NVIDIA Stock Drop: Can DeepSeek Bypass CUDA and Disrupt the GPU Ecosystem?

DeepSeek's $5.57M R1 model challenges NVIDIA's CUDA dominance. Can PTX language disrupt the GPU ecosystem and reshape the future of AI computing?

DeepSeek trained its R1 model with only $5.57 million, far less than similar models like OpenAI's and Claude's. Some believe DeepSeek bypassed CUDA and used the more fundamental PTX language for training, which is one of the reasons for its low cost.

Can DeepSeek bypassing CUDA and using PTX language disrupt the GPU ecosystem?

I think the current situation where PTX bypasses CUDA, and Nvidia's short-term stock price decline, does not indicate that Nvidia's moat has disappeared.

To answer this question, we need to understand what CUDA means for Nvidia.