Infrastructure as Leverage: Chinese AI and Its Consequences

In November 2025, as the world was swept up in the AI revolution, Singapore made a unique decision, relinquishing Meta’s Llama and instead building it’s sovereign AI model on Alibaba’s Qwen.1 This development came after Singapore’s SEA-LION program’s deliberation, selecting Qwen due to its high scoring on Southeast Asian languages on the regional benchmark SEA-LION established, SEA-HELM. The program had previously been built on Llama and released a Gemma-based model only months prior; however, Singapore’s official decision was one of advancing regional sovereignty, largely due to the fact that the model is open-source and forkable. This reflected a deeper pattern, one of the significance of Chinese technology in the default choices which stretch far beyond Beijing's formal orbit. It is this logic in China's commitment in AI standard-setting in a global scope, especially on territories in the Global South where Western influences and the vigilance for national security are not overwhelmingly prevailing, and the opportunist peripheries of the Global North (such as some less advanced economies within the EU), is predicated on the overarching logic of shared interests and thereby asymmetrical dependence. Put bluntly, dependence on infrastructure innately outlasts changes in governance.

Singapore’s decision is one that establishes a concerning pattern of powers choosing infrastructure benefits in the short term, for a cheaper price than competitors, but with a long-term cost of dependence and political leverage. China is the master of such negotiations, and this logic is embedded within Chinese domestic politics and foreign policies alike. It is an embodiment of China's strategic preference for transactional exchange of interests, i.e., Quid pro quo, and infrastructural hostage to lock down long-term political-economic benefits and adherence to the advantage of the ruling CCP both at home and abroad. One should note that this idiosyncratic tactic is deliberately devoid of Western values, ideologies, and idealism, such as democracy, universal human rights, and political accountability, as fundamental principles to shape and constrain behaviors of state actors and govern human societies. Therefore, AI standard-setting is constantly modified and updated by competing political-economic forces, rather than a result of unimpeded technological dissemination and evolution into all possibilities and scenarios.

This dynamic is far from speechcraft and rhetorical influence. Beijing's 2018 China Standards 2035 push, as well as the 2021 National Standardization Development Outline following it, explicitly direct Chinese firms to increase their technical capabilities and export Chinese-set specifications internationally in priority sectors including AI, using Belt and Road projects and standards bodies as delivery channels.2 This is a large step from how Western-led standards bodies such as the International Organization for Standardization operate, as such institutions are voluntary and multi-stakeholder, aggregating competing input from firms and national delegations without a single government directing the outcome. China's approach is deliberately different, explicitly subordinating standard-setting to the state's industrial strategy.

Put bluntly, Beijing is actively seeking to establish projects that create international obedience to China through slow and deliberate standardization of Chinese technology.

The enthusiastically promoted One Belt, One Road Initiative (the BRI) aims at not only building physical infrastructure to integrate the Eurasian continent, but also absorbing countries and economies involved into China's orbit through, for example, technological dependence, and trade and investment reliance. The latter is intangible by nature, yet nonetheless addictive, country-specific, and potentially extortionate by coercive means at some point. China's AI standard is a technological shell with a monopolized political core inside. The Chinese authority wants to popularize and embed it as the default technological undercurrent for global adopters. The BRI reveals China's global ambition and expansive tendency under the disguise of multilateral co-development and a seemingly innocuous Chinese proposal for non-Western globalism. Once Chinese technological protocols, hardware specifications, and operating standards are organically integrated into the global AI blueprint, serving as the inherently exclusive and less compatible infrastructural foundation that applications rely critically upon to function, it can create enormous technological inertia for adopters. Extremely high switch costs to separate from it or simply to diversify by incorporating alternative systems can deter almost any second thought in a future scenario.

This is reflected from a ground perspective as well as a bird's-eye view, as Ecuador's ECU-911 system, established under a 2011 agreement, was financed by a $240 million loan from Beijing.3 Ecuador contracted the Chinese state-owned CEIEC and hardware supplier Huawei to establish a nationwide surveillance and emergency-response network which went into operational capability in November 2016. Within a few years, the system relied on more than 4,300 Huawei-supplied cameras nationwide, including a facial recognition capability deployed at the airports in Quito and Guayaquil as well as the city of Cuenca.4 This tool, meant to be a crime-fighting establishment to Ecuadorans, was later reported to have fed footage to the country's domestic intelligence service. Ecuador's dependence on Chinese-built and Chinese-manufactured hardware for core public safety functions is precisely the kind of technological trap described; once installed, the cost of removing the cameras, servers, and software is high enough that a change in government will not lead to a systemic replacement. It is, put simply, a one-way entry trap.

China dominates legacy chips through its remarkable manufacturing scale and extraordinary ability to cost-down. Weaponization of this competitive advantage by China is entirely possible. However, the country is strictly excluded from Extreme Ultraviolet Lithography (EUV) equipment provision that is monopolized by the Dutch firm, ASML, a restriction the Dutch government formalized through export-control regulations that took effect in September 2023 and tightened further in 2024.5 As well as advanced and high-capacity chips under the size of 7 nanometers, predominantly manufactured by TSMC for Nvidia, notably H100 and the Blackwell series, and others, exports of which have been subject to US licensing requirements since 2022.6 China's hostile and confrontational relationship with Taiwan carries a real, if debated, military risk rather than a fixed prediction. In 2021 the then-commander of US Indo-Pacific Command warned China could be capable of moving against Taiwan within six years, the so-called Davidson window, which runs through 2027, and a 2023 CSIS wargame of a Chinese amphibious invasion found that while a joint US-Taiwan-Japan defense would likely repel it, doing so would come at extraordinarily high military and economic cost to all sides, underscoring the risk to the security and stability of global advanced-chip supply chains.7 In addition, the intensifying competition and rivalry between the US and China for global preeminence, especially over high-techs that can be instrumentalized for defense, and military purposes or revolutionize economic productivity and sophistication, notably AI, add more geopolitical and geoeconomic uncertainties and contentions to an already volatile and tumultuous situation in the highly globalized semiconductor industry. AI serves as a modern catalyst for conflict; one must be wary of its implications for Chinese leverage.

