DeepSeek develops its own AI chip to reduce reliance on Nvidia and Huawei

Chinese AI startup DeepSeek is developing its own chip for inference — the phase where a trained model generates responses — according to three people familiar with the matter Source: Milenio. The move aims to reduce dependence on Nvidia and Huawei chips, which DeepSeek currently relies on for training and running its globally popular models Source: La Republica.

Chip Details

The chip is designed specifically for inference, not training new models Source: Milenio. The initiative is still in an early stage, with DeepSeek contacting external partners including chip design firms, foundries, and memory manufacturers Source: La Republica. The project began about a year ago Source: Milenio.

Strategic Context

DeepSeek's expansion into semiconductors marks a significant strategic shift for a company hailed as China's "AI champion" and could challenge Huawei's dominance Source: La Republica. Huawei currently holds about half of China's $50 billion AI chip market, aided by US export restrictions on Nvidia's most advanced chips Source: Milenio. However, competitors Alibaba and Baidu are also developing their own AI chips, eroding Huawei's share Source: La Republica.

Hiring and Timeline

DeepSeek has increased recruitment of chip design engineers in recent months, but without posting public job listings Source: Milenio. The company maintains a low profile and did not respond to comment requests Source: Milenio.