← Back · ← Home · ← Back to list
The Paradox of US AI Export Controls: Effectiveness of Strategies Against China's AI Militarization and Challenges for Multilateral Governance
Executive Summary
Executive Summary (Executive Summary)
The strengthening of US AI export controls has been an ad hoc accumulation of security imperatives mixed with industrial-political motives. The case of Chinese startup Z.ai launching a model with comparable performance without any restrictions, immediately after the ban on non-US persons accessing Anthropic's top-tier models, symbolically demonstrates that the current control system is failing to substantially curb China's AI capabilities while weakening the global competitiveness of US companies. The most likely future scenario is a fragmented competitive landscape where US export controls maintain partial effectiveness while China's indigenous development accelerates in parallel, solidifying a complex structure where no single strategy can simultaneously resolve all risks. Consequently, a strategic shift is urgently needed for the US government to move beyond ad hoc access restrictions and establish a risk-based, tiered regulatory framework, linked with a multilateral AI governance framework with allies. Allied governments, in turn, must pursue a dual strategy of actively building their own AI sovereign capabilities, rather than passively integrating into the US-led system. For global corporations, the key survival strategy will be to accept geopolitical fragmentation as an irreversible structural change and proactively build resilient business models adaptable to any scenario through supply chain diversification and technology portfolio restructuring.
Step 1: Issue Situation Analysis
Analysis of the Issue Situation: US AI Model Export Controls and China's Response to AI Militarization
1. Background and Progress of the Issue
Artificial intelligence (AI) has rapidly emerged as a strategic asset that determines national security, military power, and economic hegemony, extending beyond the realm of mere technological innovation. Historically, each technological revolution has fundamentally altered the nature of geopolitical competition, and the AI era is no different, with the ability to learn, innovate, and adapt more rapidly becoming a key variable of national competitiveness, in addition to traditional power indicators such as territory, population, and military strength [5]. In this context, the United States and China have engaged in all-around competition over AI technology leadership, encompassing semiconductor supply chains, model development capabilities, military applications, and international standard-setting.
The US has progressively strengthened export controls based on concerns that AI technology could be directly utilized to enhance China's military and security capabilities. Initially, the focus was on restricting exports of advanced semiconductors, particularly high-performance GPUs from Nvidia. Subsequently, the scope expanded to include access controls on AI models themselves. The US government, through executive orders, has required AI developers to provide early access to the government before releasing frontier models, and based on this, has begun earnest negotiations to establish voluntary standards for model release criteria and the scope of foreign access [2]. In this process, analyses have suggested that some key players in the US AI industry intentionally highlighted the conflict with China to create a regulatory environment favorable to themselves. It is also noteworthy that while AI competition in the US is unfolding in a 'winner-takes-all' dynamic, this characteristic is relatively less pronounced in Chinese policy documents [1].
2. Current Situation (Latest Trends)
Currently, the US government is implementing measures to block foreign access to the most powerful AI models on grounds of national security. Specifically, the Trump administration directed a ban on non-US persons accessing Anthropic's top-tier models, Fable 5 and Mythos 5 [7], stemming from an intention to block high-risk capabilities that could be weaponized, such as the detection of cybersecurity vulnerabilities. However, mere days after the implementation of this measure, the emergence of a Chinese startup, Z.ai, releasing an AI model with performance close to Anthropic's models, and at a competitive price, with no access restrictions, has amplified doubts about the effectiveness of export controls [9]. Chinese cybersecurity firm 360 has also unveiled an AI tool, 'Tulongfeng,' which it claims rivals Mythos [12], suggesting that US control measures are paradoxically accelerating China's indigenous development.
In terms of semiconductor export controls, smuggling issues are also becoming prominent. With the explosive growth in demand for Nvidia AI servers from China, Chinese companies such as Alibaba and Tencent are showing willingness to pay almost any price, and smuggling activities are surfacing across the entire AI server supply chain [13]. Consequently, Nvidia is responding by further strengthening inspections of AI servers, but complete blockage is practically difficult due to the complexity of the supply chain. Meanwhile, the landscape of the large language model market itself is rapidly changing. According to the 2026 AI Industry Report, OpenAI's ChatGPT's market share in the global AI assistant market has fallen below 50% for the first time [4], indicating that numerous competitors, including those in China, are rapidly closing the gap.
China is also taking proactive steps in the competition for technical standards. China has announced seven national standards for the interoperability of AI agents, establishing an integrated framework for how AI agents identify, discover, collaborate with, and utilize external tools [15]. This is interpreted as part of China's strategy to strengthen its technological sovereignty by preemptively establishing its own AI ecosystem standards to counter export controls.
3. Key Actors and Their Positions/Interests
The US Governmentperceives the military application of AI technology as the greatest threat and utilizes export controls and model access restrictions as key instruments. However, it faces criticism that ad hoc and piecemeal measures may instead stimulate China's indigenous development and weaken the trust of allies [3]. The core challenge for the US strategy is to ensure coherence between the two goals of maintaining technological superiority and strengthening allied cooperation.
US AI Companies (Anthropic, OpenAI, etc.)hold complex positions caught between the government's tightening regulations and their own commercial interests. Anthropic has blocked foreign access to its top-tier models in compliance with the government's export restriction order [7], but this is leading to a loss of global market share. Criticisms that some AI companies exaggerated the threat from China to create a regulatory environment favorable to themselves suggest that their interests may not align solely with national security logic [1].
The Chinese Government and AI Companiesare leveraging US export controls as a justification for strengthening their technological sovereignty and accelerating the development of an independent AI ecosystem. China plans to invest approximately 2 trillion yuan ($295 billion) in AI data center infrastructure over the next five years [17], which is part of a systematic strategy toward long-term technological independence. Furthermore, China has implemented a reverse regulation, State Council Order No. 837, to strengthen state control over technology transfer during overseas investments [6]. Chinese AI companies are expanding their global presence by successively releasing open-source models that match the performance of US models at a lower cost [9][12].
Allies and Third Countriesare seeking to secure their own AI sovereignty amidst the US-China AI cold war, while simultaneously pursuing strategic autonomy to avoid subordination to either side. Canada and France have declared their commitment to strengthening their indigenous AI capabilities [4], and India is using the direct impact of US access restrictions as an opportunity to redefine its own AI sovereignty strategy [7]. The launch of Fugu, an independent model by Tokyo-based startup Sakana AI, can be understood in the same context [12]. Global South countries are at risk of becoming dependent on either the US or Chinese AI ecosystem due to the AI accessibility gap and lack of regulatory capacity [16], and the need for non-aligned AI cooperation frameworks for them is also being raised [8].
Global Corporate Usersare accelerating their shift to Chinese open-source models as an alternative to the US government's tightening export controls. According to a UBS report, there is a clear trend of companies switching from advanced closed-source models to cost-effective Chinese open-source models, such as Alibaba's Qwen series, for tasks that do not require them [10]. This demonstrates that US export controls are inadvertently helping Chinese AI companies expand their global market presence.
4. Summary of Key Issues
First, the effectiveness of export controlsis the most critical issue. While the US is blocking access to top-tier AI models from Anthropic, Chinese companies are releasing models with similar performance, raising fundamental questions about whether control measures genuinely contribute to maintaining the technological gap [9][12]. As warned by PIIE, ad hoc measures may instead stimulate China's indigenous development, leading to unfavorable outcomes for the US in the long run.
