DeepSeek Plans Inference Chips as US Curbs Leave Huawei With 50% of China Market
Updated
Updated · Ars Technica · Jul 7
DeepSeek Plans Inference Chips as US Curbs Leave Huawei With 50% of China Market
3 articles · Updated · Ars Technica · Jul 7
Summary
DeepSeek has spent about a year preparing a move into silicon, meeting hardware partners and hiring engineers to build data-center chips for AI inference, Reuters reported.
The push is driven by US export controls that have limited Nvidia in China and by DeepSeek’s aim to cut dependence on both Nvidia and Huawei.
Huawei already controls about half of China’s data-center chip market, while Alibaba and Baidu are also pursuing in-house AI chip efforts.
Custom silicon is becoming a broader AI industry strategy: OpenAI and Broadcom recently unveiled an inference chip, and Anthropic has also explored chip design.
Owning more of the stack could help DeepSeek secure scarce compute capacity as competition for data centers intensifies with AI model and service expansion.