For state actors that fall between the dueling giants in the US-China AI race, hedging between both powers is not a sustainable long-term strategy, especially from a strategic point of view. The key reasons behind this are America's secondary sanctions and American security and stability provisions for conflict-ridden regions, such as a 2023 nuclear umbrella extension to South Korea via the Washington Declaration, wherein major semiconductors such as Samsung and SK Hynix are located.8 The enormous market capacity possessed by the US and China, respectively, with politically filtered entry barriers and operational regulations, serves another role in the same rationale of unsustainable longevity of this middle ground strategy. Lastly, ongoing and deepening derisking or decoupling driven by political incompatibility, divergent economic ambition and interests, as well as colliding national security pursuits, all prove a barrier to this long-term strategy. State actors at this time can choose either the US or China, but not both, no matter how great the opportunity cost may be.

Singapore might seem to be an exception, and the country is walking on a tightrope. This tiny yet commercially prosperous financial and logistical hub with both Eastern and Western heritages in Southeast Asia positions itself as a bridge between the US and China. This strategic choice actually requires extraordinary diplomatic skills and another crucial precondition, which is that its relationship with China does not jeopardize the security interests of the US in the region and beyond, and the reverse is also true from China's strategic perspective. However, as the US-China AI race escalates, it might not be attainable in reality when the space for diplomatic maneuverings and strategic ambiguity ceases to exist in any viable form.

Singapore's position reflects the strain established by these policies; in 2024, a Singapore-registered firm, Megaspeed, backed the Chinese gaming and cloud company 7Road.9 Megaspeed committed to purchasing billions of dollars in Nvidia GPUs, even as Singapore's share of Nvidia's reported revenue vastly exceeded the quantity of chips delivered within its borders.10 This gap drew US scrutiny after China's DeepSeek was released in early 2025 as a competitive AI model that appeared to rely in part on restricted hardware from FDPRs. By early 2025, Singaporean prosecutors had separately charged individuals accused of routing Nvidia-powered servers through the city-state to China, thus underscoring the difficulty for a sophisticated, rules-based hub to keep its balancing act above suspicion once chip flows themselves become the arena for great power competition.11

It is noteworthy that China's infamous red supply chains have already infiltrated deeply into those of its neighboring countries and economies, notably Singapore, Malaysia, Taiwan, South Korea and even Japan. These destinations are plagued by rampant smuggling of advanced chips and servers to China's jurisdiction. Apart from document forgery and establishing shell companies, the Chinese state controls a large number of shadow agents and entities hidden in the gray zones and institutional loopholes of the vast and expanding logistical networks in Asia centered around major hubs and powerhouses of electronics, semiconductors, and AI-related, high-tech industries. These aforementioned Asian countries rely critically on America's military protection to a significant degree. Yet, they also desire China's enormous markets in combination. If they truly outweigh security over trade opportunity, then they should resist gradual encroachment of China's expanding sphere of influence and thoroughly eliminate any presence of red supply chains in explicit and/or covert forms from their own networks, as well as upgrade their institutionalized oversight mechanisms to better cooperate with the US authorities. Otherwise, they too might be excluded from the emerging democratic and rule-based supply chain ecosystem built on shared values, mutual trust, and collective security.

The Chinese action to deliberately expand into other nations for its own gain is nothing new; however, it is critical to halt when it takes place deep within US-allied nations and occurs in the field of the most volatile technology yet established.

Risks and their presence are not hypothetical by nature; they are either actualized or resting at a potential to become such. US federal prosecutors have documented networks that relabeled and redistricted Nvidia H100 and H200 processors, falsified end-user paperwork, and reputedly exported hardware through Southeast Asian intermediaries to reach Chinese buyers.12 One of these buyers was a US-based network the Department of Justice claimed exported or attempted to export more than $160 million in restricted GPUs between October 2024 and May 2025.13 This evidence presents a clear realization of this risk. China pursues constant AI standard-setting and infrastructure exports; it should then be evident that workarounds to the chip supply chain controls are not separate initiatives from these. Instead, it is the establishment of a long-term strategy to create a durable, asymmetrical dependence that partner states find difficult to reverse- it is fundamentally a one-way entry trap.

Both US policymakers as well as mid-sized and developing states which fall between the giants of Washington and Beijing have a response role to play. Washington must pair export-control enforcement, shuttering the transshipment loopholes exposed by Singapore, and retain an affordably priced alternative to Chinese infrastructure. Chinese leverage is a trap of economic appeal- Washington must respond with an even better one. Evidently, restriction does not work- at least not alone, and as such it is the economic counterpart of a deal to prevent abusive leverage from China taking hold. Partner governments of Washington must build institutional capacity to audit their digital and telecommunications supply chains for undisclosed Chinese state involvement. This must occur before infrastructure one-way trapping suffocates that state with a scarcity of alternatives.

Without this combination of enforcement and alternatives, the shared interests Beijing offers will prove more durable, in practice, than the ad hoc coalitions Washington assembles to respond.

Wars are not won with responsive strategies; they are won with initiative.

Notes & References