Viewpoint: NVIDIA will benefit in the short term from China’s open-source AI boom, while its structural risk lies in long-term computing power independence.

Citrini analyst Jukan published an article refuting the view that "China's open-source models are irrelevant to NVIDIA", noting that the boom in China's open-source AI is actually strengthening NVIDIA's dominant position. China continues to use NVIDIA GPUs for model training and releases versions optimized for NVIDIA hardware. As a result, models like DeepSeek V4 and Kimi K3 still rely on NVIDIA GPUs as the most advantageous platform for inference—no one uses TPU or Trainium to run inference for these models. This protects NVIDIA's moat in two ways: first, it maintains its status as the optimal platform in the inference market; second, it indirectly curbs the motivation of hyperscale cloud providers to promote self-developed ASICs as alternatives to NVIDIA GPUs. Meanwhile, Neocloud providers that rely on NVIDIA GPUs for their inference operations also stand to benefit, as the boom in China's open-source models gives them a steady source of computing power demand. Jukan points out that the real structural risk is long-term: if China achieves true computing power independence and no longer relies on NVIDIA, the consequences will go far beyond losing the Chinese market alone. NVIDIA's dominant position will be significantly weakened, the relative competitiveness of alternatives like TPU and Trainium will rise, and it will also deal a blow to the Neocloud ecosystem, which NVIDIA regards as a key strategic asset.