China's DeepSeek V4 Chips Away at Nvidia's Lead

This shift to domestic hardware is a direct response to escalating U.S. export controls that began restricting China's access to advanced semiconductors in October 2022. These restrictions were designed to slow China's progress in AI and advanced military systems by cutting off access to high-end chips from companies like Nvidia. Instead of halting progress, the policy has accelerated China's national strategy for technological self-reliance. Huawei's Ascend 910B, manufactured by SMIC on a 7nm process, is a key alternative to Nvidia's offerings. While there's debate on exact performance, Huawei claims the 910B can outperform Nvidia's A100 chip by 20% in some tests, and deliver 80% of its efficiency in others. This has made it a viable option for Chinese tech giants like Baidu and iFlytek, who have placed orders for the chips. Cambricon, another key domestic player, is also ramping up production to fill the void left by Nvidia. The company's Siyuan series of AI accelerators are gaining traction, with the Siyuan 590 reportedly achieving around 80% of the Nvidia A100's performance. Cambricon aims to deliver around 500,000 AI accelerators in 2026, a significant increase from its 2025 output, signaling strong domestic demand. The DeepSeek V4 model itself is a significant step forward, reportedly a trillion-parameter Mixture-of-Experts (MoE) model. It is designed with a native multimodal architecture, meaning it can process and generate text, images, and video from its core training. This is a departure from previous models that often had such capabilities added on later. This convergence of advanced domestic models and hardware is central to China's "AI Plus" initiative, which aims to integrate AI across key industries like manufacturing. By forcing a pivot to local hardware, U.S. policy has inadvertently spurred the creation of a more integrated and self-sufficient Chinese AI ecosystem, from chip design and manufacturing to model development and application. While challenges remain for Chinese chipmakers, including reliance on older manufacturing processes and software ecosystem maturity compared to Nvidia's CUDA platform, the pace of development is undeniable. The successful deployment of complex models like DeepSeek V4 on domestic silicon demonstrates a growing capability to optimize the entire AI stack locally, a crucial step in long-term technological competition.