Why Are We Sleeping on China’s AI Invasion?
When Airbnb CEO Brian Chesky recently admitted his company is “relying a lot” on Chinese AI models because they’re “very good…fast and cheap,” it should have set off alarm bells in Washington. Here’s a CEO choosing Alibaba’s Qwen over his friend Sam Altman’s OpenAI—not because of superior technology, but because Chinese AI is dramatically cheaper.
According to Bloomberg columnist Catherine Thorbecke, Alibaba’s Qwen has racked up roughly 385 million cumulative downloads on Hugging Face, surpassing Meta’s Llama at 346 million. Chinese AI models now account for more than 40 percent of new language models being built, while Meta’s share has collapsed to around 15 percent. Thorbecke also reports that U.S. coding startups valued at $10 billion, such as Cursor and Cognition AI, appear to be running on Chinese AI foundations—though those companies haven’t confirmed it.
Meanwhile, in China, try to access ChatGPT or Claude. You can’t. The Cyberspace Administration of China has approved over a hundred AI models for public use—every one domestically developed and none from American firms, according to Chinese state media and regulatory filings. U.S. AI products are effectively barred from the Chinese market through a combination of the Great Firewall and regulatory requirements.
That’s textbook trade asymmetry: China has erected an impenetrable wall against American AI companies while Chinese AI services freely access American customers and developers. And we’re letting it happen.
Washington has responded with a flurry of proposals—all of them missing the point. The “No Adversarial AI Act” would ban U.S. government agencies from using Chinese AI. Senator Josh Hawley’s bill would criminalize Americans working on Chinese AI research. While many of these measures deserve support, they are either narrowly targeted at government procurement or focused on preventing collaboration and investment in the Chinese AI companies.
But none of them address the core problem: American companies are freely purchasing Chinese AI services while Chinese companies have zero ability to purchase American AI services, even if they wanted to. That’s exactly the sort of one-way trade barrier that President Donald Trump has promised to end.
Moderate Reciprocity in AI Now
The case for action is straightforward.
First, there’s the reciprocity principle. Whatever China imposes on us, we should impose on it. If China blocks American AI from its market, we should block—or at least tariff—Chinese AI from ours.
Second, Chinese AI companies benefit from heavy state subsidies that allow them to undercut American competitors on price. As venture capitalist Chamath Palihapitiya recently said on the All-In podcast, Chinese AI is “frankly just a ton cheaper than OpenAI and Anthropic.” That price gap reflects Beijing’s industrial policy, not market efficiency. A tariff would level the playing field by forcing American purchasers to internalize those hidden subsidies.
Third, there are legitimate national security concerns about what American companies feed into Chinese AI systems—systems that Beijing can access under Chinese law. A tariff makes companies pay for that security risk rather than socializing it.
Now, let’s be clear about what true reciprocity would require: a complete ban. China doesn’t just restrict American AI—it prohibits it entirely. The reciprocal response would be to prohibit American citizens and businesses from using any Chinese AI products, period.
But we can start with something more moderate: a substantial tariff on Chinese AI services. This approach has several advantages. Unlike a ban, it generates revenue rather than simply cutting off commerce. It lets market forces work—companies can still choose Chinese AI if the price advantage remains compelling even after the tariff. And it’s consistent with the broader Trump administration approach to trade policy.
The mechanism is straightforward. When American companies pay for Chinese AI services—whether through API calls, subscription fees, or commercial licenses—that’s a taxable transaction. Enforcement would follow the same framework as any other imported digital service.
Critics will argue this makes American companies less competitive globally. This is the line they used for a generation as we gave away our industrial base to China. Doing it once was foolishness. Allowing it to happen twice, this time with AI, would be treason.
The real question is whether we’re serious about defending American AI leadership. We can’t simultaneously complain about China’s industrial policy while giving Chinese AI companies unrestricted access to American buyers.
It’s time to practice what we preach about reciprocity. Tariff Chinese AI.
