The artificial intelligence world is rapidly evolving, and DeepSeek, a Chinese AI model, has emerged as a prominent player as China fights back against the large US IT corporations.
Its chatbot has raced to the top of the App Store charts in 51 countries, but we had no idea it was powered by Huawei.
According to AI analyst Alexander Doria, DeepSeek’s R1 large-language model (LLM) was initially trained on Nvidia’s H100 but now generates responses using Huawei’s Ascend 910C CPU.
This demonstrates that AI computing is shifting away from Western-made hardware and toward Chinese-made alternatives, which might have important implications in the global AI race.
Doria mentioned in posts on X that DeepSeek has separated training and inference, which reduces the requirement for powerful GPUs. While Huawei’s 910C hardware is not as powerful as Nvidia’s H100 for training, it performs well for replies, allowing DeepSeek to reduce expenses while remaining competitive.
Doria does, however, reveal that DeepSeek plans to train its next AI model with 32,000 Huawei 910C CPUs. While Huawei may struggle to meet demand, it may prioritize DeepSeek to demonstrate that China can develop AI without relying on US technology.
If DeepSeek trains on Huawei chips, it has the potential to totally transform the global AI business. The US AI industry, which is already concerned about DeepSeek’s performance, may face even more competition from Chinese startups that can construct advanced AI models at a reduced cost.
Currently, Huawei’s 910C chip cannot entirely replace Nvidia’s H100, particularly for training AI models. However, with the AI struggle far from over, DeepSeek’s usage of Huawei chips demonstrates that China is bridging the gap.
Its effectiveness in executing DeepSeek replies demonstrates that China is making rapid advances in AI chip technology. If the 920C chip, the successor to the existing one, performs as expected, it could be a game changer in the AI fight.
