DeepSeek to use Huawei chips over Nvidia for small AI models training

DeepSeek has confirmed that it will use Huawei AI chips over Nvidia to train small AI models. While the company was under consideration to use the Ascend processor for its next-gen AI reasoning model, it might skip that plan for the time being.

A new report from TheInformation says DeepSeek will adopt Huawei chips for its small AI models. The aim is to reduce dependency on Nvidia for its AI processors.

Details further mentioned that DeepSeek is testing new AI GPU accelerators from Huawei, Baidu, and Cambricon to train some smaller models than the AI R2 version.

However, that doesn’t mean DeepSeek will use Ascend chipsets for its R2 AI LLM (large language model). Inputs reveal that China’s AI startup will continue to seed Nvidia processors, as they are currently the reliable source for its products.

Early reports said that DeepSeek is facing challenges in its upcoming R2 AI model. Even though Huawei sent a super-efficient engineering team to help the company with R2 development, things didn’t really prove helpful.

Following these problems, DeepSeek has to postpone the launch plans for the R2 AI model. It delayed the R2 AI model debut and decided to release it later this year.

The latest report now reveals that DeepSeek is relying on Nvidia’s chipsets to build the more powerful R2 AI reasoning model. Meanwhile, it will use Huawei Ascend processors to train and refine smaller versions of the R2 model.

DeepSeek didn’t reveal when these Huawei AI chip-powered LLM tech will debut on the consumer platform. Though we may get the details regarding this matter in the time ahead.

Following the news, A Nvidia spokesperson said:

“The competition has undeniably arrived. The world will choose the best tech stack for running the most popular applications and open-source models. To win the AI race, U.S. industry must earn the support of developers everywhere, including China.”

[source]