How China bypasses U.S. chip bans to power its AI development

The demand for AI hardware

Nvidia has become a cornerstone of the modern economy by designing Graphics Processing Units (GPUs). While these components were originally built to handle 3D visuals in video games, they possess the unique ability to process massive amounts of digital data. This capability makes them the primary tool for training and running large-scale artificial intelligence models.

The launch of ChatGPT in 2022 served as a catalyst for an investment gold rush in the AI sector. As governments and tech giants raced to build their own systems, the demand for Nvidia's hardware skyrocketed. This surge in interest has driven Nvidia's market valuation to $4.5 trillion, placing it on par with Apple as one of the world's most valuable companies.

Because advanced AI has significant military applications, the United States has implemented strict export restrictions on Nvidia's most powerful hardware. Washington specifically targets the high-end "Blackwell" and "Rubin" GPUs to maintain a global lead in technology. These policies are intended to prevent China from accessing the most sophisticated tools available for AI development.

Bypassing the export barriers

Despite these restrictions, Chinese startups like Moonshot AI, Z.ai, and DeepSeek continue to release high-performing models. Analysts note that these vendors are still utilizing cutting-edge Nvidia chips to train their systems. This suggests that the U.S. trade barriers are not completely stopping the flow of technology into the Chinese market.

One primary method for bypassing these rules involves the use of third-party data centers. Because U.S. export controls typically restrict where a chip is shipped rather than who ultimately uses it, Chinese firms can rent computing power in other countries. For example, a customer in Shenzhen could log into a data center in Malaysia's Johor region to run their workloads in full compliance with current rules.

In some instances, Chinese AI labs have been reported accessing high-end Nvidia chips in Thailand. Other reports suggest more clandestine methods, such as smuggling. One tech outlet reported that DeepSeek developed a new model using thousands of chips that were dismantled overseas and then brought into China. Nvidia has stated they have seen no evidence of such smuggling, calling the idea far-fetched.

The rise of domestic chips

While Nvidia remains the preferred choice for speed and ease of use, China is aggressively developing its own hardware alternatives. Huawei is currently the top domestic competitor, though it still lags behind the American company in terms of raw performance. However, the gap is narrowing as the Chinese government provides state subsidies to boost local production.

Training on Chinese-made chips is often more difficult and expensive than using Nvidia hardware. It typically requires more engineering effort and takes more time to achieve the same results. To compensate for these inefficiencies, Chinese companies can rely on the country's cheap electricity and government funding to scale up their operations.

Huawei recently announced that its next-generation AI chip will be ready in the first quarter of 2027. This timeline is three quarters earlier than originally planned. As these domestic options become more viable, more Chinese companies are expected to shift their reliance away from restricted Western hardware.

The geopolitical stalemate

Nvidia CEO Jensen Huang has lobbied the U.S. government to allow sales to China, arguing that restrictions are encouraging the growth of homegrown rivals. The U.S. has attempted to find a middle ground by easing controls on the H200 chip, which is less powerful than the banned models. However, the U.S. takes a 25 percent cut of these sales, and Beijing has shown little interest in buying them.

The current situation creates a complex environment for global trade and technology development. China is finding creative ways to maintain its AI progress while the U.S. seeks to protect its technological edge. This tension highlights the difficulty of enforcing hardware bans in a world where data and computing power can be accessed remotely.

As the competition intensifies, the focus is shifting toward who can build the most efficient systems at the lowest cost. Whether China succeeds in replacing Nvidia with Huawei or continues to find ways to acquire American hardware remains a central question for the future of AI. The race for dominance continues as both sides navigate the limits of trade policy and engineering capabilities.