[Salon] China’s Z.ai and America’s Self-Defeating AI Strategy



China’s Z.ai and America’s Self-Defeating AI Strategy

The company’s success despite U.S. sanctions proves export restrictions are counterproductive.

imageAn exhibition at the World Artificial Intelligence Conference in Shanghai, July 26, 2025. Photo: -/Agence France-Presse/Getty Images

China’s DeepSeek shocked the global AI community in January by building a frontier model at a fraction of Western costs. Now it has been outdone by a Chinese company subject to U.S. sanctions. It has become painfully obvious that Washington’s strategy of restricting chip exports isn’t working.

Z.ai, formerly Zhipu AI, last week launched GLM-4.5, a production-level open-source model priced at 13% of DeepSeek’s cost. It matches or exceeds Western standards in coding, reasoning and tool use. Z.ai runs on only eight Nvidia H20 chips, which Nvidia recently gained reapproval to sell in China. That’s better performance than DeepSeek with about half the hardware.

Z.ai did it despite being under Washington’s most restrictive GPU sanctions. The Commerce Department placed Zhipu and its subsidiaries on the U.S. Entity List in January for allegedly aiding China’s military modernization. (Zhipu disputed the factual basis of Washington’s decision.) About six months later—backed by $1.5 billion from Alibaba, Tencent and Chinese state funds—Z.ai delivered one of the world’s most competitive models. The company projects it will have millions of downloads and millions of dollars of revenue in 2025.

The company isn’t an outlier. It’s a signal of how well China’s AI strategy is working and how poorly America’s attempts to halt Beijing have fared. Washington’s tack so far has been to try to limit Chinese entities’ access to advanced hardware. Critics warned that export controls wouldn’t stop China from innovating and would instead push Chinese companies to develop their own chips, with which they could then fill the supply void left by the overly strict U.S. export rules. Far from controlling global chip demand, America was surrendering control to Beijing.

That’s exactly what seems to have happened. At the World Artificial Intelligence Conference in Shanghai last month, Premier Li Qiang, the second most powerful official in China, revealed a comprehensive AI plan. The country is taking a top-down approach that combines—on an international scale—GPU infrastructure diplomacy, open-source development and low-cost offerings on everything AI, from models and hardware to engineers.

The new “AI Plus” initiative aims to integrate Chinese models into key industries and export Chinese AI and hardware to the Global South—no export license, no questions asked.

The results are already clear. China has racked up more than 1,500 models, many of which are open-source. Many outperform or match the math and coding benchmarks of Western models. Huawei’s GPUs are quickly filling the gap left by the Biden administration’s adoption of stricter export controls. The research firm Bernstein projects that Nvidia’s global AI market share will drop a whopping 12% this year alone, if restrictions largely remain in place. China’s foundry capacity has vastly surpassed Washington’s expectation, and China is shipping chips abroad several years ahead of schedule. While U.S. politicians compete to see who can be more hawkish on China, Beijing is increasing international dependency on its models and hardware.

What’s the American response to a clearly failing strategy? In many parts of Washington, it’s still restrictions. But happily that isn’t true in the White House.

The Trump administration’s recently announced AI Action Plan emphasizes that U.S. strength lies in scaling supply and adoption abroad, not retreating. The president proposes exporting American AI and hardware while cutting regulations that slow production at home. Our data centers now consume more power than small cities. While China expands its energy production through whatever source is expedient, we face permitting delays and political scaremongering. America needs to streamline approvals, speed up reindustrialization, and rebuild large-scale computing capabilities. The U.S. should also make a priority of developing a Western AI supply chain with Latin America to counter China’s AI Belt and Road Initiative. This would turn the strategic manufacturing diplomacy Beijing favors against China.

Beijing is right to see exporting AI hardware and models as leverage. Each Nvidia chip sent abroad is a new point on the board for American software and values. Every U.S.-branded LLM shapes AI norms globally. Success comes from ubiquity of platforms, not exclusion or restrictions.

Hesitation isn’t the same as safety. America needs to start shipping AI to the world before it’s too late, including to China. Z.ai’s success proves that sanctions won’t stop Beijing. The next great Chinese AI model will be faster, cheaper and maybe fully self-sufficient. For America to lead, it must boost exports, infrastructure and global influence.

The AI future goes to the innovators who can establish a global platform, not to the most cautious regulators.




This archive was generated by a fusion of Pipermail (Mailman edition) and MHonArc.