DeepSeek R1 Leveraging Huawei’s Ascend 910C AI Chips, Highlighting China’s Expanding AI Enhancements

DeepSeek’s AI model is making waves as it uses Huawei’s cutting-edge Ascend 910C chips to handle inference tasks, signaling significant progress in China’s AI scene recently.

DeepSeek Leverages Huawei’s Ascend AI Chips for Inferencing

Lately, the Chinese AI company DeepSeek has been a hot topic in global technology circles. Their new R1 LLM model has caused quite a stir, impacting the US stock market to the tune of $2 trillion due to investor concerns. While the specifics about the computational power behind DeepSeek’s R1 are still murky, Huawei’s role is becoming clearer. According to tech influencer @dorialexander on Twitter, DeepSeek’s R1 model relies on the Ascend 910C chips for running inference, adding a fresh twist to the ongoing narrative.

I feel this should be a much bigger story: DeepSeek has trained on Nvidia H800 but is running inference on the new home Chinese chips made by Huawei, the 910C. pic.twitter.com/6IAgQlQ3ou

— Alexander Doria (@Dorialexander) January 28, 2025

For those not in the loop, Huawei’s Ascend 910C chip is poised as a competitor to NVIDIA’s Hopper H100 AI accelerators. While details about Huawei’s latest chip remain under wraps for now, the company has announced plans for mass production by the first quarter of 2025. This move has piqued the interest of several major Chinese AI players, including giants like ByteDance and Tencent. In China, Huawei’s AI chips have increasingly become the go-to alternative to NVIDIA, grabbing a significant share of the market, and it’s likely we’ll see their popularity continue to grow.

There’s still a lot of confusion regarding the computational resources DeepSeek’s R1 model taps into. Notably, the reported $5 million isn’t the cost of training the entire model but rather the expenditure for operating the final version. Furthermore, it’s suggested that DeepSeek has access to over 50,000 of NVIDIA’s H100 accelerators, indicating their resource needs are in line with other major AI initiatives.

The decision to use Huawei’s chips for inferencing is intriguing because they are readily available to Chinese companies and offer competitive pricing. This makes them an attractive option compared to NVIDIA’s more expensive models or even those sourced through unofficial channels. Huawei is also reputedly developing the next generation of Ascend AI chips, which are expected to rival NVIDIA’s upcoming Blackwell AI products, potentially fueling even more intense global competition.