The theory behind US chip export controls was straightforward: deny China access to the most advanced semiconductors and you slow its AI progress. Years of escalating restrictions on Nvidia chips, TSMC production, and advanced chip-making equipment were built on this logic. This week brought fresh evidence that the theory has a problem. SMIC, China's largest chipmaker, reported 2025 revenue of $9.3 billion, a 16% year-on-year record, with analysts projecting further growth to over $11 billion in 2026. A CNBC report noted that US export restrictions have added what one analyst described as "rocket fuel" to demand for domestic chips. The policy designed to constrain the Chinese AI ecosystem may be one of its more reliable growth drivers.
The story is not limited to chip manufacturing. Zhipu, one of China's leading AI startups and a so-called "AI tiger," reported that its revenue more than doubled in its first public earnings disclosure, sending its shares up 35%. Zhipu builds large language models and enterprise AI products, and its growth reflects a domestic market that is expanding fast despite — or partly because of — reduced access to US technology. When US software and chips are unavailable or restricted, Chinese companies have both the incentive and the government backing to build their own versions.
This is not a new dynamic. The export control playbook has a history, and it is not especially encouraging for the containment argument. Restrictions on advanced chip equipment have prompted China to accelerate investment in domestic semiconductor tooling. Restrictions on Nvidia GPUs created a market for Huawei's Ascend chips that might not otherwise exist at scale. Each new layer of controls removes a dependency but also removes a lever, and often provides the political cover for the kind of industrial policy spending that makes domestic alternatives viable.
The one-year anniversary of Trump's "liberation day" tariff announcement, which fell this week, adds another angle. A CNBC retrospective found that global investors have spent the past year rethinking their exposure to US assets as the combination of tariff uncertainty, AI valuation questions, and geopolitical volatility has made American exceptionalism feel less automatic than it once did. Capital that might have flowed into US AI infrastructure is finding its way elsewhere, including into the Chinese AI ecosystem the export controls were meant to starve of funding.
None of this means the export controls have failed entirely. China's chip industry remains years behind TSMC on process nodes, and that gap matters for training the most demanding frontier models. SMIC's record revenue is driven by mature-node chips for EVs, consumer electronics, and AI inference at older architectures, not by cutting-edge GPU production. The controls have imposed real costs and real delays. The question is whether they have imposed them on China more than on the companies and research culture that depends on global supply chains and open collaboration.
The tension in the export control strategy has always been this: semiconductors are a global industry, and any restriction that disadvantages China also, to varying degrees, disadvantages the US companies that sell to China, the research institutions that collaborate with Chinese counterparts, and the broader principle of open scientific exchange that the US AI lead was built on. Record revenue at SMIC does not resolve that tension, but it is a useful data point for anyone still confident the containment logic is working as intended.