Why have US chip export controls failed?

Since 2018, the US has increasingly sought to restrict China’s access to the most advanced Western semiconductors and related technologies. The objective was simple and seemingly well-reasoned: to prevent Chinese computing and AI capabilities from matching those of the United States.

Almost a decade later, however, this strategy has failed to achieve its central objective. While Chinese semiconductor manufacturing still lags years behind Taiwan’s TSMC, Beijing has effectively narrowed the AI performance gap with the US, as demonstrated by the remarkable debut of DeepSeek-R1 in early 2025. According to the latest Stanford 2026 AI Index Report, the performance of leading Chinese and American AI models has since advanced almost neck and neck.

A trigger for unprecedented financial and technological mobilization

Despite preventing convergence in advanced semiconductor technology, the US geoeconomic strategy failed to prevent China from closing the AI performance gap. Worse still, there is considerable evidence suggesting that American export controls have been not only ineffective in this regard, but also counterproductive. As Beijing was gradually deprived of access to Western technologies, it only doubled down on its efforts to reduce dependence on them.

China’s ever-increasing capital injections into domestic semiconductor companies through successive National Integrated Circuit Industry Investment Funds are a direct testament to this. The first fund mobilized $22 billion in 2014, followed by $29 billion in 2019 and a record $47.5 billion in 2024, while an additional package, reportedly planned in late 2025, could raise the bar to $70 billion. Including subsidies and other industrial policy measures, Beijing’s total support for its semiconductor supply chain reached an estimated $142 billion between 2014 and 2023 alone, more than three times that of the United States.

What proved critical was that China did not simply follow the Western model of AI development. Rather than competing in a race for computational power that it could not win due to export restrictions, China was incentivized to pursue more innovative and efficient approaches, making better use of the technological capabilities and resources already at its disposal.

DeepSeek-R1 directly exemplified this effectuation approach. Working with less advanced Nvidia chips, Chinese developers focused on optimizing model architectures and training processes to overcome computational and memory constraints. Their success challenged a key assumption underpinning the US strategy: that access to ever-greater computational power was the primary route to improving AI performance.

Even more, Chinese efforts to improve the performance of domestically produced advanced chips are increasingly focused on innovative, more efficient designs rather than following the Western path of ever-greater transistor miniaturization. Beijing appears to have concluded that catching up with the West in technologies it has spent decades developing, and in which it still holds a substantial lead, may prove too costly and time-consuming. Instead, China is increasingly seeking to circumvent existing technological barriers and leapfrog Western capabilities through alternative approaches.

In May this year Huawei presented its LogicFolding technology that makes a good case for such a bypassing strategy. Rather than relying exclusively on making transistors ever smaller and packing them more densely onto a single plane, this innovative approach seeks to reorganize circuits into more compact, potentially multilayered structures, much like a multistorey building. There, circuits are brought closer together, shortening the distance electrical signals must travel and thereby improving chip performance.

The original sin: Underestimating China's talent pool

For now, it remains uncertain whether China will succeed in its strategy. Yet there is at least one compelling reason to believe it could, one that was largely overlooked in the design of US export controls: China’s vast pool of AI talent and engineering expertise. Without it, American semiconductor restrictions might have been far more effective, as China, cut off from advanced Western technologies, would have lacked the human capital necessary to develop alternative solutions.

DeepSeek and LogicFolding technology, among others, prove that this is far from being the case. In fact, Carnegie research found that over half of the world’s top-tier AI researchers received their undergraduate education in China, compared with only 13% in the US. Their weight is especially visible in global AI patenting, which is dominated by China, accounting for 74% of AI patents in 2024, up from just 13% in 2010. In fact, even US AI giants today rely disproportionately on Chinese-origin talent in their leading engineering and research positions.

Beijing thus possesses an army of highly trained semiconductor and AI specialists, engineers, and researchers capable of innovating independently of the West, whose total number likely reaches hundreds of thousands or even millions, depending on the definition. Yet this vast pool of human capital has been largely underestimated in US calculations when imposing export restrictions, which pushed China to fully mobilize its extensive financial and human resources to develop indigenous technological capabilities and circumvent Western constraints.

Hidden Factor: China's Spy Network

Recent developments have also shed light on another Chinese asset whose influence cannot be underestimated: its spy network. While Beijing’s proficiency in industrial espionage has been widely documented, only the recent Nvidia scandal has drawn Western attention to just how close Chinese intelligence operatives can get to the most consequential figures in the American AI industry, potentially offering yet another channel for circumventing US restrictions.

Jensen Huang’s extramarital affair with Alice Huang, who was allegedly recruited by China’s Ministry of State Security to spy on him, is indicative of this vulnerability, regardless of whether she succeeded in her alleged mission. However stringent US export controls may be, they can still be circumvented through informal exchanges and personal relationships, which are inherently more difficult, if not impossible, to monitor and regulate.

What could Washington have done differently?

In retrospect, the complete opposite of the actual US geoeconomic strategy could have played much more to the US’s advantage, technologically, economically, and geopolitically. Instead of pushing China into self-reliance, the US could have focused on keeping in place and even strengthening Beijing’s dependence on the Western advanced chips. That would in practice mean retaining Chinese access to them, and effectively allowing it to use them to develop its AI models.

Maintaining US firms’ dominance in China’s advanced semiconductor market would not only have slowed Beijing’s push for technological self-reliance, but also weakened incentives to develop alternative, potentially leapfrogging technologies. At the same time, it would have preserved a lucrative market for American companies like Nvidia, generating substantial revenues and capital for further R&D while reducing the need for costly US government subsidies.

Geopolitically, and perhaps most importantly, it would have preserved powerful American leverage over China for a moment of genuine crisis or even war, when comprehensive export controls could have been imposed all at once, with maximum impact and little time for Beijing to adapt. Instead, Washington gradually exposed its leverage, giving China years to adjust and accelerate its domestic capabilities, while sacrificing substantial corporate revenues and eroding one of its most potent geoeconomic instruments. In return, it may have merely accelerated the emergence of a competitive Chinese AI industry that would likely have developed anyway.

The paradoxes of weaponized dependencies in today's world

The case of the failure of US export controls against China reveals much about the broader nature of today’s geoeconomic dependencies and the consequences of their weaponization. In a world where concerns about the security of supply chains have returned to the forefront, economic dependencies, once seen as a natural consequence of specialization under the liberal free trade regime, are now increasingly perceived as potential sources of leverage, especially in relations with geopolitically unaligned countries.

Once these dependencies are exposed, let alone weaponized, the dependent party will seek to reduce them through substitution, diversification, or both. This makes strategic leverage, in many cases, a one-time weapon whose effectiveness begins to diminish after its first exposure and use. The first US export controls in 2018 already prompted China to accelerate its efforts towards technological self-reliance. Similarly, Chinese export restrictions on rare earths in 2025 are now driving intensified Western efforts to reduce dependence on Beijing. The logic works in both directions.

That is why gradual weaponization can be particularly counterproductive, as it alerts the dependent party to its vulnerabilities and gives it time to react as restrictions intensify, which was precisely the case with the US. Yet even Beijing, known for its long-term strategic thinking, risked making the same mistake with rare earths, having leveraged its dominance gradually rather than exercising its full potential at once.

Consequently, one could argue that strategic dependencies thrive in conditions of stability and predictability, when they can deepen over time, especially if they bring substantial economic benefits. From this perspective, their weaponization should be reserved for moments of genuine political necessity, such as a major crisis or war, when they could be exploited with their full force, the element of surprise, and therefore maximum effect.