In several of his public talks, Jensen Huang has mentioned a term: Sovereign AI.

▶ Listen to summary
AI-synthesized voice, cloned from the author's own voice

When most people hear the word “sovereignty,” what comes to mind is territory, armies, diplomacy. But today, as AI is deeply embedded in every nation’s economy, defense, and public services, the definition of sovereignty is being rewritten.

A nation that cannot autonomously develop, deploy, and control its AI systems has incomplete sovereignty — no matter how strong its army or how vast its territory.

Four Layers of Autonomy

Sovereign AI is not a single concept. It has four layers, and none can be missing.

Technological sovereignty. Can you build the core technology yourself — chips, high-performance computing, AI frameworks? If your AI systems run on another nation’s chips, the moment they cut off supply, you’re paralyzed. Taiwan’s semiconductor industry holds a unique advantage at this layer, but this advantage is not eternal; it must be sustained through continuous investment.

Data sovereignty. Where is your data stored? Who can access it? If a nation’s medical data, financial data, and demographic statistics are all stored in foreign clouds, that’s not digitalization — it’s digital colonialism. Data sovereignty means controlling the storage and processing of data, protecting sensitive information from external access.

Algorithmic sovereignty. AI models are not neutral. They reflect the biases and values of their training data. If a nation’s public AI systems (education, healthcare, justice) all use foreign-trained models, the values embedded in those models may be entirely at odds with the needs of the home society. Algorithmic sovereignty means ensuring AI models are transparent, controllable, and aligned with the nation’s interests and social values.

Application and service sovereignty. Command over the AI infrastructure of critical public services. If your traffic management system, power grid control system, and defense simulation systems rely on foreign platforms, that’s not cooperation — it’s dependency.

Together, these four layers constitute complete Sovereign AI. Lacking any one layer leaves a breach in sovereignty.

From Oil to Data

In the last century, the foundation of national power was oil and minerals. Whoever controlled energy controlled the world order.

In this century, data and compute are taking the place of oil.

In “The Negentropy Strategy: The Survival Logic of Taiwanese Enterprises,” I discussed how enterprise survival in Taiwan requires establishing order amid chaos. The same applies at the national level — in the chaos of AI geopolitics, establishing one’s own order of technological autonomy is a precondition for survival.

The difference is this: the geographic location of oil is fixed — you either have it or you don’t. Data and compute can be constructed — as long as you have the will and the capacity to invest. For a nation like Taiwan, which has no oil but does have technological capability, this is actually an opportunity.

But opportunity is not a guarantee. If Taiwan does not proactively build its own AI infrastructure, this window of opportunity will be seized by others.

Technological Autonomy vs. Technological Authoritarianism

The development of Sovereign AI has a dangerous dark side.

When a nation vigorously develops AI in the name of “technological autonomy,” how do we ensure these AIs are not used to surveil the people, suppress dissent, or manipulate public opinion? China’s social credit system is one example — technically, it is a realization of “Sovereign AI,” but ethically, it is a demonstration of technological authoritarianism.

This is the dilemma every nation pursuing Sovereign AI must face: you need enough technological control to protect national security, but if that control lacks the checks and balances of democratic institutions, it can easily slide into abuse.

In “Safer-4 and the Future of Technology Governance,” I discussed how the core problem of AI governance is not technology, but the distribution and balancing of power. The same goes for Sovereign AI — technological autonomy is necessary, but it must be embedded within a democratic, transparent, and accountable institutional framework, lest it become another form of oppression.

Taiwan’s Position

Taiwan occupies a unique position on the chessboard of Sovereign AI.

Our semiconductor manufacturing capability gives us an inherent advantage at the first layer of technological sovereignty. But at the other three layers — data sovereignty, algorithmic sovereignty, and application and service sovereignty — our investment falls far short.

Our public AI applications rely heavily on foreign platforms. Our data governance framework is still immature. Our AI talent continues to flow abroad.

Taiwan does not lack technological capability — what it lacks is strategic-level integrative thinking: extending the semiconductor advantage across the full AI stack, from chips to models to applications to data governance, to build a complete Sovereign AI system.

This is not something a single company can do, nor something a single ministry can do. It requires strategic will at the national level.

A Civilizational Choice

Sovereign AI is not merely technological competition — it is a civilizational choice.

Do you choose to depend on a great power’s AI ecosystem, or to build your own? Do you choose technological control to safeguard security, or democratic checks to prevent abuse? Do you choose to hand your data to the most efficient platform, or insist on keeping it where you can command it?

Every choice carries a cost. But not choosing — letting things unfold on their own — carries the greatest cost of all. Because the direction of “natural development” always tilts toward the concentration of power and capital.

To establish true autonomy in the digital age requires not only technological investment, but a nation’s clear-eyed answer to the question of “what do we want to become.”