Southeast Asian countries are seeking to adopt some measure of autonomy in the AI technology competition between China and the US. The question is whether such efforts can be successful.
In an ISEAS Perspective paper in October, John Lee argued for developing a niche outside the “integration of Artificial Intelligence (AI) technologies with robotics” in the technological competition between China and the US. Building on his paper, it should be noted that Southeast Asian nations are extending their Cold War-era political hedging strategies into the AI development race. They are seeking to balance between two great powers and develop a third pathway independent of either. Whereas Cold War-era hedging allowed Southeast Asian nations to develop a level of political autonomy, the current technological reality is much more diffuse and complex. It raises a key question: whether these early hedging efforts would be successful in developing or achieving full control over independent AI stacks.
It should be noted that great power competition in Southeast Asia is as present in the technological domain as in the diplomatic, economic and security domains. The US pursues this through initiatives like the American AI Exports Program, which forcefully endorses establishing the full American AI technology stack of hardware and software tools, data systems and models as the “gold standard for AI worldwide“. There are even voices arguing for an AI Marshall Plan, which takes a leaf from US efforts to rebuild Europe after World War Two.
Meanwhile, China is investing in and building domestic stack capabilities. It seeks to replace the US as the most dominant AI actor and shape international AI standards in the region. In response to this, and not wanting to be mere consumers of AI, Southeast Asian nations are exploring the feasibility of the “third stack” movement to utilise national leverage in AI development and reduce reliance on foreign AI products. This consists of domestic stack development, while simultaneously introducing new policy frameworks that aim to establish sovereignty over AI stacks and data streams.
Vietnam has introduced Southeast Asia’s first law on AI and the Law on the Digital Technology Industry. Hanoi seeks to establish greater autonomy over domestic stacks by classifying digital systems as strategic assets subject to localised data requirements. These policy advancements go hand-in-hand with a hedging approach based on strategic reconstruction. The state-linked firm FPT, for example, anchors its core compute capacity in the FPT AI Factory using US Nvidia chips while deploying models through a sovereign cloud termed the FPT Smart Cloud. Private companies have developed localised models such as PhoGPT – a Vietnamese language Large Language Model (LLM) developed by VinAI and Semikong — which was developed with the help of FPT software. Using the nation’s native language, these models are attuned to specific contexts and aim to reduce reliance on foreign LLMs. This modular system aims to balance between technological borrowing and the risk of dependency, demonstrating that Vietnam’s advantage lies in asymmetrical scaling and adaptation rather than complete self-sufficiency.
Similarly, Singapore seeks to use public policy as a tool to establish greater domestic control over AI stacks. In January, it launched the Model AI Governance Framework for Agentic AI, an update to the 2024 Generative AI policy. State-linked initiatives such as AI Singapore have developed SEA-LION family, an open, multimodal AI model to address the under-representation of Southeast Asian languages and cultures in frontier AI models. Malaysia, Indonesia and Thailand are aiming to develop greater control over domestic AI systems by hedging between US and Chinese AI investments and developments.
It is evident that national AI development strategies that rely on hedging strategies as a means to establish technological autonomy are only partially successful. Therefore, instead of pursuing nationally focused strategies, nations in the region would benefit from collaborative hedging.
These examples underscore some progress toward greater technological autonomy. It remains to be seen, however, to what extent Southeast Asian nations will be able to develop or achieve full control over independent AI stacks necessary to achieve AI sovereignty. They are still reliant on American or Chinese software and hardware and will remain so, as the economic and technological feasibility of developing domestic tools remains difficult, given the advanced economic and technological capabilities of Chinese and US firms. Jacob Taylor and Joshua Tan rightly note that national efforts are simply no match for “the scale of compute, talent or high-quality data of US or Chinese AI ecosystems”. In fact, efforts to develop a third pathway are more in line with localisation of existing (American and Chinese) AI models, supported by “technology transfers” from the same American and Chinese firms, creating domestically adapted platforms instead of domestically controlled AI stacks.
It is evident that national AI development strategies that rely on hedging strategies as a means to establish technological autonomy are only partially successful. Therefore, instead of pursuing nationally focused strategies, nations in the region would benefit from collaborative hedging. Taylor and Tan use Airbus as a template for successful regional technological development. This involves a public–private consortium of middle powers’ national AI laboratories which seek to build scalable AI products that can be sold under a public utility model. Even though the EU and ASEAN are notably different, the authors’ idea of an “Airbus for ASEAN AI” deserves some consideration.
First, hedging strategies are most effective when multiple states align. During the Cold War, Southeast Asian nations secured strategic autonomy through collective hedging rather than individual alignment. The same Cold War logic is just as relevant today. This means that at the diplomatic level, nations should be straightforward and firm in their goals and concerns, ensuring policy alignment that eschews ideological alignment with either the US or Chinese technological ecosystems. Instead, policy alignment is necessary and would help to maintain regulatory and infrastructure standards across the region.
Second, it would require a regional effort beyond intergovernmental alignment. Equally necessary is public-private cooperation in which states and regional firms build capabilities together. Whereas states can develop regionally aligned policies, public-private cooperation could develop regional chip and cloud capacity, and possibly even competitive foundation models. The sum of hardware and software development, in tandem with regionally aligned policies, would require significant and, at times, uncomfortable national choices. In the end, collaborative hedging by these means is the best way to avoid becoming further entrapped in Sino-US technological competition.
2026/237