TLDR
- Major Chinese tech firms including Alibaba and ByteDance are training AI models in Southeast Asian data centers
- The move allows access to Nvidia’s advanced chips that face U.S. export restrictions in China
- Chinese companies lease data center space from non-Chinese operators to access the hardware
- DeepSeek trains models in China using Nvidia chips purchased before export bans began
- DeepSeek partners with Huawei and other Chinese manufacturers to develop domestic AI chips
Chinese technology companies are relocating their artificial intelligence training operations outside mainland China. The strategy provides access to computing hardware restricted by U.S. export policies.
Major firms like Alibaba and ByteDance now train their newest large language models in Southeast Asian facilities. Two sources with direct knowledge confirmed these arrangements to Financial Times.
The companies secure access through lease agreements with data centers owned by non-Chinese operators. This approach bypasses restrictions on chip sales directly to Chinese entities.
Offshore AI training activity increased steadily after April 2025. That month marked when the United States expanded restrictions on Nvidia’s H20 chip sales to China.
NVIDIA Corporation, NVDA
The H20 chip represents advanced computing technology designed for AI workloads. U.S. export controls target these products to limit Chinese capabilities in artificial intelligence development.
How Chinese Firms Access Restricted Hardware
The lease-based model provides Chinese companies with legal access to prohibited hardware. Data centers in Southeast Asian countries stock Nvidia chips for their operations.
Chinese firms rent computing time and storage space at these facilities. The arrangement allows them to train large AI models without directly purchasing restricted chips.
This workaround addresses both technical and regulatory challenges. Companies maintain their AI development pace while staying within legal boundaries.
Multiple Chinese technology firms have adopted this offshore training approach. The trend reflects growing demand for advanced computing power in AI development.
DeepSeek’s Domestic Training Strategy
DeepSeek follows a different path than its competitors. The AI startup continues training models within China’s borders.
DeepSeek built a large inventory of Nvidia chips before U.S. export restrictions took effect. This stockpile supports the company’s current domestic operations.
The company also invests in long-term alternatives to foreign chips. DeepSeek works with Chinese semiconductor manufacturers on next-generation processors.
Huawei leads the consortium of chipmakers collaborating with DeepSeek. The partnership focuses on optimizing existing designs and creating new AI-specific chips.
These development efforts aim to establish Chinese alternatives to Nvidia products. Success would reduce dependence on foreign technology suppliers.
China’s government has implemented policies discouraging foreign chip purchases. State-funded data centers now face bans on using foreign AI processors.
Authorities also discourage private companies from buying advanced Nvidia chips. These measures complement U.S. restrictions by reducing Chinese demand.
President Trump is reviewing whether to permit Nvidia sales of H200 chips to China. The H200 offers more advanced capabilities than currently restricted models.
Chinese companies continue expanding their Southeast Asian data center footprint. The offshore training model provides reliable access to necessary computing resources.
The post Chinese Tech Companies Find Workaround for Nvidia Chip Restrictions appeared first on Blockonomi.
Source: https://blockonomi.com/chinese-tech-companies-find-workaround-for-nvidia-chip-restrictions/