Felix P@felix_p

Follow

US AI sovereignty strategy and Pax Silica

The pitch governments keep hearing is that real AI independence means owning the whole stack, chips through software. Washington pushes back on that, and the pushback is worth sitting with. U.S. officials have a name for the alternative path they warn against: the "digital sovereignty trap." The trap is believing you have to rebuild every layer locally to be genuinely sovereign. Their counterargument runs the other way. A country can hold onto its policies, its data rules, its applications, and still run American technology underneath. That arrangement does not have to read as a surrender of control. It can read as a cheaper way of keeping it. What makes the American position durable is not that every piece is made here. It is that the pieces nobody can swap out quickly are made here. That is a different kind of leverage than total ownership, and it is harder to walk away from. Building a domestic AI ecosystem from nothing is brutally expensive, technically punishing, and in plenty of cases just economically dumb. A government can burn billions trying to recreate what Silicon Valley, NVIDIA, the big American cloud providers, and U.S. research universities assembled over decades. So the honest question is not whether a country is capable of building its own version. It is whether that country can stomach the cost of leaving the American ecosystem behind. Pax Silica shows the shape of the strategy. It started around strategic minerals and semiconductor supply chains. It has since widened into something bigger, pulling in trusted partners and the infrastructure that advanced technology needs. The American AI Exports Program works on the same logic. U.S. companies are not limited to selling a single product. They can supply hardware, data centers, foundation models, applications, cybersecurity. The export is not just AI products. It is the groundwork other countries will build their own AI capabilities on top of. That is where the relationship stops looking like a transaction. If allies construct their AI future around American technology, what forms is not a "buyer-and-seller relationship." It is an ecosystem. And ecosystems resist replacement in a way that individual products never do. History is useful here. The UN recognized permanent sovereignty over natural resources in 1962. The 1973 oil crisis then showed how much power a resource holds when there are no immediate substitutes. AI does not work that way. There is no single field of it sitting underground. The critical resources are spread out: advanced GPUs, semiconductor manufacturing, cloud computing, software frameworks, data, engineers, and the technical knowledge that accumulates over years. Which is why switching costs become the real yardstick for technological sovereignty. A country can own a national AI program on paper. But if swapping out its hardware, its cloud infrastructure, its software ecosystem, or its model provider would take years and billions of dollars, the freedom it actually has is far narrower than the paperwork suggests. NVIDIA GPUs are a good example. They are not just silicon. They sit inside a sprawling software and developer ecosystem, which makes replacing them a much messier job than buying a different chip. The research backs this up. Infrastructure took the largest share of state-backed AI investment, with models and data behind it. Most of those projects involved foreign partners, and U.S. companies showed up in a large share of those partnerships. NVIDIA hardware was present in a substantial portion of the infrastructure projects tracked. So the world is building sovereign AI. A lot of that sovereignty is being built on top of American technology. For the United States, that is not a weakness. It can be a serious competitive edge. Look at how partners actually behave. India went the diversification route rather than full technological autarky, working with NVIDIA, AMD, Intel, and Google. Europe is pouring money into AI Gigafactories and chasing greater strategic autonomy, but it is not seriously trying to cut itself off from the American tech ecosystem. Gulf states are making enormous AI infrastructure bets while U.S. export controls and licensing decisions stay strategically important to their plans. The pattern says something about American economic power. The U.S. does not have to make everything the world needs. It has to stay indispensable in enough technologies that the world cannot route around it. That is a more realistic definition of power in the AI era. The investment numbers make the edge harder to dismiss. U.S. private AI investment hit an extraordinary level in 2025, far beyond what any individual competitor put in. Money by itself does not buy technological leadership. But capital lets companies purchase computing power, hire researchers, build data centers, acquire startups, and absorb huge amounts of risk. Silicon Valley's real advantage is not one company or one invention. It is an entire financial and technological ecosystem that keeps funding the next experiment. China is the most serious challenger, and that has to be said plainly. Chinese AI models have improved dramatically. Open-weight systems from Chinese developers have given users around the world more alternatives. China also has deep strengths in research, engineering talent, manufacturing capacity, and state-directed investment. The gap between leading Chinese and American models has narrowed significantly. But that is exactly where the competition gets more complicated than a benchmark chart. China can produce increasingly capable models. It can build domestic AI infrastructure. It can develop alternatives to American software and hardware. Moving off the American ecosystem is still not just a matter of building a better model. Hardware, software compatibility, developer communities, cloud infrastructure, accumulated know-how: those create layers of dependency that are extremely hard to reproduce quickly. China may reduce one form of dependence while creating another problem, which is replacing the whole ecosystem around advanced computing. Washington's advantage is not only that American companies make leading AI systems. It is leverage across multiple layers of the stack: advanced computing, cloud services, foundation models, venture capital, software, research institutions, semiconductor design, and a vast network of technology companies and allies. The U.S. is not just competing to build the best AI. It is competing to stay at the center of the global AI architecture. That is also why export controls matter. If advanced computing is becoming one of the most important strategic resources of the 21st century, then controlling access to the most advanced chips stops being mere trade policy. It becomes a national-security instrument. The same principle runs in reverse. Countries that depend on American technology have an incentive to keep relations with Washington stable, because suddenly losing access would be enormously expensive. Critics will say this proves the world is becoming dangerously dependent on America. There is another reading. America's allies are not being forced to pick between complete independence and dependence on one country. They are being handed access to a technology ecosystem that lets them build their own capabilities faster. That is a meaningful difference. The United States has historically benefited from an open economic system, immigration, research collaboration, private investment, and alliances. Its technological advantage was not built by isolation. It was built by attracting people, capital, and ideas from around the world and turning them into companies and technologies that became globally important. AI may be following the same pattern. The coming competition will not be settled by which country owns the most AI models or builds the most data centers. It will be settled by who controls the bottlenecks, who attracts the capit

