DeepSeek builds own inference chip to escape Nvidia, Huawei dependency

DeepSeek, the Chinese AI startup behind globally popular models, is reportedly developing its own AI inference chip. According to three sources familiar with the matter, the chip is designed specifically for inference — the stage where a trained model generates responses — not for training new models. The move aims to reduce DeepSeek‘s reliance on Nvidia and Huawei, whose chips currently power its operations. The project began about a year ago, and the company has been quietly hiring chip-design engineers without public job postings. DeepSeek has also held discussions with chip-design, foundry, and memory companies as it seeks external partnerships.

However, DeepSeek faces significant hurdles. Designing a competitive AI chip typically takes years and requires massive capital. More critically, US export controls block Chinese designers from accessing the most advanced overseas foundries and high-bandwidth memory, essential for inference chips. Analyst Richard Windsor noted that Nvidia’s China business is essentially zero, and DeepSeek has almost no chance of selling silicon outside China without access to leading-edge manufacturing. The chip push coincides with DeepSeek’s first outside funding round, raising $7 billion at a $52-59 billion valuation. If successful, DeepSeek would join OpenAI (which unveiled its Jalapeno inference chip) and Anthropic in pursuing custom silicon for AI inference, the fastest-growing segment of AI computing demand.