Nvidia's CUDA Empire Crumbles as Open-Source AI Gains Ground
Nvidia's dominance in AI hardware and software is being challenged by an influx of open-source AI from the East. This shift has been driven by Western sanctions, which have forced mainland Chinese developers to adapt and align domestic hardware with open-source software.
Huawei's CANN (Compute Architecture for Neural Networks) has emerged as a key player in this space, offering a proprietary parallel computing platform and API that competes directly with Nvidia's CUDA. By open-sourcing CANN, Huawei has enabled local chipmakers to coalesce around open acceleration frameworks that are natively decoupled from American APIs.
The use of custom silicon and open-source AI is gaining traction globally, with Google's TPUs (Tensor Processing Units) and AWS Trainium and Inferentia leading the charge. Meta and Microsoft are also rolling out bespoke silicon at data center scale. Meanwhile, models from Chinese labs such as Alibaba's Qwen, DeepSeek, Kimi, and Zhipu GLM make up a massive share of global model downloads and fine-tuning derivatives.
The rise of open-source AI has significant implications for Nvidia's business model, which relies heavily on proprietary software. As the market shifts towards custom silicon and open-source frameworks, Nvidia's dominance in AI hardware and software is being eroded.