Deepseek builds its own AI chip to cut Nvidia reliance

Chinese AI startup Deepseek is designing its own chip, according to Reuters. The chip is specifically for inference — the phase when a trained model generates answers for users — not for training new models. This could reduce Deepseek’s dependence on Nvidia and Huawei hardware. The project is early-stage: Deepseek is talking to chip design, manufacturing, and memory companies, and has quietly been hiring chip engineers for months without posting public listings. The challenges are significant. US export controls block Chinese companies from accessing cutting-edge chips and memory. OpenAI and Anthropic are also working on custom chips. Meanwhile, Deepseek is raising outside capital for the first time — reportedly seeking $7 billion at a $52–59 billion valuation. Why it matters: This move signals that even fast-moving AI startups see custom silicon as a strategic necessity, not just a cost play — and that US chip restrictions may accelerate, not stop, China’s semiconductor ambitions.