Every nation wants AI sovereignty. Only two have anything close to it. The rest are buying "sovereignty as a service" from the countries they're trying to become independent from.
Audit your AI supply chain. If training data, compute, or models depend on foreign providers, you don't have sovereignty—you have dependency.
Updated January 2026: Added Sovereignty Dependency Audit for supply chain risk assessment.
Digital sovereignty has become a shared global instinct. A new World Economic Forum paper examines how economies can strengthen AI competitiveness through strategic investment choices and trusted international partnerships. Countries that disagree on everything else agree on this: they need their own AI capabilities. The problem is that AI sovereignty requires controlling an entire stack of capabilities that almost no nation possesses.
The Full-Stack Reality
Only two nations, the United States and China, enjoy anything close to full-stack control. This means chip design, chip fabrication, hyperscale cloud infrastructure, and frontier model development all happening within their borders.
Everyone else connects at various points along global supply chains. This determines where they sit in a digital hierarchy, rendering sovereignty more mirage than reality.
The AI stack has multiple chokepoints. Nvidia GPUs are essential for training AI models. As CEO Jensen Huang has noted, "Nvidia GPU is the only platform that's available to everybody." That availability comes with lasting vendor lock-in to a U.S. company.
Sovereignty as a Service
Major tech companies have identified a lucrative market: selling "sovereignty as a service" to governments. Microsoft, Amazon Web Services, Nvidia, and Huawei all offer partnerships that promise self-sufficiency while actually embedding deeper dependencies.
Researcher Rui-Jie Yew frames the question correctly: "Are you selling your chips and calling it a day, or are you using your dominant position to bundle additional services that rope your clients into ongoing dependencies?"
The answer, consistently, is the latter. Each "sovereign AI" deal creates new touchpoints for foreign influence rather than reducing them. I've seen this pattern in vendor relationships across domains. The promise of independence becomes a mechanism for deeper capture.
When Export Controls Bite
The fragility of these arrangements becomes visible when geopolitics intervenes. Kazakhstan's supercomputer project was delayed due to U.S. export licensing holds on Nvidia shipments. Malaysia initially announced a Huawei sovereign AI partnership, then retracted it under U.S. pressure, later unveiling a less-capable domestic edge chip instead.
These aren't edge cases. They're the normal functioning of a system where "sovereign" AI depends on hardware controlled by foreign governments with their own strategic interests.
Countries without compute and energy capacity risk becoming rule-takers, even if they write ambitious AI laws. The legislation means nothing if the infrastructure runs on someone else's chips.
The Scale of What's Required
Building real AI sovereignty requires staggering investment. Europe is racing to build AI infrastructure with several multi-gigawatt projects: MGX and Mistral AI's Campus in France (1.4 GW), SINES in Portugal (1.2 GW), and the U.K.'s AI Growth Zone (~1.1 GW).
This sounds impressive until you compare it to the U.S., which has a 25-GW pipeline of announced AI infrastructure projects. As McKinsey analysis notes, 71% of executives characterize sovereign AI as an "existential concern" or "strategic imperative." Europe's entire effort amounts to a few gigawatts, a rounding error in the global race.
Canada has committed $2 billion to a Sovereign AI Compute Strategy. Brazil is spending approximately $4 billion under its AI Plan 2024-2028. Japan is building ABCI 3.0 with 6 AI exaflops of performance. These are serious investments that still won't eliminate foreign dependencies.
Even Advanced Economies Can't Escape
South Korea has robust domestic technology industries, world-class semiconductor manufacturing, and strong research institutions. Despite this, South Korean companies still train AI models on Nvidia GPUs and develop data centers through AWS.
If South Korea can't achieve AI independence, the path for most nations is even harder. The layer tax in technology stacks compounds: each dependency creates vulnerabilities that cascade upward.
Researcher Sam Winter-Levy states the uncomfortable truth: "For most states, sovereign AI is 'a very, very expensive proposition'" with the reality that nations "still won't be able to eliminate dependencies and vulnerabilities on foreign states."
