Vietnam's AI Act Is a Wake-Up Call for Europe's Sovereign AI Debate

Anne Bajart, Head of Unit for Artificial Intelligence Policy at the European Commission's DG CONNECT, has noted publicly that the EU AI Act is designed to be risk-proportionate rather than infrastructure-prescriptive. Vietnam's law inverts that logic entirely: it is infrastructure-prescriptive first, and risk classification second. Whether that trade-off delivers better outcomes for citizens or simply better outcomes for the state remains the central question.

The 12 Billion Dollar Sovereign Stack

Vietnam's government launched its AI Vietnam 2030 strategy in 2024 with a 12 billion US dollar capital commitment. A national AI development fund, established under Decision No. 367/QD-TTg for the 2026 to 2027 period, co-finances startups, datacentres, and training programmes. AI startups pay 15 per cent corporate tax, roughly one-third the standard rate. Engineers qualify for fast-track Tech Visas. The government is not merely funding AI; it is actively shaping which companies and which problems receive support.

FPT Corporation, one of Vietnam's largest technology conglomerates, is sponsoring the Au Lac Grand Prize, a one million US dollar competition seeking Vietnamese AI innovations with measurable socio-economic impact. Submissions close 31/07/2026, with awards announced in November 2026. The eligibility criteria are explicit: innovations must be validated through real-world deployment, not laboratory benchmarks or venture-capital pitch decks. This is a pointed contrast with the speculative funding cycles that have characterised parts of the European startup ecosystem.

The comparison to EU industrial policy is instructive. The European Commission's AI Office, established in early 2024 to coordinate AI Act enforcement and strategic investment, is working alongside programmes such as Horizon Europe and the European High Performance Computing Joint Undertaking to build sovereign compute capacity. But European funding remains fragmented across member states, and no single EU country has matched the coherence or velocity of Vietnam's integrated policy package. Germany, France, and the Netherlands are all investing in AI infrastructure, but they are doing so under different frameworks, at different speeds, and with different state aid constraints.

What European AI Labs and Regulators Are Watching

Mistral AI, the Paris-based large language model developer that has positioned itself as Europe's answer to OpenAI, operates in a regulatory environment that is becoming more prescriptive but remains broadly permissive compared to Vietnam's framework. Mistral's Chief Executive Arthur Mensch has argued consistently that European regulation should enable competitive AI development, not constrain it. Vietnam's model, with its mandatory compute localisation and government-directed funding, represents a more interventionist alternative that some in Brussels are beginning to take seriously.

At ETH Zurich, researchers working on AI governance have pointed to Vietnam's law as an early example of what the academic literature calls "infrastructure sovereignty", the idea that meaningful AI independence requires not just domestic talent and data, but domestic compute. Professor Aurelia Tamara Merenmies and colleagues at the university's AI Centre have noted that data residency mandates, while administratively burdensome, can accelerate domestic capability building in ways that purely regulatory frameworks cannot. The European High Performance Computing Joint Undertaking's investment in EuroHPC supercomputers follows a similar logic, though without the private-sector compliance lever that Vietnam has introduced.

Vietnam's AI sector reportedly grew 187 per cent year-on-year in 2025, against 23 per cent growth in Silicon Valley over the same period. The country attracted 80 million US dollars in AI investment in 2025 and ranks third globally in AI trust according to recent surveys, behind Singapore and Japan. The EU, by contrast, has strong AI research credentials but has struggled to translate academic excellence into commercial AI leadership. The reasons are well-documented: fragmented capital markets, risk-averse procurement, and regulatory uncertainty during the AI Act's long phase-in period.

The Fragmentation Problem and What It Means for European Firms

For European technology companies with operations across multiple jurisdictions, Vietnam's law introduces a new compliance layer. Any EU-headquartered firm processing Vietnamese citizen data, whether through a local subsidiary, a SaaS product, or a data processing agreement, must now meet Vietnamese data residency requirements in addition to GDPR obligations and AI Act requirements. The two frameworks are not contradictory, but they are additive. Compliance teams that were already stretched by the EU AI Act's phased rollout now face a third major regime.

The broader lesson is one of fragmentation. As more countries enact their own AI laws with their own infrastructure mandates, the cost of global AI operations increases in ways that disproportionately affect mid-sized companies. A large European enterprise with a dedicated legal and compliance function can absorb Vietnam's requirements. A Series B startup expanding from Berlin to Ho Chi Minh City faces a materially different calculation. The EU has championed interoperability and mutual recognition in its trade agreements; it has not yet applied the same logic systematically to AI governance.

This fragmentation dynamic is precisely why the EU AI Act's international dimension matters. The European Commission's approach to third-country AI systems, and the AI Office's engagement with non-EU regulators, will determine whether the EU becomes a pole of regulatory convergence or simply one among many competing frameworks. Vietnam has chosen the latter path deliberately. Whether the EU can sustain the former is the defining question of the next regulatory cycle.

Speed as Signal

Vietnam's decision to implement on 1 March 2026 without extensions or grace periods sends a clear signal: regulatory velocity is a form of industrial policy. Countries that move slowly on AI governance are not being cautious; they are ceding first-mover advantage on the standards and infrastructure choices that will define the next decade of AI deployment. The EU passed its AI Act ahead of most major economies, but the phase-in timeline has allowed the implementation narrative to drift. Vietnam has demonstrated that a middle-income country with a fraction of the EU's administrative resources can move faster when political will is present.

For the European public sector in particular, Vietnam's model raises a pointed question. EU member state governments are among the largest potential deployers of AI systems, from social welfare processing to judicial decision support to public health analytics. The AI Act imposes significant obligations on high-risk public sector AI. But it does not compel governments to build sovereign AI capacity; it merely regulates what they deploy. Vietnam's approach does both simultaneously. Whether that combination produces better governance outcomes or merely tighter state control is a question Europe will need to answer before the next AI Act review cycle begins.