DeepSeek and Huawei Develop AI Chip to Reduce Nvidia Dependence

DeepSeek and Huawei are working together on an AI chip designed to cut China's reliance on Nvidia processors, according to a report by TokenPost.

The partnership pairs China's most-watched AI model developer with the country's largest telecommunications equipment maker. Both companies have already built products that reduce their exposure to US-origin hardware.

Why target Nvidia?

Washington's export controls restrict the sale of Nvidia's most advanced AI accelerators to Chinese customers. Those restrictions have pushed Chinese AI developers toward domestic alternatives, and Huawei's Ascend chip line has become the most prominent candidate.

DeepSeek, which gained global attention for training competitive models at low cost, relies on compute. A domestic chip supply would insulate its training runs from future export restrictions.

What do we know so far?

The report confirms the collaboration and its goal — reducing dependence on Nvidia — but does not specify:

• the chip's technical specifications;
• a production or release date;
• financial terms of the partnership;
• which foundry or process node will manufacture the part.

Huawei designs its own processors and, through HiSilicon, has shipped Ascent-series AI accelerators to Chinese cloud providers since 2019.

Who benefits if it works?

A viable DeepSeek–Huawei chip would serve two audiences at once. DeepSeek would secure compute for model training without export-control risk. Huawei would gain a flagship AI customer for its silicon at a moment when Nvidia's absence from the Chinese market has left demand unmet.

For Nvidia, the stakes are measurable. China accounted for a significant share of the company's data-center revenue before Washington tightened export rules, and each domestic alternative that reaches production erodes the market Nvidia might eventually recover.

The effort also signals a broader shift: Chinese AI firms are no longer only building models on US hardware and looking for substitutes later. They are co-designing the silicon underneath.

No launch date has been announced. The project's success will show first in benchmarks, not press releases.