The race for AI is not simply a contest between the US and China. Dr Siyi Liu argues that middle powers can gain influence by building trusted partnerships around complementary strengths rather than pursuing unattainable AI independence.
The new anatomy of AI and digital power
The ambition of the Bennett School’s AIxGEO project is to offer an alternative to the mainstream narrative surrounding artificial intelligence (AI) as a zero-sum, titanic clash and arms race between the United States (US) and China. Since we launched the project, there has been a growing focus on other major, small and non-great powers (so-called middle powers), which – from Europe and East Asia’s tech hubs – are rapidly pivoting to ‘sovereign AI’ strategies. Such national strategies (for example in South Korea, the United Kingdom (UK), and the European Union’s (EU) new technological/digital sovereignty agenda) generally advocate for countries to build the AI full stack (from data, raw hardware, to foundation models, orchestration models and deployments) domestically and to treat any dependence on others as a vulnerability to be eliminated.
There is only one problem: total AI self-reliance is an illusion and unachievable (Kumar 2023; Cha 2023; Yeung 2026). Attempting to do so doesn’t create security and sovereignty; on the contrary, it can backfire and turn ordinary economic links into weaponised leverage. For example, in 2019, Japan restricted exports of chip-making chemicals to South Korea, but in return South Korea fostered domestic industries to decouple from Japan on the product. The event marks how a tradable commodity was used for geopolitical coercion and thus fractured a regional supply chain. Policymakers first need to discard the habit of measuring geopolitical influence through raw economic scale or military spending. In the AI era, power is structural, specific, and profoundly decentralised.
For non-great powers, there could be a “third path” between submission and autarky. Rather than trying to replicate every layer of the AI ecosystem, these agile states are leveraging international organisations to encourage domain-specific collaboration. The mechanisms aim to achieve functional equivalence – designing different regulatory and technical mechanisms that yield equivalent security and accountability outcomes – giving them collective bargaining power. These actions prepare the formation of minilateral coalitions, which are small, agile and focused networks that act as a closely knitted and balanced web to prevent and mitigate fragmented supply chains.
As the AI supply chain undergoes a messy restructuring, the next phase of the AIxGEO project will ask how non-great power actors can participate and shape AI governance through the structurally distributed and fragmented AI supply chains by turning the chokepoints and bottlenecks into complementary leverage based on mutual trust. This requires a fundamental shift from a passive and reactive posture of building walls to a proactive understanding of supply chain complementarity and complexity. The project will involve a three-part analytical framework.
1. Global supply chain complexity
The global AI supply chain is not just about the chips or the computing. It spans from energy, internet and utility infrastructure, raw materials such as silicon and rare earth materials, to semiconductors (different types), cloud computing, data centres, capital, data, markets and talent. Middle power actors sit in different parts of the chain and hold different bargaining levers.
Countries possess leverage that can far exceed conventional measures of diplomatic weight. For example, ASML’s lithography machines can be leveraged as a ‘weapon’ to cut any country from securing advanced AI chips. The US Cloud Act can be used as an extraterritorial leverage to reach datasets that are not within the US territory but hosted by US-owned servers. Germany and South Korea purified about 69 percent of the silicon for China, indicating a high dependency despite China being the top 1 raw silicon provider. This snapshot shows that the general understanding of the AI supply chain is too thin.
Identifying different types of leverage provides stakeholders with different narrative angles and situated perspectives to see and understand countries’ strengths, needs and power in the global AI landscape. Understanding and leveraging the global AI supply chain complexity is a practical way for policymakers to develop effective strategic goals and diplomats to tailor partnerships and agreements.
2. Industrial policy as a master key for AI policy
Most countries frame AI policy in terms of sovereignty. However, the security and geopolitical aspects are not the only ones to consider. Countries in practice use industrial policy to fund numerous projects that build, scale and protect their power and positions in the global AI supply chain. One example of this is the Sovereign AI unit in the UK, which functions as a venture capital firm, investing in start-ups and companies to help them scale. The investment in Isomorphic Labs aims to cultivate the UK’s comparative advantage in AI for drug design and development. Vietnam’s strategy uses its existing manufacturing power to become a hub for assembly, packaging and testing (APT) for backend chipmaking support, aiming to increase its share in global APT services from one to eight percent in ten years.
Industrial policies should be at the centre of asking how to shape the AI supply chain. We build on the supply-side capability building, defensive management of dependencies, and demand-side coordination to identify where trade, investment and partnerships can reinforce countries’ positions in the wider AI network.
3. Coalition equilibrium for the future
Countries build many partnerships, agreements and alliances. We want to know how they are connected. ASEAN unites Southeast Asian countries in adopting a non-alignment approach amidst the geopolitical tension between China and the US (potentially a balanced coalition). The US banned China’s access to the Electronic Design Automation (EDA) software to decrease China’s chip design capability under the Foreign Direct Product Rule (FDPR) (a weaponised interdependence). The goal should be balanced interdependence for a trusted global AI future.
Balanced interdependence involves multiple countries collaborating and co-developing certain products. The resulting trusted minilateral coalitions should be built into national strategies. Each country excels in different elements. One example is the photonic supply chain for advanced AI chips. The French Soitec, Israeli Tower Semiconductor, US Lumentum and the Swedish Sivers Semiconductors specialise in substrate, wafer and laser, respectively. The interconnected collaboration through technology companies allows countries to participate in and shape global AI development. Critically, it keeps the leverage “evenly” distributed across countries, building more communities of shared interests and reducing risks of hegemony and monopoly.
Testing and categorising real-world developments against our framework will help understand how countries around the world are acting, responding and mediating the AI world order and offer suggestions for shaping national AI destinies that do not fall prey to the illusion of isolated sovereignty.
The minilateral future
The current fixation on a US-China AI order misreads the nature of modern technological/digital power. The global AI architecture is not an empire that can be conquered by a single superpower; it is a complex production and governance system where absolute self-reliance is an economic fiction that leads to heavy penalties. True strategic agency in the next decade belongs to the actors that reject the false dichotomy of choosing between submission and autarky.
This reality opens a window of opportunity for supply-chain coalitions. By shifting focus toward domain-specific, high-trust coalitions, agile states can transform their individual structural vulnerabilities into collective bargaining power. When distinct niche capabilities in supply chain are intentionally clustered into trusted minilateral alliances, they create a robust “balanced interdependence”. This structural equilibrium distributes leverage evenly enough to deter superpower coercion and effectively prevent market monopolies.
The views and opinions expressed in this post are those of the author(s) and not necessarily those of the Bennett School of Public Policy.