Second, the uncontrollability of open-source modelsreveals the structural limitations of the export control system. If Chinese companies release high-performance models in open-source form, any US access restriction measures will have little practical effect. The trend of global companies switching to Chinese open-source models for cost savings clearly illustrates the severity of this problem [10].
Third, the issue of alignment with alliesis significant. The US's unilateral and abrupt access restriction measures infringe upon the AI access rights of its allies, stimulating them to pursue independent AI sovereignty paths [4][7]. The proposed establishment of a free world AI computing alliance by the Carnegie Endowment is an alternative approach to address these issues, but coordinating the interests of allied nations itself presents another challenge.
Fourth, China's AI militarization and the competition for technical standardsare critical. The AI innovation activities of China's People's Liberation Army Strategic Support Force suggest that the military application of AI is already considerably advanced. China's preemptive announcement of national standards for AI agents reflects a strategic intention to expand its influence in the global AI ecosystem by securing technical standards first. The attempt to secure superiority in wireless infrastructure through preempting 6G spectrum is also an important issue intertwined with AI militarization [14].
Fifth, the absence of global AI governanceis evident. As the US and China build their respective technological ecosystems and strengthen export controls and technology transfer restrictions, Global South countries are at risk of becoming dependent on one side due to the AI accessibility gap and lack of regulatory capacity [16][8]. As the preliminary report of the UN Independent International Panel of Scientists points out, the unequal distribution of AI benefits and the lack of regulatory capacity in developing countries remain fundamental challenges for AI governance.
Step 2: In-depth Issue Analysis
In-depth Analysis of the Issue: US AI Model Export Controls and China's Response to AI Militarization
1. Analysis of the Root Causes of the Issue
The US's strengthening of AI export controls ostensibly aims to prevent technology leakage, but at its core lies a more complex structural anxiety. The most critical root cause is the shift in perception that AI has transcended the category of mere commercial software to become an element of national security infrastructure. AI encompasses a wide range of military applications, including automated cyberattacks, intelligence gathering and analysis, autonomous weapon systems, and support for military decision-making. The US government particularly believes that if frontier AI models capable of automatically detecting software vulnerabilities and exploiting them fall into the hands of adversarial forces, a non-symmetric security threat could materialize [20]. This perception has led to measures such as the ban on non-US persons accessing top-tier models like Anthropic's Fable 5 and Mythos 5 [7].
However, delving deeper into this root cause reveals that the US export control policy cannot be explained solely by security logic. An analysis scheduled for publication in a Cambridge University journal suggests that some key players in the US AI industry have intentionally highlighted the technological conflict with China to elicit a regulatory environment and government support favorable to themselves [1]. In other words, the strengthening of export controls stems not only from pure security imperatives but also from the industrial-political motives of US AI companies seeking to institutionally solidify their competitive advantage within a 'winner-takes-all' dynamic [1]. In this regard, export controls function as both security policy and industrial policy, operating in a mutually reinforcing manner.
Another root cause stems from the inherent limitations of control due to the intrinsic nature of AI technology. Unlike hardware with a physical form, such as semiconductors, AI models are digital information that can be easily replicated and disseminated, and their boundaries can be circumvented through the open-source ecosystem. Indeed, immediately after the US blocked access to Anthropic's models, Chinese startup Z.ai released a model with comparable performance without any access restrictions [9], and Chinese cybersecurity firm 360 unveiled 'Tulongfeng,' an AI tool claimed to rival Mythos [12]. This demonstrates that US control measures are not substantially curbing China's indigenous development capabilities and are instead paradoxically creating an opening for Chinese companies to freely penetrate the global market.
2. Structural Context
Political Structure
At the political level, the US's AI export controls reveal a structural vulnerability where ad hoc measures led by the executive branch accumulate without a medium- to long-term strategic framework. The Trump administration, through executive orders, required AI developers to provide early government access before releasing frontier models and pursued the establishment of voluntary standards for model release criteria and the scope of foreign access [2]. However, this process, relying on executive orders without a legislative basis, makes policy consistency and predictability fragile. Furthermore, the dual relationship where the US government treats private companies like Anthropic and OpenAI as both regulated entities and partners in national security further complicates policy design [3].
In contrast, China is pursuing a long-term and systematic state-led AI strategy. Through State Council Order No. 837, effective July 1, 2025, China has established a system where the state strictly controls the scope of technologies, dispatched personnel, and shareable know-how that Chinese companies can transfer during overseas investments [6]. This indicates that China is strengthening its institutional safeguards for preventing technology leakage and protecting strategic technologies, moving in a direction that counters US export controls while simultaneously enhancing its own technological sovereignty. Moreover, China is accelerating the standardization and interoperability of its AI ecosystem by announcing seven national standards for AI agent interoperability [15]. Preempting technical standards is not merely a technical issue but also a political act that determines the future governance structure of the global AI infrastructure.
Economic Structure
At the economic level, US export controls entail a structural dilemma that self-limits the global competitiveness of its own AI industry. While the US blocks foreign access to top-tier AI models, Chinese open-source models are penetrating the global market without restrictions. According to an analysis by Swiss UBS Securities, there is a clear trend of global companies switching from high-performance closed-source models to cost-effective Chinese open-source models, such as Alibaba's Qwen series, for tasks that do not require them [10]. This implies that US control measures are paradoxically facilitating the expansion of Chinese AI models' global market share.
China's scale of AI investment is also a crucial variable for understanding the economic structure. China is pursuing an ambitious plan to invest approximately 2 trillion yuan (about $295 billion) in AI data center infrastructure over the next five years [17], which is not merely corporate investment but strategic infrastructure development at the national level. Such large-scale investment aligns with China's long-term strategy to achieve self-sufficiency in AI computing infrastructure and gradually reduce its dependence on US semiconductor export controls. Indeed, the fact that smuggling activities are spreading throughout the supply chain due to the explosive increase in demand for Nvidia AI servers from China [13] demonstrates that export controls are not suppressing China's desire for technological access but are instead creating side effects that invigorate the underground economy.
Security Structure
At the security level, the core of this issue lies in the military application potential of AI and the asymmetric response structure to it. The AI innovation activities of China's People's Liberation Army Strategic Support Force are a focus of intense analysis within the US security community, and the integration of AI into core elements of military operations, such as cyber warfare, electronic warfare, information warfare, and autonomous weapon systems, is rapidly increasing. China's AI-based cybersecurity capabilities are also rapidly improving, leading to assessments that they are eroding US cyber superiority [11]. In this security environment, US export controls serve as a means to block short-term threats, but they possess a structural limitation where their effectiveness is bound to rapidly diminish over time, considering the speed at which China is developing its indigenous capabilities.
The competition for 6G wireless infrastructure also forms an important layer of the AI security structure. Paul Rosenzweig, former Deputy Assistant Secretary for Policy at the US Department of Homeland Security, warned that if the US falls behind China in 6G spectrum allocation, its mobile AI competitiveness could be severely weakened [14]. This indicates that the competition for AI hegemony is expanding beyond model development and semiconductors to include wireless infrastructure. China is establishing early superiority in the 6G field through long-term planning and systematic investment, suggesting that the convergence of AI and communication infrastructure will be a key foundation of future military power [14].
