China's DeepSeek developing its own AI chip, sources say

• The chip targets inference computing (generating user responses) rather than training, the fastest-growing segment of AI demand as applications expand beyond model development
• DeepSeek's move could challenge Huawei, which currently holds roughly half of China's $50 billion AI chip market, though Huawei already faces competition from Alibaba and Baidu developing their own chips
• Success faces significant hurdles: designing competitive AI chips typically requires years and substantial capital, while U.S. export controls bar Chinese companies from advanced overseas foundries and high-bandwidth memory components

Summary: DeepSeek Developing Proprietary AI Chip

Key Development: Chinese AI startup DeepSeek is developing its own artificial intelligence chip, marking a significant strategic shift for China's AI champion. The chip is designed specifically for inference computing—generating user responses—rather than training new models.

Strategic Implications: This move would reduce DeepSeek's dependence on chips from Nvidia and Huawei, joining a global trend of AI developers seeking greater hardware control. The company has increased hiring of chip-design engineers in recent months and initiated discussions with chip-design, foundry, and memory companies. Development reportedly began approximately one year ago.

Market Context: The initiative comes as Huawei currently controls roughly half of China's $50 billion domestic AI chip market, though competitors like Alibaba and Baidu are also developing proprietary solutions. U.S. export controls ban Chinese companies from purchasing Nvidia's most advanced chips, pressuring Beijing to build domestic alternatives.

Company Background: DeepSeek gained global recognition over a year ago with highly efficient AI models, including its R1 reasoning model that triggered a U.S. tech stock rout in January 2025. The company is raising capital for the first time, with a funding round expected to value it between $52-59 billion.

Challenges Ahead: Success is uncertain, as competitive AI chip design typically requires years and substantial capital. U.S. restrictions prevent Chinese designers from accessing advanced overseas foundries and limit access to high-bandwidth memory, a critical component for AI inference chips.

Target Market: The inference chip would address the fastest-growing AI computing segment, as industry demand shifts from model training to deployment applications.