DeepSeek is reportedly building its own AI inference chip to cut reliance on Nvidia and Huawei

Chinese AI lab DeepSeek is developing its own AI inference chip to reduce its dependence on Nvidia and Huawei processors, according to three people familiar with the effort. The chip targets inference, the stage where trained models answer queries, rather than model training.

The move is another sign that China’s leading AI developers want to control their own silicon as US export limits tighten supply. DeepSeek began the effort about a year ago and it remains at an early stage, with the company reportedly in discussions with external chip-design, foundry and memory partners. A design finishing and reaching volume production is far from assured.

DeepSeek’s hardware history explains the push. It previously trained its R1 model on Nvidia’s China-market H800 chip, which Washington banned in late 2023, and later adapted its V4 model for Huawei’s Ascend line, with Huawei processors used in part of V4-Flash training. An in-house inference chip would loosen its reliance on both suppliers.

The report moved markets. Nvidia shares fell about 2 percent in premarket trading, while memory and storage names dropped harder: Micron fell about 4.7 percent, Western Digital about 6.3 percent and SanDisk about 4.6 percent.

The plan rests on unnamed sources rather than any DeepSeek announcement; the company has not commented, and details on the chip’s design, foundry or timeline remain unconfirmed. An early-stage program is also years from displacing established suppliers even if it succeeds.

Whether DeepSeek tapes out a working inference chip, and which foundry builds it, will be the clearest test of how far Chinese labs can go toward hardware independence.