3. Comparative Analysis of Historical Precedents and Similar Cases
Lessons from COCOM during the Cold War
The US AI export control regime shares structural similarities with the COCOM (Coordinating Committee for Multilateral Export Controls) system, which controlled technology exports to the Soviet Union during the Cold War. Established in 1949, COCOM was a multilateral system where Western allies jointly controlled the export of strategic technologies to the Soviet bloc. Two key lessons emerge from COCOM. First, the effectiveness of export controls is fundamentally limited without the joint participation of allies. Even if the US prohibits the export of a specific technology, its impact is halved if allies export the same technology. In the current context of AI export controls, the fact that allied companies are developing similar capabilities outside the scope of US control, as exemplified by Japan's Sakana AI releasing a model comparable to Mythos, Fugu [12], reaffirms this lesson. Second, COCOM did not completely suppress the Soviet Union's indigenous technological development; instead, it spurred the Soviet Union to strengthen its self-reliance in certain technological fields. This finding aligns precisely with analyses suggesting that current US AI control measures are paradoxically accelerating China's indigenous development.
Precedent of Semiconductor Export Controls: The Case of the Japanese Semiconductor Industry
The US pressure on the Japanese semiconductor industry in the 1980s and the Toshiba incident in 1987 offer another relevant precedent. The Toshiba incident, involving the illegal export of precision machine tools capable of military diversion to the Soviet Union, exposed the vulnerability of technology leakage channels through allied companies, leading to the strengthening of export control systems. However, this case also demonstrated that export controls can create tensions in alliances and incentivize the target country to accelerate its indigenous development. The current situation where India, facing access restrictions to Anthropic's models due to US export control measures, is re-recognizing the need to strengthen its own AI sovereignty [7] suggests that this historical pattern is repeating.
Comparison with the Nuclear Non-Proliferation Regime
The Nuclear Non-Proliferation Treaty (NPT) and the Nuclear Suppliers Group (NSG) represent another historical precedent for controlling the proliferation of dual-use technologies. The nuclear non-proliferation regime shares a similar objective with AI export controls in that it established international agreements and verification systems to prevent the military diversion of technology. However, fundamental differences exist between nuclear technology and AI technology. Nuclear technology requires highly specialized physical infrastructure and rare materials, whereas AI technology, being digital information, is inherently easy to replicate and disseminate. This difference implies that applying nuclear non-proliferation control methods to AI will inevitably have structural limitations. The Carnegie Endowment's proposal for a coalition of free-world AI computing can be interpreted as an attempt to address these limitations and enhance control effectiveness through a multilateral approach.
Precedent of Internet Governance Fragmentation
The US-China conflict surrounding internet governance since the 2010s holds significant implications as a precedent for AI standards competition. Just as China has advocated for 'cyber sovereignty' and moved towards separating its domestic internet ecosystem from global standards, it is now moving towards establishing its own national standards and an interoperability framework for AI agents [15]. This increases the likelihood of the AI ecosystem fragmenting into a US-led global standards system and a China-led independent standards system, similar to the internet. In such a scenario, the effectiveness of export controls would be further weakened.
4. Key Variables in Issue Development
Variable 1: Speed of Open-Source AI Model Proliferation
One of the most decisive variables in the future development of this issue will be the speed at which open-source AI models improve in performance. ChatGPT's global market share as an AI assistant has fallen below 50% for the first time [4], and the large model market is rapidly shifting from a monopoly by a single giant corporation to a multi-player competitive landscape. The faster Chinese open-source models achieve performance levels close to US closed-source frontier models, the more meaningless US export controls will become. In particular, if the trend of global companies switching to Chinese open-source models for cost savings accelerates [10], US control measures could end up merely weakening the global market position of US AI companies.
Variable 2: Policy Coherence among Allied Nations
Policy coherence among key allies is essential for the effectiveness of US export controls. However, Western countries such as Canada and France have declared their intention to build independent AI capabilities [4], and India is seeking to strengthen its AI sovereignty due to access restrictions imposed by US control measures [7]. The emergence of calls for a non-aligned movement in an AI Cold War [8] suggests that global AI governance may not converge into a single US-led system. Whether the coalition of free-world AI computing proposed by the Carnegie Endowment can be concretely realized, and how allied nations' interests will be coordinated in that process, will be key variables.
Variable 3: Speed of China's Semiconductor Self-Sufficiency
The ultimate effectiveness of US export controls hinges on when China achieves the capability to domestically produce advanced semiconductors. Currently, China is resorting to smuggling to acquire Nvidia AI servers [13], demonstrating its continued high dependence. However, if China continues its efforts toward semiconductor self-sufficiency alongside a 2 trillion yuan investment in AI data centers [17], US hardware export controls will lose their efficacy over time. This variable is the key factor determining the 'shelf life' of export controls.
Variable 4: Speed and Transparency of AI Militarization
The speed of the Chinese People's Liberation Army's AI militarization and its level of transparency are also important variables in the issue's development. As the military application of AI accelerates and its reality becomes clearer, pressure for stronger export controls from the US and its allies will increase. Conversely, if the reality of AI militarization proves to be unclear or exaggerated, the justification for export controls will weaken, and industry backlash may grow. This variable determines the political sustainability of export control policies.
Variable 5: Outcome of Global AI Standards Competition
Finally, the outcome of the competition surrounding AI technology standards is a crucial variable. If China successfully establishes its own national standards for AI agents [15] and promotes them internationally, the effectiveness of US export controls will be further weakened by the fragmentation of the standards system itself. Conversely, if the US, along with its allies, preempts the establishment of AI technology standards and governance systems, more effective technological hegemony can be maintained through the synergy of export controls and standard setting. The outcome of this standards competition will be the most critical structural variable determining the landscape of the global AI ecosystem over the next decade.
Phase 3: Scenario Analysis
US AI Model Export Controls and China's Response to AI Militarization: A Scenario Analysis
Introduction: Premises of Scenario Analysis
The current US-China AI technology competition exhibits a complex system structure where three key variables—the effectiveness of export controls, China's indigenous development capabilities, and allied cooperation coherence—interact. Whether US AI export controls can effectively curb China's military use of AI, or paradoxically accelerate China's self-reliance, remains an open question. The following presents optimistic, baseline, and pessimistic scenarios to systematically organize this uncertainty, followed by an analysis of the ripple effects of each scenario on the global economy and industry. The probability estimates for each scenario are based on currently available public information and structural analysis, assuming they are subject to fluid change as events unfold.
1. Optimistic Scenario (Estimated Probability: 15-20%)
Development Trajectory
The optimistic scenario unfolds with the US recognizing the limitations of ad-hoc export controls early on and establishing a structural and coherent AI governance system through allied cooperation. Specifically, it envisions a path where the US substantially institutionalizes the 'Coalition for Free World AI Computing' concept proposed by the Carnegie Endowment, forming a multilateral framework where democratic allies adopt common export control standards and AI safety standards. In this process, the US moves beyond piecemeal administrative orders to establish a medium- to long-term AI strategy with a legislative foundation, introducing a differentiated regulatory system that precisely categorizes AI models by risk, imposing strict controls on high-risk military diversion capabilities while allowing flexible approaches for commercial applications [2]. Concurrently, key allies such as Japan, South Korea, and the EU adopt positions aligned with the US on semiconductor supply chains and AI model access criteria, effectively blocking China's path to acquiring advanced AI capabilities through circumvention.
