DeepSeek Develops Custom AI Chips to Bypass US Export Controls

The Chinese AI startup DeepSeek is developing its own artificial intelligence chips specifically designed for inference to reduce its dependence on hardware from Nvidia and Huawei. The initiative began approximately one year ago and involves the private recruitment of chip-design engineers and consultations with foundries and memory companies. This strategic shift targets specialized inference chips, which offer higher power efficiency than general-purpose GPUs.

The move is driven by United States export controls that restrict Chinese access to advanced Nvidia chips and high-bandwidth memory, alongside pressure from the Chinese government to create indigenous semiconductor alternatives. This trend mirrors actions by U.S.-based firms; OpenAI recently partnered with Broadcom to announce the Jalapeño chip, while Anthropic is exploring similar custom designs to mitigate compute constraints.

DeepSeek's entry into semiconductors challenges Huawei's current hold on approximately 50% of China's $50 billion domestic AI chip market, where it already faces competition from Alibaba and Baidu. Simultaneously, the company is pursuing a maiden funding round estimated to raise $7 billion, which would value DeepSeek between $52 billion and $59 billion.