Chinese Tech Companies Train AI Models Overseas to Access Nvidia Chips

TLDR

• Chinese tech companies like Alibaba and ByteDance are training AI models in Southeast Asian data centers to access Nvidia chips
• The offshore training strategy helps Chinese firms avoid U.S. export restrictions on advanced chips like the H20
• Companies use lease agreements for data centers owned by non-Chinese entities to access the hardware
• DeepSeek is an exception, training models domestically using Nvidia chips acquired before U.S. export bans
• DeepSeek is working with Chinese chipmakers including Huawei to develop next-generation domestic AI chips

Chinese technology companies are moving their artificial intelligence model training operations to Southeast Asia. The shift allows them to access Nvidia chips that face restrictions in mainland China.

NVIDIA Corporation, NVDA

Alibaba and ByteDance are among the major firms training their latest large language models in data centers located in Southeast Asian countries. This information comes from two sources with direct knowledge of the arrangements.

The companies rely on lease agreements with data centers owned and operated by non-Chinese entities. This structure provides access to advanced computing hardware that would otherwise be unavailable in China.

The offshore training trend increased after the United States moved to restrict sales of Nvidia’s H20 chips in April 2025. U.S. export controls aim to limit Chinese progress in advanced technology development.

Alternative Training Approaches

DeepSeek represents an exception to the offshore training pattern. The AI startup continues to train its models domestically in China.

DeepSeek accumulated a large stock of Nvidia chips before U.S. export bans took effect. This inventory allows the company to maintain domestic operations while other firms move abroad.

The company is also pursuing a different long-term strategy. DeepSeek is collaborating with Chinese chip manufacturers to reduce dependence on foreign technology.

Huawei leads the group of domestic chipmakers working with DeepSeek. These partnerships focus on optimizing current chips and developing next-generation AI processors.

The collaboration aims to create viable alternatives to Nvidia’s products using Chinese manufacturing capabilities. This development effort runs parallel to DeepSeek’s current AI training operations.

China has taken steps to reduce reliance on foreign AI chips. The government recently banned the use of foreign AI chips in state-funded data centers.

Chinese authorities have also discouraged local technology companies from purchasing advanced Nvidia chips. These policies complement U.S. export restrictions by reducing demand within China.

Current Chip Access Questions

President Donald Trump is now considering whether to allow Nvidia to sell H200 chips to China. The H200 represents a more advanced product than the H20 model currently restricted.

The decision could affect how Chinese companies approach AI development. Access to newer chips might reduce the need for offshore training facilities.

Chinese firms continue to expand their Southeast Asian data center operations. The lease-based model provides flexibility while maintaining access to restricted hardware.

The post Chinese Tech Companies Train AI Models Overseas to Access Nvidia Chips appeared first on CoinCentral.