In this scenario, China's AI militarization significantly slows down due to technological bottlenecks. The illicit trade routes for advanced semiconductors essential for high-performance AI training are effectively blocked by enhanced supply chain surveillance by Nvidia and cooperative enforcement by allies [13]. The AI innovation activities of China's Strategic Support Force face constraints due to a lack of computing resources. Furthermore, as China's indigenously developed AI models show limitations in narrowing the performance gap, the US's technological superiority is maintained in military and security domains.
Impact of the Optimistic Scenario
If this scenario materializes, the US AI industry, while accepting short-term regulatory burdens, will benefit in the long run by solidifying its position as the global AI standards setter. As allies are integrated into the US-led AI governance system, US AI companies like OpenAI, Anthropic, and Google DeepMind will be able to utilize the entire democratic bloc as a de facto domestic market. Conversely, Chinese AI companies will face structural disadvantages due to restricted access to the global market, potentially leading to the isolation of China's AI industry ecosystem within its domestic sphere. However, the low probability assigned to this scenario stems from three persistent structural barriers: conflicting interests between US industry and government, differing economic interests among allies, and the inherent diffusion potential of AI technology [1][3].
2. Baseline Scenario (Estimated Probability: 50-55%)
Development Trajectory
The baseline scenario involves the gradual strengthening of US export controls, which maintain their current ad-hoc nature, but their effectiveness remains limited. The US government continues to add individual measures, such as prohibiting non-US persons from accessing Anthropic's Fable 5 and Mythos 5, and moves towards institutionalizing voluntary model disclosure standards [2][7]. However, these measures operate at the administrative order level without a legislative foundation, failing to overcome structural limitations. Cooperation with allies progresses partially but does not achieve full coherence due to differences in economic interests, as seen in semiconductor export controls.
In this scenario, China continues to strengthen its indigenous AI development capabilities in response to US control measures. Chinese companies like Z.ai and 360 are successively releasing models with performance close to US frontier models [9][12], and the Chinese government is pursuing self-sufficiency in computing infrastructure by promoting a plan to build AI data centers worth 2 trillion yuan (approximately $295 billion) [17]. Semiconductor smuggling continues, and a cat-and-mouse game repeats between enhanced US supply chain surveillance and China's circumvention strategies [13].
The most notable paradox in this process is that US export controls inadvertently provide a competitive advantage to Chinese AI companies. While US companies are constrained by government regulations, Chinese open-source models are targeting the global market without any access restrictions [9], and cost-competitive Chinese models are increasingly eroding the market share of US companies. Indeed, the data showing ChatGPT's global AI assistant market share falling below 50% for the first time clearly illustrates this trend [4]. Meanwhile, according to UBS analysis, there is a clear trend of companies switching from high-performance closed-source models to more affordable Chinese open-source models [10], putting structural pressure on the profitability of US AI companies.
Impact of the Baseline Scenario
In the baseline scenario, the global AI industry progresses towards bifurcation into two parallel ecosystems. A US-led ecosystem of high-performance closed-source models and a China-led ecosystem of low-cost open-source models coexist, forcing companies and governments worldwide to make strategic choices between them. Developing countries, in particular, may increasingly opt for Chinese models due to cost accessibility and technological sovereignty [16], potentially leading to an expansion of China's influence in the long run of AI standards competition. Countries like India, facing access restrictions from US AI companies, will feel an even greater need to build their own sovereign AI capabilities [7]. In this scenario, companies in US allied nations like South Korea and Japan face increasing strategic uncertainty under the dual pressure of relying on the US technology ecosystem while needing to maintain access to the Chinese market.
3. Pessimistic Scenario (Estimated Probability: 25-30%)
Development Trajectory
The pessimistic scenario unfolds with US ad-hoc export controls failing to curb China's AI militarization and instead undermining US AI industry competitiveness. The core mechanism of this scenario is the entrenchment of an asymmetric dynamic where the US government intensifies regulations on AI companies, slowing down innovation [3], while China accelerates AI development and deployment without constraints. As French media has pointed out, Chinese AI companies in Beijing, Shanghai, and Shenzhen may welcome the US tying down its AI giants with regulations, thereby expanding their competitive advantage [3].
In this scenario, China's AI militarization progresses at a faster-than-expected pace. The PLA's Strategic Support Force achieves substantial capability enhancements in automating cyberattacks, information warfare, and developing autonomous weapon systems using AI, gradually eroding US cyber security superiority [11]. Notably, research findings indicating that Chinese AI models have achieved performance close to Anthropic's Mythos in detecting software vulnerabilities support the realism of this scenario [11]. The semiconductor smuggling issue also escalates to uncontrollable levels, with Chinese companies like Alibaba and Tencent demonstrating a willingness to pay any price, diversifying smuggling routes across the entire supply chain [13].
China also secures a preemptive advantage in the 6G wireless infrastructure competition, intensifying US strategic anxiety on the new front of AI-infrastructure convergence [14]. As a former US Under Secretary of Homeland Security warned, China's early acquisition of 6G spectrum could structurally weaken US capabilities in mobile AI competition [14], leading to a reshaping of the physical infrastructure base for AI militarization in China's favor. To make matters worse, as allies like Canada and France react against US AI export controls by pursuing their own independent AI capabilities [4], the US-led AI alliance framework fragments.
Another key risk factor in this pessimistic scenario is the AI technology competition taking on the character of an 'AI Cold War,' leading non-aligned countries to seek their own third path [8]. As developing countries refuse to be fully incorporated into either the US or China AI bloc, the fragmentation of AI governance deepens, potentially making the establishment of international AI safety standards practically impossible. As the AI Preliminary Report by the UN Independent International Panel of Scientists points out, the deepening imbalance in AI benefit distribution between the Global South and the Global North [16] solidifies a structure where AI technology acts as a new fault line for geopolitical conflict.
Impact of the Pessimistic Scenario
In the pessimistic scenario, the impact on the global economy and industry is multi-layered and extensive. Firstly, the US AI industry's global competitiveness is structurally weakened by the dual pressures of regulatory burdens and market fragmentation. The decline of OpenAI's ChatGPT market share below 50% [4] could mark the beginning of a structural decline rather than a temporary adjustment, and Chinese open-source models increasingly erode the corporate market share with their cost competitiveness [10]. Nvidia and other semiconductor companies face a structural dilemma, having to contend with both restricted access to the Chinese market due to export controls and brand risks from smuggling issues [13].
On the security front, the likelihood of China's AI cyber capabilities posing a substantial threat to US digital infrastructure increases [11]. This extends beyond mere military threats, potentially spreading to vulnerabilities across the civilian economy, including financial systems, energy infrastructure, and supply chain management. Companies in allied nations such as South Korea, Japan, and Taiwan are exposed to the dual shock of supply chain restructuring and market access restrictions as US-China technological decoupling intensifies. In particular, a structural reorganization of the global supply chain in the fields of AI semiconductors and AI servers becomes inevitable.
