Chinese AI model DeepSeek is making waves with its cost-efficient, open-source chatbot, which has topped the App Store in 51 countries.
Now, it’s been revealed that Huawei’s Ascend 910C AI chip is powering the model’s inference, a crucial step in AI response generation.
Huawei’s AI Hardware Push
DeepSeek trained its LLM (large-language model) on Nvidia’s H100 GPUs but switched to Huawei’s Ascend 910C for inference.
While the Ascend 910C lacks the power for training, it is sufficient for running the model, making it a viable alternative amid US-China tech tensions.
Huawei’s Next Move: The 920C Chip
To compete with Nvidia’s cutting-edge Blackwell B200, Huawei is developing the Ascend 920C, which aims to enhance AI training capabilities and reduce reliance on US-made chips.
With DeepSeek’s rise and Huawei’s expanding AI hardware ecosystem, China is positioning itself as a formidable player in the AI arms race.
Leave a comment