DeepSeek is developing its own inference chip to reduce its reliance on Nvidia and Huawei chips, according to three people familiar with the matter.

The effort started about a year ago and is still in its early stages. DeepSeek is discussing partnerships with chip design, foundry, and memory companies, while quietly hiring chip-design engineers, the people said.

The push follows tighter US export controls on advanced AI chips sold to China.

Nvidia’s H800 became one of the main advanced AI chips Chinese companies could legally buy after Washington restricted the H100 and A100 in 2022, before expanding the rules in late 2023 to ban the H800 and A800 as well.

DeepSeek has used both Nvidia and Huawei chips. It adapted its V4 model for Huawei’s Ascend chips, while Alibaba and Baidu also are developing their own AI chips.

In June, DeepSeek was reportedly set to raise US$7 billion in its first external funding round at a valuation of US$52 billion to US$59 billion.

🔗 Source: Reuters