4. Scenario-Based Impact Analysis on the Global Economy and Industry
AI Semiconductor and Hardware Industry
In all three scenarios, the AI semiconductor industry is at the forefront of geopolitical risk. In the optimistic scenario, supply chain restructuring through allied cooperation allows allied semiconductor companies like TSMC, Samsung Electronics, and SK Hynix to secure stable demand as key suppliers in the US-led ecosystem. In the baseline scenario, while Nvidia's access to the Chinese market is partially restricted, Chinese domestic AI semiconductors like Huawei's Ascend series continue to erode the Chinese domestic market. In the pessimistic scenario, the semiconductor smuggling issue escalates to uncontrollable levels [13], fundamentally undermining the reliability of the global AI semiconductor supply chain, and the industry structure reorganizes towards greater focus on domestic semiconductor production capabilities in each country.
AI Software and Model Development Industry
In the AI model development industry, the divergence of scenario-based impacts hinges on how the competition between open-source and closed-source models unfolds. In the optimistic scenario, US-led high-performance closed-source models maintain a premium in security and military domains, protecting the profitability of companies like Anthropic and OpenAI. In the baseline scenario, Chinese open-source models increasingly erode the market share of US companies in the corporate market by leveraging cost competitiveness [10], making it imperative for US AI companies to strategically focus on high-value specialized markets. In the pessimistic scenario, US AI companies face a slowdown in innovation due to the dual pressures of regulatory burdens and market fragmentation [3], while Chinese AI companies, unhindered, target the global market, making the risk of a reversal in technological leadership a reality.
Cybersecurity Industry
The cybersecurity industry is directly affected by the militarization of AI, and the threat landscape varies significantly depending on the scenario. In the baseline and pessimistic scenarios, Chinese AI models enhance their capabilities in detecting software vulnerabilities [11], accelerating the automation and sophistication of cyberattacks. This structurally increases the demand for cybersecurity investments from companies worldwide. In particular, companies in critical infrastructure sectors such as finance, energy, and defense will face pressure to make substantial investments in building defensive AI capabilities to counter AI-based cyber threats. In this regard, the cybersecurity industry may paradoxically emerge as a beneficiary of the intensifying AI technology competition.
Emerging Economies and Global South Economies
The impact on emerging economies and Global South countries varies most starkly by scenario. In the optimistic scenario, the US-led AI governance framework could be designed to ensure a certain level of AI accessibility for emerging economies. However, in the baseline and pessimistic scenarios, emerging economies become increasingly reliant on high-cost Chinese AI models[16]. While this lowers the cost of AI utilization for emerging economies in the short term, it harbors structural risks of technological dependency on the Chinese AI ecosystem and violations of data sovereignty in the long term. Amidst the trend of recognizing AI technology as strategic infrastructure[20], emerging economies will face an increasingly acute dilemma between building their own sovereign AI capabilities and external dependency[7].
Conclusion: Strategic Implications of Scenario Analysis
Synthesizing the three scenarios, even the baseline scenario, which is currently judged to have the highest probability of unfolding, reveals structural limitations that make it difficult for US AI export controls to fully achieve their intended objectives. The inherent diffusion potential of AI technology, coupled with China's improving indigenous development capabilities and the diverging interests of allies, suggests that export controls are likely to remain a limited tool for temporarily moderating the pace of technological competition, rather than an effective means of deterring China's AI militarization. This analysis aligns with concerns, as warned by the PIIE, that ad hoc AI model control measures could inadvertently benefit China. Amidst such uncertainties, it is crucial for companies to build flexible strategic portfolios that maintain resilience across multiple scenarios, rather than overly relying on strategies optimized for a single scenario.
Step 4: Analysis of Response Strategies
Responding to Enhanced US AI Export Controls and China's AI Militarization: Analysis of Response Strategies
Introduction: Framework for Analyzing Response Strategies
Under the dual pressure of intensified US AI export controls and accelerated Chinese AI militarization, relevant actors face a situation where they must choose different response paths based on their respective interests and capabilities. The three pathways—optimistic, baseline, and pessimistic—derived from the preceding scenario analysis serve not merely as a framework for future prediction, but as a strategic map defining the choices each actor must make under specific conditions. The following presents response options for each actor—the US government, allied governments, and global corporations—under each scenario, systematically evaluating the advantages, disadvantages, and feasibility of each option. The analysis focuses not on simple risk avoidance, but on maximizing strategic flexibility in a highly uncertain environment.
1. Response Strategies under the Optimistic Scenario
Presentation of Response Options
If the optimistic scenario materializes—that is, the US establishes a structured and coherent AI governance system in cooperation with its allies—the response options available to each actor are relatively clear. For the US government, institutionalizing the 'Coalition for AI Computing in the Free World' proposed by the Carnegie Endowment becomes a key response option. This includes transitioning from fragmented measures at the executive order level to a mid-to-long-term AI strategy with a legislative foundation, and introducing a tiered regulatory system that precisely categorizes the risk levels of AI models, imposing strict controls on high-risk military applications while allowing flexible access for commercial applications[2]. For allied governments, a conditional participation strategy that actively integrates into the US-led AI governance framework while protecting their own AI sovereignty and industrial interests emerges as a viable option. This includes a dual strategy, such as India's 'Sovereign AI' initiative, of building an independent foundation for their own AI capabilities within the alliance framework[7]. For global corporations, proactively aligning with the trend of the US-led AI standards system becoming the de facto standard for the entire democratic bloc, and building product and service portfolios that comply with these standards, becomes a core option.
Analysis of Advantages and Disadvantages
The response options under this scenario share the common strength of long-term stability and predictability. If the US government establishes consistent AI governance with a legislative basis, companies can move beyond regulatory uncertainty to plan long-term investments, and allies can deepen technological cooperation under common standards. In particular, a tiered regulatory system provides structural advantages that allow US AI companies to utilize the entire democratic bloc as a de facto domestic market. However, these options also entail significant disadvantages and risks. To elicit voluntary participation from allies, the US must invest substantial diplomatic resources, and the sensitivity surrounding each nation's industrial interests and AI sovereignty can limit the scope and depth of cooperation. Canada's and France's declarations of strengthening their autonomous AI capabilities suggest that allies will not unconditionally integrate into the US-led framework[4]. Furthermore, operating a tiered regulatory system presupposes the technical and institutional capacity to accurately assess the risk levels of AI models, a capacity that no country currently possesses sufficiently.
Feasibility and Risk Assessment
The response options under the optimistic scenario have a medium level of difficulty in terms of feasibility. Political polarization within the US and the lack of legislative cooperation between the executive and legislative branches act as structural impediments, and coordinating the interests of allies requires considerable time and diplomatic resources. The core risk lies in the transition costs incurred during the realization of this scenario. While US AI companies bear short-term regulatory burdens, Chinese AI companies, unhindered in their global market pursuit, can expand their market share[9], paradoxically undermining the foundation of long-term technological superiority intended by the optimistic scenario.
Priority Response Measures
The highest priority response measure in the optimistic scenario is the early establishment of a classification system for AI model risks. This is essential for finding a balance that minimizes industrial damage from excessive controls while achieving security objectives, by laying the technical groundwork for tiered regulation. The second priority is to institutionalize channels for AI governance cooperation with allies, which plays a crucial role in thwarting China's strategy of individually targeting allies to divide their solidarity.
