Source: Global Times

Article from the U.S. "Wall Street Journal" on August 5, original title: China's Zhipu Model and the U.S. Self-Destructive AI Strategy Subtitle: Despite U.S. sanctions, this company has achieved success, which proves that export restrictions are counterproductive. The Chinese DeepSeek Company has built cutting-edge models at a far lower cost than the West, shocking the global artificial intelligence (AI) community. Now, a Chinese company under U.S. sanctions has surpassed DeepSeek. Obviously, Washington's strategy of restricting chip exports (to suppress China) has not worked.

Z.ai (Zhipu) officially launched its new generation flagship large model GLM-4.5 last week. This production-level open-source model runs using only 8 NVIDIA H20 chips, achieving or exceeding Western standards in coding, reasoning, and tool usage. Zhipu achieved this achievement despite the harshest GPU sanctions from Washington. The U.S. Department of Commerce added Zhipu and its subsidiaries to the Entity List in January. About six months later, Zhipu delivered one of the most competitive models globally.

The company is not an exception. This shows that China's AI strategy has been highly effective, while Washington's attempts to prevent Beijing have had minimal impact. So far, the U.S. strategy has been to try to restrict Chinese entities from accessing advanced hardware. Critics warn that export controls cannot stop Chinese innovation, but instead force Chinese companies to develop their own chips, thus filling the supply gap left by overly strict U.S. export rules. Instead of controlling global chip demand, the U.S. has handed control over to Beijing.

This is apparently what is happening right now. China takes a top-down approach, combining GPU infrastructure diplomacy, open-source development, and low-cost solutions in the field of AI on the international stage, covering all aspects from models, hardware to engineers. The new "AI+" program aims to integrate Chinese models into key industries and export Chinese AI and hardware to countries in the Global South.

The results are already evident. China has developed more than 1,500 models, many of which are open-source. Many of these models have exceeded or matched Western models in mathematical and coding indicators. Huawei's GPUs are rapidly filling the gap left after the Biden administration implemented stricter export controls. Bernstein research institution predicts that if the restrictions remain largely unchanged, NVIDIA's global AI market share will decrease by as much as 12% this year. China's chip manufacturing capacity has far exceeded Washington's expectations, and it is exporting chips overseas several years ahead of schedule. While U.S. politicians compete to show a tough stance toward China, Beijing is enhancing dependence on its models and hardware internationally.

What is the response of the U.S. to this clearly failed strategy? In many places in Washington, the answer remains restrictions. However, the recently announced "Artificial Intelligence Action Plan" by the Trump administration emphasizes that the U.S. advantage lies in expanding overseas supply and applications, rather than retreating. He proposed exporting U.S. AI and hardware while reducing regulatory measures that hinder domestic production.

Beijing views the export of AI hardware and models as a leverage point, which is correct. Each exported chip adds new weight to the relevant software and values, and each brand's large model shapes global AI standards. Hesitation does not equal security. The U.S. needs to start exporting AI to the world, including to China, before it's too late. Zhipu's success proves that sanctions cannot stop Beijing. The next great Chinese AI model will be faster, cheaper, and perhaps completely independent. (Author Aaron Ginn, translated by Qiao Heng)

Original: https://www.toutiao.com/article/7535608122489061915/

Statement: The article represents the personal views of the author. Welcome to express your attitude below 【top/down】 buttons.