The Countries Behind AI's Chokepoints
In my last post I presented that AI looks broad at the application layer. Underneath it runs on a small number of chokepoints. This post names them by country.
Taiwan’s TSMC makes the most advanced chips in the world. The Netherlands’ ASML makes the machines that make those chips. The U.S. dominates EDA software through Synopsys and Cadence; without which chip design does not happen. Nvidia owns the compute stack and the CUDA ecosystem that most AI training runs on. South Korea supplies the memory through SK Hynix and Samsung. Japan supplies the materials and equipment the entire chain depends on. That is the foundation, and it is spread across a handful of countries.
No country controls all of it; not even the U.S. America leads across compute, hyperscalers, frontier models, capital, and R&D, but it still depends on Taiwan, South Korea, and Japan at the semiconductor layer. Full AI sovereignty does not exist today. Every country has dependency somewhere underneath.
Governance is another way nations exercise control, and the U.S. and EU use it very differently. The U.S. export restrictions on Nvidia GPUs to China are not a trade dispute. They are a strategic attempt to limit China’s ability to scale advanced AI. The U.S. increasingly appears to be applying a similar logic to that used during the Cold War around strategic technology restrictions. The goal is the same; slow a rival’s ability to build capability. The EU does not own the deep infrastructure layers. It owns access to its market; the world’s largest economic bloc. Anyone who wants large-scale AI adoption in Europe will need to comply with its frameworks. That is leverage of a different kind; it is still leverage.
India participates heavily in AI; software talent, digital public infrastructure, enterprise services, a large workforce. But participation is not control, as India does not own semiconductors, advanced compute, hyperscalers, or frontier models. It builds on infrastructure owned elsewhere. Saudi Arabia and the UAE are positioning differently; capital, energy, and large-scale datacenter investment. But even they rely on semiconductors, compute, and hyperscaler ecosystems they do not control. Countries that do not own the chokepoints localize AI models to retain data sovereignty and some local control. That is a reasonable move, though it does not change dependency underneath.
The line between commercial AI and national security AI is gone. Surveillance, cybersecurity, intelligence, cross-border monitoring; these are no longer separate from the AI stack. Whoever controls the chokepoints controls infrastructure that is now a strategic asset for every nation.
China does not have end-to-end control today. It remains dependent on compute and hyperscaler infrastructure it does not own. But it is playing a long game. SMIC is a state-backed attempt to build domestic semiconductor capability. DeepSeek shows what China can do at the model layer when pushed, as it has already put pressure on frontier model providers to justify their pricing and value proposition, particularly among startups. The strategy is not to win immediately. It is to reduce dependency on external chokepoints, layer by layer, through domestic substitution and state-backed industrial policy. That will take time. It is happening. COVID made one thing impossible to ignore. Globalization creates structural vulnerabilities. The U.S. learned that manufacturing is not just an economic question. It is a chokepoint question. I wrote about this in detail in my first post: The U.S. Isn’t Rebuilding Production. It’s Managing Risk.
Looking ahead…..
The U.S. holds the strongest position across the stack and is unlikely to lose it quickly. China is closing gaps where it can; models first, infrastructure over time. The East Asian semiconductor ecosystem remains the foundation the global AI stack runs on. The EU, India, and the Gulf exercise influence at different layers; governance, deployment, capital/ energy, respectively, but not at the deepest infrastructure layers. The countries that already control the deepest chokepoints have structural advantages that do not disappear fast. Everyone else is managing dependency, not eliminating it.