2. Response Strategies under the Baseline Scenario
Presentation of Response Options
The baseline scenario posits a structural competitive state where US export controls are makeshift and China's indigenous development accelerates, gradually narrowing the technological gap between the two. In this scenario, the US government's core response option is to strengthen the enforcement of the current export control system while simultaneously pursuing parallel strategies to compensate for its limitations. Specifically, this includes enhancing surveillance of NVIDIA's AI server supply chain to block illicit channels[13] and, concurrently, expanding public investment in the US AI research ecosystem to deepen technological superiority itself. The logic behind this option is that the US must also strive to maintain the gap through strategic investment in computing infrastructure, given China's ambitious plan to invest 2 trillion yuan (approximately $295 billion) in AI data center infrastructure[17].
For allied governments, a dual hedging strategy—maintaining cooperation with the US while building an independent foundation for their own AI capabilities—emerges as a realistic response option. The launch of Fugu, an independent AI model by the Tokyo-based AI startup Sakana AI, demonstrates that allies are moving towards reducing their reliance on US models and strengthening their own capabilities[12]. For global corporations, a strategy of diversifying their portfolios, strategically leveraging the cost competitiveness of Chinese open-source AI models while maintaining US models in security-sensitive areas, is effective. Indeed, according to UBS analysis, a trend of companies switching to Chinese open-source models like Alibaba's Qwen series for tasks not requiring high-performance closed-source models is already becoming visible[10].
Analysis of Advantages and Disadvantages
The response options under the baseline scenario have the strength of high realistic feasibility. Strengthening supply chain surveillance and enhancing enforcement can be pursued within the existing institutional framework, and the dual hedging strategy offers allies flexibility to protect their national interests while maintaining relations with the US. Portfolio diversification at the corporate level also has the advantage of achieving both cost efficiency and risk diversification. However, the fundamental disadvantage of these options is that they only manage, rather than solve, structural problems. Blocking illicit channels is technically impossible to completely seal off, and there are limits to slowing down the pace of China's indigenous semiconductor technology development. The dual hedging strategy risks weakening alliance trust with the US, and corporate portfolio diversification paradoxically exposes companies to security risks as their reliance on Chinese AI models increases. In particular, companies extensively using Chinese open-source models in an escalating AI cold war may face further intensified regulations in the future[8].
Feasibility and Risk Assessment
The response options under the baseline scenario are generally highly feasible, but their sustainability is questionable. While strengthening supply chain surveillance has the effect of deterring illicit trade in the short term, enforcement costs increase exponentially as China's circumvention strategies become more sophisticated. The core risk is the paradoxical effect where US export controls inadvertently accelerate China's indigenous development; the concern, as warned by the PIIE, that ad hoc control measures could benefit China, operates most realistically in this scenario[1][3]. Furthermore, if the AI cold war triggers a non-aligned movement, leading Global South countries to seek a third path independent of both the US and China[8], the universality of the US-led AI governance framework itself could be challenged.
Priority Response Measures
The highest priority response measure in the baseline scenario is a 'two-pronged strategy' that refines the enforcement mechanisms of export controls while simultaneously expanding public investment in the US AI research ecosystem to deepen technological superiority. Controls alone cannot prevent China's technological catch-up, making it more effective in the long run for the US to maintain its position as a technological leader as a deterrent. At the corporate level, establishing internal governance systems that clearly distinguish the use of models based on their application and security sensitivity becomes a key priority.
3. Response Strategies under the Pessimistic Scenario
Presentation of Response Options
The pessimistic scenario posits a path where US makeshift export controls have the counterproductive effect of accelerating China's AI self-sufficiency, and divisions deepen as allies react against unilateral US actions, pursuing their own independent AI strategies. In this scenario, the US government's response options are broadly divided into two. The first is the 'hardline continuation' option, which maintains the current export control stance and strengthens enforcement further. The second is the 'strategic adjustment' option, which acknowledges the counterproductive effects of export controls and shifts to a multilateral approach centered on allied cooperation. The former has lower short-term political costs but risks exacerbating structural problems, while the latter entails the cost of damaged credibility from unilateral US actions in the short term but can open a more sustainable path in the long term.
For allied governments, as the costs imposed by unilateral US AI export controls on their domestic companies and research institutions become more apparent, political pressure to build independent AI capabilities increases. Canada's and France's declarations of strengthening their autonomous AI capabilities indicate that movement in this direction has already begun[4], and India's Sovereign AI strategy can be understood in the same context[7]. In this scenario, the key response option for allies is 'strategic autonomy,' which involves maintaining formal cooperation with the US while substantively building an independent AI ecosystem. For global corporations, preparing for a situation where the global single market splits into US and Chinese blocs due to the deepening AI cold war, a 'dual-track structuring' strategy—building separate product and service systems for each bloc—emerges as an unavoidable choice.
Analysis of Advantages and Disadvantages
The response options under the pessimistic scenario commonly face the dilemma of high costs and limited effectiveness. The US government's 'hardline continuation' option has the advantage of maintaining political consistency in the short term, but it has the structural counterproductive effect of accelerating China's AI self-sufficiency and deepening allied estrangement. The case where Chinese startup Z.ai released a model with comparable performance without restrictions immediately after the US blocked access to Anthropic's Fable 5 and Mythos 5 clearly illustrates the paradox where intensified controls increase incentives for China's indigenous development[9][12]. The 'strategic adjustment' option is more effective in the long term but entails short-term costs of damaged credibility from previous measures and domestic political backlash.
The allies' 'strategic autonomy' option brings the positive effect of fostering their own AI industries, but it carries the risk of becoming vulnerable to China's divide-and-conquer strategy due to weakened technological cooperation channels with the US. The global corporations' 'dual-track structuring' strategy is a realistic measure for maintaining market access, but the surge in operating costs and the complexity of meeting the regulatory requirements of both blocs simultaneously pose a significant burden. In particular, as China moves towards establishing its own AI agent standards at the national level to build an interoperability ecosystem[15], companies face the dual burden of adapting to two different technological standard systems.
Feasibility and Risk Assessment
The response options under the pessimistic scenario have the highest uncertainty in terms of feasibility. The 'hardline continuation' option is feasible in the short term, but its sustainability rapidly declines as structural counterproductive effects accumulate. The 'strategic adjustment' option requires both political will and institutional capacity, and it is uncertain whether the political environment in the US will permit it. The allies' 'strategic autonomy' is feasible for countries with sufficient financial capacity and technical personnel, but it is not a realistic option for others. The global corporations' 'dual-track structuring' is feasible for large multinational corporations but is virtually impossible for small and medium-sized enterprises. The core risk is that once the pessimistic scenario becomes entrenched, reversing it will require immense costs and time. In particular, the fragmentation of AI technology standards, once solidified, will have long-term consequences that structurally reshape the global technological ecosystem over decades[20].
Priority Response Measures
The highest priority response measure in the pessimistic scenario is paradoxically a preemptive measure to prevent this scenario from materializing: 'strategic adjustment,' which involves recognizing the counterproductive effects of export controls early on and shifting to a multilateral approach centered on allied cooperation. If already in the pessimistic scenario, 'damage limitation' strategies become key to minimizing harm. Specifically, restoring dialogue on AI governance with allies and providing companies with sufficient predictability regarding regulatory changes become the top priorities. At the corporate level, building a 'scenario-neutral' portfolio that can survive any scenario—that is, a strategy of avoiding excessive dependence on a particular technological bloc while internalizing core capabilities—becomes the most realistic priority.
