DeepSeek has begun developing an in-house AI inference chip, Reuters reported Tuesday, potentially reducing its reliance on Nvidia and Huawei as it expands beyond AI model development.

The Hangzhou-based startup launched the effort about a year ago and has been working with external semiconductor partners while privately recruiting chip-design talent.

The initiative comes as Chinese technology firms including Huawei, Alibaba and Baidu race to develop domestic AI chips, with Nvidia’s access to the Chinese market constrained by US export restrictions.

The Chinese AI startup, known for its open-weight large language models, has continued expanding its AI ambitions.

DeepSeek has developed foundation models focused on reasoning, coding and agent-based AI, positioning itself as an open-source alternative to leading US AI labs.

In April, it introduced DeepSeek-V4-Pro and V4-Flash, saying the new models could improve mathematical reasoning, coding performance and tool use while supporting larger agent-training environments.

DeepSeek has also strengthened ties with China’s domestic AI ecosystem by optimizing its latest models for Huawei’s Ascend processors.

Separately, reports indicate that the company is seeking to raise around 50 billion yuan, with investors including Tencent, CATL and other major Chinese technology firms expected to participate.