The launch of DeepSeek V4 marks the first complete "NVIDIA-free" loop in China's AI industry at the trillion-parameter big model level. Its performance may not be the top tier globally (the official stance admits a 3-6 month lag behind closed-source flagship models), but it proves that domestic computing power can support cutting-edge model training and inference, and can deliver long-context capabilities of millions of words at a very low cost. $NVDA