Model supremacy in a token-efficient macro environment plagued by HBM, energy and datacenter compute bottlenecks. The 2026 story of AI is getting geopolitical.
The real “China problem” is that open-weight Chinese models are exposing how fragile the economics of closed-source American AI may be. Washington may call it national security, but Silicon Valley also has a strong commercial interest in keeping low-cost competitors out.
Mike, I strongly recommend you pay attention to China’s developments in heterogeneous computing (combining different types of chips into one computing system) and “token factories” (using multiple models and infrastructure layers to deliver AI solutions). I believe these will become the next major trends in AI.
One more thing: the current situation reminds me of how people in China describe Napoleon’s famous quote: “I am the wave.”
That’s fascinating CT Zhao. I do think China’s ecosystem is the most innovative right now even in a stark absence of foreign investment and a peculiar geopolitical backdrop. In a world where TSMC is raising prices, ASML is as well - China’s crusade to catch up in the semiconductor industry is critical. China’s concentration of AI and robotics talent is going to pay dividends in the years ahead.
It most definitely seems like a turning point in AI, the open source wave seems to be miles ahead of where it was even three months or so ago.
There's no question that today the open source Chinese models seem like the largest threat by far - and when you think how much data AI has, it gets really scary.
I'm curious what you think the government's role is in protecting the U.S - should they be banning Chinese models from U.S. usage? Is that even possible? Is it ethical?
I think there's a big chance Meta comes out of this with one of the frontier open source models, they are investing $200 billion in Capex that's gotta go somewhere.
The Trump Administration restricting Anthropic models, was a mistake. Restricting Chinese models in any way would be a mistake. I don't believe in American protectionism. The world needs to set up a global bipartisan regulatory AI body.
The market needs to work itself out and the United States needs to compete without trying to pick the winners.
Neither should governments own a stake in these companies.
The real question here is whether AI is treated like economic competition -don't regulate - or a national security threat - regulate maybe?
At the end of the day, while I do agree with you in an ideal world on a global regulatory AI body, the Chinese coming to the table at this point seems a little far fetched.
Would leave the U.S. in an awkward situation where they set up a regulatory body which is followed by everyone besides for... China, the biggest threat.
To the contrary, China is far more pro AI regulation than the U.S. Trump’s mistake was trying to compromise with big Banks and incumbents on the perceived cybersecurity risk of Mythos class models. You can’t have it both ways.
You can’t restrict your best company in your country and expect a good outcome for business - the literal AI Enterprise leader. Anthropic already do their due diligence on risk mitigation and are if anything, over cautious. It was a pivotal mistake by this Administration.
The real “China problem” is that open-weight Chinese models are exposing how fragile the economics of closed-source American AI may be. Washington may call it national security, but Silicon Valley also has a strong commercial interest in keeping low-cost competitors out.
You can find the public chat about this article here: https://substack.com/chat/396235/post/43e45fd9-6558-4ad2-b5e3-16ec91df9973
Mike, I strongly recommend you pay attention to China’s developments in heterogeneous computing (combining different types of chips into one computing system) and “token factories” (using multiple models and infrastructure layers to deliver AI solutions). I believe these will become the next major trends in AI.
One more thing: the current situation reminds me of how people in China describe Napoleon’s famous quote: “I am the wave.”
That’s fascinating CT Zhao. I do think China’s ecosystem is the most innovative right now even in a stark absence of foreign investment and a peculiar geopolitical backdrop. In a world where TSMC is raising prices, ASML is as well - China’s crusade to catch up in the semiconductor industry is critical. China’s concentration of AI and robotics talent is going to pay dividends in the years ahead.
Hey Michael,
A few of my thoughts:
It most definitely seems like a turning point in AI, the open source wave seems to be miles ahead of where it was even three months or so ago.
There's no question that today the open source Chinese models seem like the largest threat by far - and when you think how much data AI has, it gets really scary.
I'm curious what you think the government's role is in protecting the U.S - should they be banning Chinese models from U.S. usage? Is that even possible? Is it ethical?
I think there's a big chance Meta comes out of this with one of the frontier open source models, they are investing $200 billion in Capex that's gotta go somewhere.
Thanks for the quality post!
The Trump Administration restricting Anthropic models, was a mistake. Restricting Chinese models in any way would be a mistake. I don't believe in American protectionism. The world needs to set up a global bipartisan regulatory AI body.
The market needs to work itself out and the United States needs to compete without trying to pick the winners.
Neither should governments own a stake in these companies.
A true libertarian :)
The real question here is whether AI is treated like economic competition -don't regulate - or a national security threat - regulate maybe?
At the end of the day, while I do agree with you in an ideal world on a global regulatory AI body, the Chinese coming to the table at this point seems a little far fetched.
Would leave the U.S. in an awkward situation where they set up a regulatory body which is followed by everyone besides for... China, the biggest threat.
To the contrary, China is far more pro AI regulation than the U.S. Trump’s mistake was trying to compromise with big Banks and incumbents on the perceived cybersecurity risk of Mythos class models. You can’t have it both ways.
You can’t restrict your best company in your country and expect a good outcome for business - the literal AI Enterprise leader. Anthropic already do their due diligence on risk mitigation and are if anything, over cautious. It was a pivotal mistake by this Administration.
A comedy of errors.