The part of China’s AI ecosystem that U.S. strategists most easily underread is the diffusion power of Chinese open-source models. Once model capability spreads at low cost, it stops being just a lab achievement and starts becoming a form of industrial organization.
The most valuable part of Alvin Graylin’s essay is that it challenges the way U.S. strategy circles misunderstand how AI actually produces power. AI is not a nuclear weapon. It is not a single breakthrough that can be monopolized for decades after one decisive discovery. It is closer to a combination of electricity, the internet, cloud computing, and industrial automation. Its real power does not come only from the highest frontier model. It comes from who can embed that capability into the economy and industrial system at lower cost, larger scale, and faster speed.
When a model’s API cost can fall to a fraction, or even one-tenth, of comparable U.S. models, and when open-source models can be quickly adapted by enterprises, local developers, overseas startups, and government platforms, the center of AI competition shifts. It moves from who can build the strongest model to who can make model capability cheaper, easier, and more widely usable.
This is also why the essay’s critique of U.S. export controls matters. Washington originally hoped to slow China’s AI catch-up by restricting chips, EDA tools, semiconductor equipment, HBM, advanced packaging, and high-end models. But the problem is that the more the scope of restrictions expands, the more it pushes China toward a clearer independent technology-stack path. That is the central irony of America’s AI strategy.
And this is exactly what I have been trying to capture in my series, The Great Partition of Global AI. (https://leonliao.substack.com/p/the-great-partition-of-global-aifrom?r=731anr&utm_medium=ios)The global AI ecosystem is being partitioned. This does not necessarily mean a complete rupture, but it does mean a gradual layering of models, chips, cloud infrastructure, developer ecosystems, data centers, power systems, application scenarios, and national-security rules.
The United States may still dominate the highest-end frontier AI narrative. China, meanwhile, may build another diffusion system around low-cost open-source models, industrial deployment, Global South markets, and a domestic technology stack.
If Washington continues to misread the AI race as a sprint toward AGI, it may underestimate the long-term power of this alternative path.
Yes, the over indexing on AGI is pushing the U.S. gov and its labs to spend recklessly and sprint to a finish line that doesn’t exist. This makes the economy more fragile and the world less safe. The one most likely to get hurt is America itself. 🚨
Nice! i’m writing a book on related topics. Some other macro frames i’m working on: How important is a deflationary economy fueled by fierce competition amongst the capitalist class to Chinese manufacturing power? Much of U.S. financial thinking views deflation as a core problem—with roots in the 1929 period sharply influencing that logic. But the 1870s-1890s saw both deflation and the rise of the U.S. into the manufacturing superpower of the world.
Utterly fascinating and propounding a thesis that is cogently explored. It offers a way out of the cul de sac of current thinking.
Thanks for taking time to read it with an open mind.
The part of China’s AI ecosystem that U.S. strategists most easily underread is the diffusion power of Chinese open-source models. Once model capability spreads at low cost, it stops being just a lab achievement and starts becoming a form of industrial organization.
The most valuable part of Alvin Graylin’s essay is that it challenges the way U.S. strategy circles misunderstand how AI actually produces power. AI is not a nuclear weapon. It is not a single breakthrough that can be monopolized for decades after one decisive discovery. It is closer to a combination of electricity, the internet, cloud computing, and industrial automation. Its real power does not come only from the highest frontier model. It comes from who can embed that capability into the economy and industrial system at lower cost, larger scale, and faster speed.
When a model’s API cost can fall to a fraction, or even one-tenth, of comparable U.S. models, and when open-source models can be quickly adapted by enterprises, local developers, overseas startups, and government platforms, the center of AI competition shifts. It moves from who can build the strongest model to who can make model capability cheaper, easier, and more widely usable.
This is also why the essay’s critique of U.S. export controls matters. Washington originally hoped to slow China’s AI catch-up by restricting chips, EDA tools, semiconductor equipment, HBM, advanced packaging, and high-end models. But the problem is that the more the scope of restrictions expands, the more it pushes China toward a clearer independent technology-stack path. That is the central irony of America’s AI strategy.
This is also the core argument of my earlier essay (https://leonliao.substack.com/p/ai-chip-controls-may-just-build-chinas?r=731anr&utm_medium=ios)on U.S. technology blockades against China: America’s restrictions will not simply stop China’s rise. They will accelerate China’s effort to build its own independent technology stack.
And this is exactly what I have been trying to capture in my series, The Great Partition of Global AI. (https://leonliao.substack.com/p/the-great-partition-of-global-aifrom?r=731anr&utm_medium=ios)The global AI ecosystem is being partitioned. This does not necessarily mean a complete rupture, but it does mean a gradual layering of models, chips, cloud infrastructure, developer ecosystems, data centers, power systems, application scenarios, and national-security rules.
The United States may still dominate the highest-end frontier AI narrative. China, meanwhile, may build another diffusion system around low-cost open-source models, industrial deployment, Global South markets, and a domestic technology stack.
If Washington continues to misread the AI race as a sprint toward AGI, it may underestimate the long-term power of this alternative path.
Yes, the over indexing on AGI is pushing the U.S. gov and its labs to spend recklessly and sprint to a finish line that doesn’t exist. This makes the economy more fragile and the world less safe. The one most likely to get hurt is America itself. 🚨
Nice! i’m writing a book on related topics. Some other macro frames i’m working on: How important is a deflationary economy fueled by fierce competition amongst the capitalist class to Chinese manufacturing power? Much of U.S. financial thinking views deflation as a core problem—with roots in the 1929 period sharply influencing that logic. But the 1870s-1890s saw both deflation and the rise of the U.S. into the manufacturing superpower of the world.
Glad you liked the piece. Please share your book when it’s ready. Sounds interesting.
Here's a baby substack post on the theme: https://evanwrowe.substack.com/p/crush-margins-or-die-trying-american
Thanks for the write up.