A More Realistic Approach
India offers a pragmatic alternative model. Rather than attempting full-stack sovereignty, India focuses on a single component: language-specific large language models. The IndiaAI Mission and India Stack prioritize strategic control over data, digital identity, and foundational digital services.
This targeted approach acknowledges reality. You can't compete across the entire supply chain. But you can identify which components matter most for your national interests and focus there.
Japan provides another instructive example. Rather than pursuing full independence, Japan has negotiated privileged access agreements with U.S. chip manufacturers while investing heavily in specific application domains like robotics and manufacturing AI. The Japanese approach treats sovereignty as a portfolio problem: accept dependencies in some layers while building defensible positions in others. It's less rhetorically satisfying than "complete independence" but more achievable.
The question isn't "how do we become independent" but "where should we prioritize resilience, and what dependencies are acceptable?" That's a harder conversation than nationalist rhetoric allows, but it's the only honest one.
The Data Sovereignty Alternative
There's a quieter approach some nations are taking: controlling data instead of infrastructure. If you can't build the chips or train the frontier models, you can at least govern how your citizens' data gets used.
Europe's GDPR and AI Act focus on this layer. The infrastructure might be American, but the rules governing what can be done with European data are European. This is sovereignty through regulation rather than technology.
It's not as satisfying as full technological independence. But it might be more achievable. Setting rules requires legislative capacity and enforcement mechanisms. Building AI infrastructure from scratch requires capital, expertise, energy, and supply chains most countries simply don't have.
The limitation is enforcement. Data sovereignty only matters if you can detect violations and impose consequences. When the AI providers are foreign entities operating at global scale, enforcement becomes complicated fast. Regulations without teeth are just suggestions.
Still, it's an option. And for nations without the resources for technological sovereignty, regulatory sovereignty might be the only realistic path available.
The Tipping Point
This year marks a transition. Major EU AI obligations begin applying. The U.S. is rewiring export controls. China is hardening security-first oversight. The rules governing AI sovereignty are crystallizing.
Nations making decisions now are choosing their position in the emerging hierarchy. Those pursuing the illusion of total independence will waste resources. Those taking a pragmatic approach, prioritizing targeted resilience and strategic leverage points, might actually achieve meaningful autonomy in the areas that matter most.
Regional coalitions offer another path. Collective bargaining through aligned nations can create leverage that individual countries lack. But this requires admitting that sovereignty isn't something you buy from a cloud provider.
The EU's approach illustrates both the potential and limitations. By pooling demand and setting common standards, Europe creates market power that individual member states lack. But the infrastructure gap with the U.S. remains vast, and regulatory leverage only works if there are alternatives to regulate toward.
Sovereignty Dependency Audit
Score your organization's AI supply chain exposure. Click your dependency level for each layer:
The Strategic Question: For each layer scoring 2-3, ask: "What happens to our AI capabilities if this supply chain is interrupted tomorrow?" If you can't answer, that's where your sovereignty strategy should focus.
The Bottom Line
AI sovereignty is a strategic imperative that almost no nation can actually achieve. The infrastructure requirements are too vast, the dependencies too deep, and the technological chokepoints too concentrated in two countries.
Countries pursuing sovereign AI should ask hard questions: What specific capabilities do we actually need? Which dependencies are acceptable? Where can regional cooperation substitute for national capacity? And what would we do if export controls cut us off tomorrow?
The honest answer is that most nations will remain dependent on U.S. or Chinese technology for the foreseeable future. The question is whether they structure that dependency intelligently or pretend it doesn't exist while writing sovereignty strategies on hardware controlled by foreign powers.
"Countries without compute and energy capacity risk becoming rule-takers, even if they write ambitious AI laws."
Sources
- Rest of World: Chinese and U.S. tech keeps countries dependent on foreign AI — Analysis of sovereignty-as-a-service model
- McKinsey: The sovereign AI agenda - Moving from ambition to reality — National strategy analysis
- IDC: AI Sovereignty - National Economic Competitiveness and Security — Infrastructure requirements
Technology Strategy
Understanding AI infrastructure dependencies requires seeing through vendor promises. Assessment from someone who's evaluated technology stacks across industries.
Get Assessment