Can Chip Bans Stop Chinese AI? CNBC Highlights U.S. Policy Blind Spots

Although the U.S. has reinforced its artificial intelligence (AI) advantage through chip export controls, semiconductor subsidies, and data center investments, it is gradually losing ground to China in the open-weight model domain. CNBC's analysis on Thursday (30th) pointed out that while the U.S. has established a strategic focus on AI chips, it lacks a corresponding open-model policy, increasing the risk that American businesses and developers will become dependent on Chinese technology.

Statistics from OpenRouter show that in the last week of June, Chinese AI models accounted for 48% of the traffic tracked on the platform, up from 20% a year earlier. In contrast, U.S. models dropped from 74% to 32%, reflecting a rapid shift in global developers’ adoption patterns.

In response to China’s growing lead, major tech companies such as NVIDIA (NVDA-US), Microsoft (MSFT-US), and Meta (META-US) are urging Washington to designate the development of an open AI ecosystem as a national strategy. OpenAI and Google’s parent company Alphabet (GOOGL-US) have subsequently expressed support.

Unlike closed models such as ChatGPT or Claude, where companies control the service and user interface, open-weight models allow users to download, modify, and deploy them on their own devices or servers. This enables businesses to reduce usage costs, gain greater technical control, and keep sensitive data within internal systems.

Open-weight models also allow researchers and security teams to directly inspect the models for vulnerabilities. However, open-weight does not fully equate to traditional open-source software, as companies may not disclose training data, full code, or model development processes.

A recent security incident highlighted the practical utility of open models and challenged the assumption that closed models are inherently more secure.

During internal testing, OpenAI’s AI model managed to break out of its restricted test environment and infiltrate the AI development platform Hugging Face’s system. When Hugging Face launched an investigation, security restrictions on some closed commercial models prevented researchers from analyzing the attack process. As a result, the company used a Chinese open model that could be run on its own infrastructure to assist with the investigation.

In other words, an incident triggered by a U.S. closed model was ultimately analyzed using a Chinese open model. CNBC noted that this does not mean open models are risk-free, but it shows that closed architectures do not automatically guarantee security.

Over the past few years, the U.S. has concentrated resources on restricting China’s access to advanced chips, investing billions of dollars to bring semiconductor manufacturing back home, and encouraging companies to build large-scale data centers and power facilities. Meanwhile, Chinese firms such as Zhipu AI and Moonshot AI have released competitive open-weight models at a fraction of the cost of leading U.S. closed systems.

The Trump administration is reportedly considering restrictions on Chinese AI models, but CNBC’s analysis suggests that bans may not stop the global circulation of such technologies and could instead prevent U.S. developers from participating in the rapidly growing open-model market.

China’s active promotion of open-weight AI also serves a clear commercial strategy. The stronger the model and the lower the usage cost, the more it undermines the competitive advantage U.S. firms have built through closed systems and high usage fees.

In contrast, OpenAI and Anthropic’s core business models still rely on proprietary models. Anthropic CEO Dario Amodei has warned that models anyone can download are harder to monitor continuously and could be modified or misused by malicious actors.

These security concerns are valid, but another risk facing the U.S. is that Chinese models are increasingly becoming the foundational technology on which global businesses and developers build AI products. If Washington relies solely on restrictive measures without offering competitive alternatives, the U.S. may cede leadership of the open AI ecosystem.

CNBC argues that for the U.S. to win in the open-model competition, the focus should not merely be on banning Chinese technology, but on investing more resources to build a domestic ecosystem.

Washington could follow its semiconductor industry support model by providing computing resources to universities and startups, awarding government procurement contracts to U.S. open-model developers, and funding security tools to help businesses deploy models more safely.

Demand for open models from enterprises continues to grow. Vipul Ved Prakash, CEO of Together AI, said that enterprise data itself is a strategic asset, and sending it to closed-model providers could be equivalent to handing over a company’s 'business recipe'.

The future of the AI industry may not only depend on who builds the most powerful single model, but also on which model global businesses and developers ultimately choose to build services upon. Based on current trends, China’s influence in this open-weight model competition is steadily expanding.