The United States currently leads the artificial intelligence hardware stack race, with Nvidia dominating global AI chip sales and installed computing capacity, accounting for roughly half the world’s installed AI chip stock and two-thirds of computing capacity. China's Huawei, despite technological inferiority, holds a strong domestic market position due to US export controls and Beijing's demand-side support, including subsidies and procurement preferences.
Nvidia's enduring advantage extends beyond hardware to its proprietary CUDA software platform, which has become the de-facto standard for AI development, generating powerful network effects and high margins of 70-80 percent. Huawei is actively challenging this software moat with its CANN and MindSpore ecosystems, open-sourcing its toolkit, and developing torch_npu for PyTorch compatibility to lower switching costs. Collaborating with DeepSeek, Huawei aims to bridge the gap, leveraging open-weight models to accelerate its software ecosystem momentum. While obstacles remain, the strategic implication is that China's success in crossing the CUDA moat, even domestically, would narrow the US advantage. Europe, lacking a significant domestic AI chip designer or software layer, remains an input provider in this critical competition, facing questions about achieving genuine sovereignty through partial strategies.
No comments:
Post a Comment