Chinese open-source models like DeepSeek, Kimi, and Qwen are primarily developed, optimized, and deployed on NVIDIA's CUDA ecosystem, reinforcing its platform effects for inference. This dynamic benefits NVIDIA and neocloud providers like CoreWeave, Lambda, and Together AI, expanding the addressable inference market beyond hyperscalers. The long-term risk to NVIDIA is China achieving a fully independent AI compute stack, shifting network effects away from CUDA.
NVIDIA's software moat is currently reinforced by Chinese open-source AI adoption, driving demand for CUDA-optimized inference compute across a diversified ecosystem.