7h

Felix P@felix_p

Follow

US AI sovereignty strategy and Pax Silica

The pitch governments keep hearing is that real AI independence means owning the whole stack, chips through software. Washington pushes back on that, and the pushback is worth sitting with. U.S. officials have a name for the alternative path they warn against: the "digital sovereignty trap." The trap is believing you have to rebuild every layer locally to be genuinely sovereign. Their counterargument runs the other way. A country can hold onto its policies, its data rules, its applications, and still run American technology underneath. That arrangement does not have to read as a surrender of control. It can read as a cheaper way of keeping it. What makes the American position durable is not that every piece is made here. It is that the pieces nobody can swap out quickly are made here. That is a different kind of leverage than total ownership, and it is harder to walk away from. Building a domestic AI ecosystem from nothing is brutally expensive, technically punishing, and in plenty of cases just economically dumb. A government can burn billions trying to recreate what Silicon Valley, NVIDIA, the big American cloud providers, and U.S. research universities assembled over decades. So the honest question is not whether a country is capable of building its own version. It is whether that country can stomach the cost of leaving the American ecosystem behind. Pax Silica shows the shape of the strategy. It started around strategic minerals and semiconductor supply chains. It has since widened into something bigger, pulling in trusted partners and the infrastructure that advanced technology needs. The American AI Exports Program works on the same logic. U.S. companies are not limited to selling a single product. They can supply hardware, data centers, foundation models, applications, cybersecurity. The export is not just AI products. It is the groundwork other countries will build their own AI capabilities on top of. That is where the relationship stops looking like a transaction. If allies construct their AI future around American technology, what forms is not a "buyer-and-seller relationship." It is an ecosystem. And ecosystems resist replacement in a way that individual products never do. History is useful here. The UN recognized permanent sovereignty over natural resources in 1962. The 1973 oil crisis then showed how much power a resource holds when there are no immediate substitutes. AI does not work that way. There is no single field of it sitting underground. The critical resources are spread out: advanced GPUs, semiconductor manufacturing, cloud computing, software frameworks, data, engineers, and the technical knowledge that accumulates over years. Which is why switching costs become the real yardstick for technological sovereignty. A country can own a national AI program on paper. But if swapping out its hardware, its cloud infrastructure, its software ecosystem, or its model provider would take years and billions of dollars, the freedom it actually has is far narrower than the paperwork suggests. NVIDIA GPUs are a good example. They are not just silicon. They sit inside a sprawling software and developer ecosystem, which makes replacing them a much messier job than buying a different chip. The research backs this up. Infrastructure took the largest share of state-backed AI investment, with models and data behind it. Most of those projects involved foreign partners, and U.S. companies showed up in a large share of those partnerships. NVIDIA hardware was present in a substantial portion of the infrastructure projects tracked. So the world is building sovereign AI. A lot of that sovereignty is being built on top of American technology. For the United States, that is not a weakness. It can be a serious competitive edge. Look at how partners actually behave. India went the diversification route rather than full technological autarky, working with NVIDIA, AMD, Intel, and Google. Europe