The battle for AI dominance is intensifying, with China’s DeepSeek AI making headlines for its cost-efficient performance. DeepSeek’s open-source chatbot has surged in popularity, topping the App Store charts in 51 countries. Now, new revelations confirm that the AI model relies on Huawei’s Ascend 910C chip for inference—marking a significant shift in China’s push for AI self-sufficiency.
DeepSeek AI has positioned itself as a strong alternative to Western AI models, boasting superior cost efficiency. While its DeepSeek R1 LLM (Large Language Model) was originally trained on Nvidia H100 GPUs, it now runs inference—processing and generating responses—on Huawei’s Ascend 910C chip.
Tech analyst Alexander Doria (@Dorialexander) highlighted this breakthrough on social media, emphasizing that DeepSeek’s reliance on Huawei’s homegrown chips for inference could be a game-changer.
Why This Matters
- Reduced Dependence on Nvidia: By using the Ascend 910C for inference, DeepSeek lowers its reliance on Nvidia’s high-end GPUs, which are often in limited supply due to U.S. trade restrictions.
- Cost Savings: Running AI models on Huawei’s chips is reportedly more cost-effective, giving DeepSeek an edge over competitors in terms of efficiency.
- China’s AI Independence: This move aligns with China’s broader strategy of reducing reliance on Western technology for its AI ambitions.
Huawei’s Future Plans: The Ascend 920C
While the Ascend 910C is effective for inference, its performance is not ideal for training AI models. To address this, Huawei is developing the Ascend 920C, a next-generation AI chip aimed at competing with Nvidia’s Blackwell B200, one of the most advanced AI processors available.
The Road Ahead
DeepSeek’s reliance on Huawei chips signals a major shift in the AI landscape. As China continues to strengthen its domestic AI ecosystem, the competition with Western AI giants like OpenAI and Google is set to intensify. With Huawei gearing up to launch even more powerful AI chips, this could be just the beginning of a new era in artificial intelligence.
