis therefore describes a long-run equilibrium, not a short-run policy option. Both decisions are the same mistake, made from opposite ends of a telescope.
This is not the moment to predict how the campaign ends, which is the discipline kept when the documentary record is hot — the workers will continue to ship the chips they can ship, the smugglers will continue to move the chips they can move, the senior Party officials will continue to brief the word traitor, and the four-year-olds of 2031 will be applying to universities whose curricula were rewritten by people who do not exist publicly. It is the moment to say what has been argued from the start: a chip campaign built on the word traitor is a campaign whose planners have run out of engineering answers and are reaching for political ones. The Moonshot Kimi K3 suspension after two days is the receipt. The doubling of grey-market Nvidia prices is the receipt. The 14 percent of U.S. compute is the receipt. The campaign is producing models that work, chips that lag, smugglers that prosper, and a vocabulary in which the engineers who say so are called traitors. The four forces will eventually reassert themselves, because they always do when the planning committee is the constraint, and the question worth asking now is whether Beijing wants to be the one holding the chokepoint when they do.