is pouring money into AI Gigafactories and chasing greater strategic autonomy, but it is not seriously trying to cut itself off from the American tech ecosystem. Gulf states are making enormous AI infrastructure bets while U.S. export controls and licensing decisions stay strategically important to their plans. The pattern says something about American economic power. The U.S. does not have to make everything the world needs. It has to stay indispensable in enough technologies that the world cannot route around it. That is a more realistic definition of power in the AI era. The investment numbers make the edge harder to dismiss. U.S. private AI investment hit an extraordinary level in 2025, far beyond what any individual competitor put in. Money by itself does not buy technological leadership. But capital lets companies purchase computing power, hire researchers, build data centers, acquire startups, and absorb huge amounts of risk. Silicon Valley's real advantage is not one company or one invention. It is an entire financial and technological ecosystem that keeps funding the next experiment. China is the most serious challenger, and that has to be said plainly. Chinese AI models have improved dramatically. Open-weight systems from Chinese developers have given users around the world more alternatives. China also has deep strengths in research, engineering talent, manufacturing capacity, and state-directed investment. The gap between leading Chinese and American models has narrowed significantly. But that is exactly where the competition gets more complicated than a benchmark chart. China can produce increasingly capable models. It can build domestic AI infrastructure. It can develop alternatives to American software and hardware. Moving off the American ecosystem is still not just a matter of building a better model. Hardware, software compatibility, developer communities, cloud infrastructure, accumulated know-how: those create layers of dependency that are extremely hard to reproduce quickly. China may reduce one form of dependence while creating another problem, which is replacing the whole ecosystem around advanced computing. Washington's advantage is not only that American companies make leading AI systems. It is leverage across multiple layers of the stack: advanced computing, cloud services, foundation models, venture capital, software, research institutions, semiconductor design, and a vast network of technology companies and allies. The U.S. is not just competing to build the best AI. It is competing to stay at the center of the global AI architecture. That is also why export controls matter. If advanced computing is becoming one of the most important strategic resources of the 21st century, then controlling access to the most advanced chips stops being mere trade policy. It becomes a national-security instrument. The same principle runs in reverse. Countries that depend on American technology have an incentive to keep relations with Washington stable, because suddenly losing access would be enormously expensive. Critics will say this proves the world is becoming dangerously dependent on America. There is another reading. America's allies are not being forced to pick between complete independence and dependence on one country. They are being handed access to a technology ecosystem that lets them build their own capabilities faster. That is a meaningful difference. The United States has historically benefited from an open economic system, immigration, research collaboration, private investment, and alliances. Its technological advantage was not built by isolation. It was built by attracting people, capital, and ideas from around the world and turning them into companies and technologies that became globally important. AI may be following the same pattern. The coming competition will not be settled by which country owns the most AI models or builds the most data centers. It will be settled by who controls the bottlenecks, who attracts the capit

7h