DeepSeek R1 Utilizes Huawei’s Ascend 910C AI Chips for Inference, Highlighting China’s Advancements in AI Technology

There’s a buzz in the AI industry with news that DeepSeek’s latest AI model, the R1, is leveraging Huawei’s cutting-edge Ascend 910C chips for inference tasks. This development showcases China’s rapidly advancing AI sector, highlighting its competitiveness on the global stage.

Let’s dive into the nitty-gritty of DeepSeek’s AI compute strategy. Recently, this Chinese AI firm sent shockwaves through global markets. Their R1 model notably triggered a moment of investor panic, slicing a whopping $2 trillion from the U.S. stock market. While the exact computational horsepower behind the R1 model remains a topic of hot debate, Huawei’s involvement seems significant. Reports from industry insiders, including a notable tweet from @dorialexander, reveal that DeepSeek is running its model’s inference tasks on these very chips, adding an intriguing layer to the unfolding narrative.

In the realm of AI chips, Huawei’s Ascend 910C has been touted as a formidable contender against NVIDIA’s renowned Hopper H100 accelerators. Although details about Huawei’s chip specifics are still under wraps, it’s known that the company plans to kick off mass production by early 2025. Notable Chinese tech giants like ByteDance and Tencent are already lining up to express interest, solidifying Huawei’s reputation as the preferred domestic alternative to NVIDIA in China, and potentially expanding its market share in the coming months.

Regarding the resources fueling DeepSeek’s R1 model, things aren’t entirely clear. There seems to be a lot of confusion in the media. For instance, when talking about expenses, the often quoted “$5 million” actually refers to the cost involved in running the final model rather than the training phase itself. Moreover, it’s said that DeepSeek has more than 50,000 of NVIDIA’s H100 chips at its disposal, suggesting that its resource allocation is on par with other leading AI initiatives.

What makes the use of Huawei’s chips so fascinating is not just their availability to domestic players, but the competitive pricing compared to some of NVIDIA’s limited or even gray market options. Looking ahead, Huawei is gearing up for the next wave of innovation, developing a new generation of Ascend AI chips. These are expected to give NVIDIA’s upcoming Blackwell series a run for their money, setting the stage for intensified global competition.

It’s evident the future of AI computing is shaping up to be a thrilling contest of innovation and strategy, with major players like DeepSeek and Huawei at the heart of it all.