4. Comprehensive Evaluation: Strategic Principles Across Scenarios
Synthesizing the analysis of response strategies across the three scenarios yields several strategic principles that cut across scenarios. First, the structural limitation that export controls alone are unlikely to effectively curb China's AI capability development is consistently confirmed across all scenarios. The paradox where intensified US control measures increase China's incentives for indigenous development operates as a mechanism in all three scenarios, suggesting that export controls should function as a supplementary tool, not a standalone strategy[1][3]. Second, the coherence of allied cooperation is confirmed as the key variable determining the effectiveness of US AI strategy across all three scenarios. Unless allies share the same export control standards as the US, China can exploit loopholes through allied nations, fundamentally limiting the effectiveness of controls. Third, companies must recognize that securing strategic flexibility to adapt to rapid changes in the AI technology landscape is the most critical task, regardless of which scenario unfolds. Strategies overly optimized for a specific scenario incur substantial transition costs when the scenario shifts; therefore, ensuring 'robustness' through internalization of core capabilities and portfolio diversification operates as a common strategic principle across all scenarios[5][20].
Step 5: Final Recommended Response Strategies
Responding to Enhanced US AI Export Controls and China's AI Militarization: Final Recommended Response Strategies
1. Comprehensive Judgment and Recommended Response Strategies
Comprehensive Judgment of the Current Situation
Synthesizing the preceding issue situation analysis, in-depth analysis, scenario analysis, and response strategy analysis leads to the conclusion that the current US-China AI technology competition is in a complex structure where optimal results cannot be achieved by the unilateral strategies of any single actor alone. US export controls have been accumulated in a makeshift manner, with a mix of security imperatives and industrial-political motives, simultaneously undermining policy consistency and effectiveness. The case where, immediately after the US implemented measures to prohibit non-US persons from accessing Anthropic's Fable 5 and Mythos 5, Chinese startup Z.ai released a model with comparable performance without any restrictions[9], symbolically demonstrates the paradoxical outcome where the current export control system fails to effectively curb China's AI capability development while only weakening the global competitiveness of US companies. The release of Thorunfeng by Chinese cybersecurity firm 360, claimed to be comparable to Mythos, is also interpreted in the same context[12].
In this situation, the most realistically unfolding path is the baseline scenario, characterized by a fragmented competitive landscape where US export controls maintain partial effectiveness while China's indigenous development accelerates in parallel. In this landscape, no single response strategy can resolve all risks simultaneously; therefore, a portfolio-based response strategy requiring flexibility and adaptability as core principles for each actor is necessary. AI is already treated as strategic infrastructure beyond mere software[20]. Starting from this recognition, each actor must pursue a balanced approach across maintaining technological superiority, ensuring supply chain resilience, and deepening allied cooperation.
Core Recommended Response Strategies
Based on the comprehensive judgment, this report recommends differentiated core response strategies for actors at three levels. First, for the US government, it recommends a strategic shift from ad hoc model access restrictions to establishing a risk-based tiered regulatory system, linked with a multilateral AI governance framework with allies. Second, for allied governments, it recommends a dual strategy of actively building their own sovereign AI capabilities beyond passively integrating into the US export control system, and proactively participating in the 'Coalition for AI Computing in the Free World' proposed by the Carnegie Endowment[7]. Third, for global corporations, it recommends accepting geopolitical fragmentation as an irreversible structural change and building resilient business models that can operate under any scenario through supply chain diversification and technological portfolio restructuring.
2. Short-Term/Mid-Term/Long-Term Implementation Plan
Short-Term Implementation Plan (0-6 Months): Minimizing Risk Exposure and Immediate Adaptation
In the short term, the most urgent task is to minimize operational risks arising from the uncertainty of current export control measures. The US government must face the reality that Chinese companies are rapidly filling the market void created by the access ban on Anthropic models[9][12] and promptly clarify the scope and criteria of current measures. Specifically, it should expedite the establishment of a system for classifying AI models into high-risk military applications, intermediate-risk dual-use applications, and low-risk commercial applications, and disclose the access control criteria applicable to each tier, enabling companies and allies to make decisions in a predictable environment[2].
In the short term, allied governments must rapidly assess the impact of US export control measures on their domestic companies and research institutions and activate diplomatic communication channels to minimize damage. Countries like India, whose AI sovereignty strategies are directly affected by US restrictions on AI model access[7], must immediately initiate negotiations to secure their status as trusted partners with the US and seek exceptions through bilateral consultations. Concurrently, they must closely monitor the rapid spread of low-cost Chinese open-source models, such as Alibaba's Qwen series, as alternatives to US models among their domestic companies[10], and assess the implications for supply chain dependency and data security.
In the short term, global corporations must conduct internal audits to assess the geopolitical risk exposure of their currently used AI models and services. In particular, contingency planning is necessary to identify business areas reliant on models subject to US export control measures and to secure alternative options in advance. Furthermore, given the serious emergence of smuggling issues in the AI server supply chain[13], checking the possibility of their supply chains being involved in sanctions violations and strengthening compliance systems should also be included as short-term tasks.
Mid-term Execution Plan (6 months to 2 years): Structural Adaptation and Strategic Positioning
In the mid-term, structural adaptation to the evolving technological landscape is required, moving beyond short-term risk management. The U.S. government should transition from fragmented measures at the executive order level to a mid-to-long-term AI strategy with a legislative foundation. In this process, the concept of a liberal AI computing coalition proposed by the Carnegie Endowment should be substantially institutionalized, establishing a multilateral framework where democratic allies adopt common export control standards and AI safety standards. Given that China has announced the Interconnection Standard for Seven National AI Agents, promoting standardization of its own AI ecosystem[15], there is a risk of losing leadership in the technology standards competition if the U.S. does not proactively establish an alternative standard system with its allies.
In the mid-term, allied governments must actively invest in building their own AI sovereignty capabilities. As Canada and France have declared their pursuit of AI autonomy[4], a strategy to move away from excessive reliance on U.S. models and build AI infrastructure based on domestic or allied foundations should be set as a mid-term objective. This is not about abandoning cooperation with the U.S., but a strategic hedge to increase bargaining power and reduce vulnerability to a single source. Particularly, as warnings are raised that China is preemptively gaining an advantage in infrastructure areas directly linked to AI competitiveness, such as 6G spectrum allocation[14], allies must promptly establish investment plans in this area.
In the mid-term, global corporations must implement the geographical diversification of their AI supply chains. In a structure of deepening U.S.-China technological fragmentation, business models that rely entirely on the technological ecosystem of a single bloc have low sustainability. Therefore, a dual supply chain system that utilizes both U.S.-based models and European or allied-based models in parallel should be established, creating a structure that mitigates the impact of regulatory changes in a specific region on the entire business. Simultaneously, AI-related compliance capabilities must be internalized to ensure organizational capacity to respond swiftly to changes in export control regulations.
Long-term Execution Plan (2 years and beyond): Securing Sustainable Competitiveness in a New Technological Order
In the long term, the core objective is to establish a structure for maintaining sustainable competitiveness, regardless of how the current U.S.-China AI competition stabilizes. The U.S. government should place greater emphasis on offensive strategies to enhance its own AI innovation capabilities, rather than defensive measures like export controls, to maintain its AI technological superiority in the long run. Given that China is pursuing a plan to invest 2 trillion yuan (approximately $295 billion) in AI data center infrastructure[17], the U.S. must also strengthen its computing infrastructure and AI talent ecosystem through public-private collaborative investments of a comparable scale. Export controls alone cannot permanently block China's progress; ultimately, superiority in the speed of innovation becomes the decisive factor in technological hegemony[5]. This should be the starting point for the long-term strategy.
