How Middle Powers Can Navigate the AI Duopoly

Artificial intelligence is consolidating around the world’s two technological superpowers. Far from democratising innovation, the economics of AI have reinforced a digital hierarchy in which the United States and China dominate every critical layer of the value chain, where systems are shaped by scale, capital, and deep integration that cannot easily be replicated. For middle powers, the challenge is no longer how to catch up, but how to remain relevant within a landscape where the foundations of AI are increasingly controlled by the two superpowers. The question that follows is not how to compete across every layer, but how to secure strategic influence within areas of the value chain that remain accessible and controllable.

China and the AI Stack

The Consolidated AI Value Chain. The AI stack is composed of several interdependent layers, with the most strategically decisive already closed to smaller nations. At the hardware level, semiconductor design and fabrication illustrate the depth of this consolidation. The United States and its allies dominate global semiconductor manufacturing equipment, while American firms such as Nvidia and AMD dominate chip design, with Nvidia alone capturing more than 90 per cent of the data-centre GPU market in 2023 (DCD, 2024). While the fabrication of these chips takes place primarily in Taiwan, through TSMC, who manufacture the majority of the world’s most advanced semiconductors, the United States consolidates its advantage through design leadership and control of chip design software.

China, by contrast, is investing heavily to localise its entire semiconductor supply chain, though it remains several generations behind (CETaS, 2024). For smaller nations, the capital required to participate meaningfully in this layer, often exceeding ten billion dollars for a single fabrication facility, renders entry effectively impossible. Cloud infrastructure reinforces this asymmetry. Cloud service providers (CSPs), such as Amazon, Google, and Baidu, already dominate the market, with CSPs likely to host between 60-65 percent of AI workloads as of 2030 (McKinsey, 2024). These companies alone plan to spend over 300 billion dollars on AI infrastructure in 2025, consolidating their advantage further (NY Times, 2025). In parallel, China has prioritised state-backed cloud expansion through firms such as Alibaba Cloud and Huawei, securing capacity on a national scale.

The Control of Data

The control of data and foundational models adds another layer of structural dominance. American companies benefit from the global reach of English, allowing them to collect vast and high-quality datasets. Whilst Chinese companies draw on a domestic population of 1.4 billion, generating unmatched volumes of Chinese-language data.

These structural advantages are reinforced over time through scale effects and feedback loops, ensuring that the most decisive layers of the AI stack are effectively foreclosed to smaller nations. To be clear, other states do play critical roles within certain niches. Taiwan remains indispensable for chip fabrication, Japan leads in robotics, and the European Union has assumed a global role in regulation.

Yet these contributions are partial and dependent on upstream infrastructure that is ultimately controlled by the United States and China. This is not simply a matter of temporary leadership but a structural outcome of population size, capital intensity, and network effects. For middle powers, the conclusion is inescapable: sovereignty across the entire AI value chain is unattainable. In order for middle powers to compete, they must recognise their constraints and instead focus on how to add meaningful value within the current ecosystem.

Strategic Relevance for Middle Powers

The experience of Singapore illustrates how smaller nations can carve out significance despite these constraints. Singapore is poorly positioned for leadership in the hardware and compute layers. Its population of only 5.4 million, and a tropical climate that makes compute-intensive operations costly. Yet rather than pursuing unattainable self-sufficiency, the city-state has deliberately targeted downstream layers where it can exert leverage.

Singapore’s National AI Strategy 2.0, announced in 2023 with $743 million committed, builds on decades of methodical digital planning dating back to the 1980 National Computerisation Plan (EDB, 2024). It aims to triple the country’s AI practitioners to 15,000 within five years and already supports more than 200 AI deployments across healthcare, transport, and urban planning (Financial Times, 2025).

The government itself functions as a production testbed, creating demand that de-risks enterprise adoption while fostering practical expertise. This institutional efficiency allows AI firms to incorporate in less than 24 hours. It is supported by regulatory clarity through instruments such as the Personal Data Protection Act. The success of this has quickly become clear with Singapore’s government-supported R&D spending as a share of GDP now exceeding the United States by a factor of eighteen (CSET, 2023).

Singapore Technologies

At the same time, Singapore has cultivated a role as a neutral hub where both American and Chinese firms operate. Google, Meta, and Salesforce run research centres alongside Alibaba and Huawei, enabling knowledge transfer and sustained investment from both ecosystems. This positioning within the coordination layer of the value chain creates a unique advantage. Singapore becomes the platform where frontier technologies are significant in real-world contexts, and global talent congregates without triggering geopolitical tensions.

This model is not replicable in its entirety. Singapore’s efficiency, wealth, and geopolitical position are exceptional. Yet the broader lesson is clear. Middle powers cannot hope to match the superpowers in computing or data. Yet, it can secure roles of consequence by focusing on governance, regulation, applied innovation, and specialised research. Some may become convening hubs for international collaboration. Others may cultivate niches in ethics, talent development, or particular applications aligned with national strengths. What matters is the recognition that sovereignty is impossible, but strategic relevance is achievable.

Ultimately, there will likely not be a multipolar AI world anytime soon. The United States and China have consolidated the foundational layers of the value chain. Yet this does not consign middle powers to irrelevance. By identifying areas where they can exercise influence, they can still shape the global AI landscape in meaningful ways. Singapore demonstrates that it can maximise the constraints rather than denying them, offering a lesson that others would do well to heed.

Bibliography

• Centre for Security and Emerging Technology (2024) ‘Examining Singapore’s AI Progress’. Available at:
  https://cset.georgetown.edu/wp-content/uploads/CSET-Examining-Singapores-AI-Progress.pdf
• Chandler, C. (2024) ‘Singapore’s AI ambitions: How the city-state is keeping up in an arms race dominated
  by the US and China’. EDB Singapore. Available at:
  https://www.edb.gov.sg/en/business-insights/insights/singapores-ai-ambitions-how-the-city-state-is-keeping
  -up-in-an-arms-race-dominated-by-the-us-and-china.html
• Data Center Dynamics (2024) ‘Nvidia GPU shipments totaled 3.76m in 2023, report’. Available at:
  https://www.datacenterdynamics.com/en/news/nvidia-gpu-shipments-totaled-376m-in-2023-equating-to-a-9
  8-market-share-report/
• Financial Times (2024) ‘Singapore: where trade meets innovation and digitalisation’. Available at:
  https://www.ft.com/partnercontent/enterprise-singapore/singapore-where-trade-meets-innovation-and-digita
  lisation.html
• Janjeva, A., Baek, S., and Sellars, A. (2024) ‘China’s Quest for Semiconductor Self-Sufficiency: The impact
  on the UK and Korean industries. CETAS Briefing Paper. Available at:
  https://cetas.turing.ac.uk/publications/chinas-quest-semiconductor-self-sufficiency
• McKinsey (2024) ‘AI power: Expanding data center capacity to meet growing demand. Available at:
  https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/ai-power-ex
  panding-data-center-capacity-to-meet-growing-demand
• New York Times (2025) ‘The Global AI Divide’. Available at:
  https://www.nytimes.com/interactive/2025/06/23/technology/ai-computing-global-divide.html