In the long term, allied governments should keep the possibility of a non-aligned movement in the AI cold war as a strategic option. If the U.S.-China AI competition solidifies into a structure that forces a choice of blocs upon a wide range of actors, including Global South countries, maintaining strategic autonomy for smaller and medium-sized nations to maximize their own interests without complete subordination to either bloc will become a crucial long-term objective[8]. This requires their own AI sovereignty capabilities to grow beyond a certain level, which justifies long-term investment.
In the long term, global corporations must establish innovation strategies that actively respond to the development direction of AI technology itself. The performance gap between AI models is rapidly narrowing[9][11], and the trend of the open-source ecosystem emerging as an alternative to closed models weakens the economic rationality of exclusive reliance on specific models in the long run. Therefore, corporations must reduce their dependence on specific AI models or suppliers and build technological architectures and organizational capabilities for flexible utilization of various models in the long term.
3. Monitoring Indicators and Trigger Points
Key Monitoring Indicators
In the current uncertain environment, a systematic monitoring system is essential for effective strategy execution, enabling early detection of situational changes and adjustment of response directions. It is effective to manage monitoring indicators by categorizing them into three groups: technological, policy, and market indicators.
As for technological indicators, first, the trend of the performance gap of Chinese AI models is most important. The speed at which models released by Chinese companies such as Z.ai, 360, and Alibaba are closing the performance gap with U.S. frontier models is a key metric for judging the effectiveness of export controls[9][12]. Second, the status of China's acquisition of AI training computing resources. By tracking the number of advanced semiconductor smuggling cases and the progress of China's indigenous semiconductor development, it can be assessed whether hardware bottlenecks are acting as a practical constraint[13]. Third, changes in the gap between the U.S. and China in 6G spectrum allocation and AI infrastructure development should also be used as leading indicators for assessing long-term technological superiority[14].
As for policy indicators, first, the legislative status and scope of U.S. AI export control measures are key. If they remain at the executive order level, policy consistency and predictability will remain low, but if institutionalized through congressional legislation, policy sustainability and the possibility of coordination with allies will increase[2]. Second, the practical progress of the liberal AI computing coalition proposed by the Carnegie Endowment. Whether allies adopt common export control standards or pursue a fragmented approach based on their respective interests will determine the feasibility of multilateral AI governance. Third, the trend of strengthening China's AI-related foreign investment regulations. With the enforcement of China's State Council Order No. 837, national control over technology transfer in overseas investments by Chinese companies has been strengthened[6]; the impact of these regulations on the global expansion strategies of Chinese AI companies must be continuously tracked.
As for market indicators, first, changes in the global market share of U.S. AI services, including ChatGPT, are important. The fact that ChatGPT's global AI assistant market share has fallen below 50% for the first time suggests a fundamental shift in the competitive landscape[4]; whether this trend continues will be a barometer for assessing the global competitiveness of the U.S. AI industry. Second, the global adoption rate of Chinese open-source models. The phenomenon of Chinese open-source models, such as Alibaba's Tianwen series, being adopted by U.S. companies for cost reduction purposes is spreading[10]; the extent to which this trend progresses will serve as an indicator that indirectly shows the practical effect of export controls.
Trigger Points and Response Transition Criteria
Trigger points are defined as decisive events that signal a transition from the current baseline scenario to an optimistic or pessimistic scenario. Triggers indicating a transition to an optimistic scenario include: first, the U.S. and major allies signing a multilateral agreement containing common AI export control standards; second, China's indigenous development of advanced semiconductors encountering technological limitations, creating a practical bottleneck in AI training capabilities; and third, the speed of performance improvement of Chinese AI models significantly slowing down, widening the gap with the U.S. again.
Conversely, triggers indicating a transition to a pessimistic scenario include: first, the Chinese People's Liberation Army deploying AI-based autonomous weapon systems in actual combat or large-scale AI-enabled cyberattacks occurring; second, U.S. export control measures provoking backlash from allies, leading to the fragmentation of the multilateral cooperation system; and third, Chinese AI companies replacing U.S. companies in the global market at a faster-than-expected pace, significantly weakening the U.S.'s ability to set technological standards. If these triggers occur, each actor must immediately activate pre-prepared contingency plans to transition their response strategies.
4. Summary and Conclusion
The dual pressures of strengthened U.S. AI export controls and China's response to AI militarization are not merely issues of technological competition, but signals of structural transformation in the formation of a new technological order. The current U.S. approach excessively relies on ad-hoc blocking of model access, paradoxically failing to effectively deter China's indigenous development while weakening the global competitiveness of U.S. companies[1][9]. To resolve this paradox, export controls must be refined, allied cooperation multilateralized, and domestic innovation capabilities strengthened in a balanced manner.
From the perspective of global corporations, the key message is to accept geopolitical fragmentation not as temporary disruption but as a structural change, and to immediately build resilient strategies that can operate under any scenario. Diversification of AI supply chains, internalization of compliance capabilities, and flexibility of technology portfolios are no longer optional choices but essential conditions for survival. The reality of Chinese open-source models being adopted by U.S. companies for cost reduction[10] demonstrates that the geopolitical boundaries of technology are constantly being reshaped by market logic, independent of policymakers' intentions.
Ultimately, the success or failure of this competition will not be determined by who more effectively blocks the other's technological access, but by who learns, innovates, and adapts more quickly[5]. Export controls can serve as a means to buy time, but if that time is not invested in strengthening innovation capabilities, they lose their strategic significance. This is the most fundamental strategic challenge facing the U.S., its allies, and all global corporations caught in the vortex of this competition.
References
[1] [Børsen] Ny analyse: “Alvorlige” risici ved amerikansk AI-strategi
[2] [The News International] White House prepares voluntary standards for advanced AI releases
[3] [Le Temps] La Maison-Blanche dicte sa loi aux géants de l’IA, une stratégie suicidaire
[4] [Xinhua (新华社)] 特稿丨从三个关键词看6月全球AI领域发展
[5] [South China Morning Post] In the AI era, US-China competition hinges on who can adapt faster
[6] [The Diplomat] China’s New Investment Regulations Block Strategic Technology Transfer
[7] [The Hindu] Reimagining sovereign AI for India’s strategic future
[8] [Nikkei Asia] The AI cold war needs a nonaligned movement
[9] [Australian Financial Review] How Chinese AI models are closing the gap on Anthropic and OpenAI
[10] [财新 (Caixin)] 瑞银：60%的企业正通过设立“护栏”机制控制AI支出
[11] [Exame] Avanço da IA da China reduz vantagem dos EUA em cibersegurança
[12] [TechCrunch] Asian AI startups launch Mythos-like models as Anthropic’s export ban drags on
[17] [The Diplomat] China’s $295 Billion Ambitions for AI Will Drive up Domestic High-Tech Stocks
[19] [La Diaria] The United States Has a Plan
[20] [Economic Times] From OpenAI to Sarvam, Govts May Want Skin in AI Game
*This text is an AI translation of an original written in Korean. Some translations or nuances may be inaccurate.