DeepSeek V4 preview on Huawei chips narrows US-China AI gap
Updated
Updated · Fortune · May 6
DeepSeek V4 preview on Huawei chips narrows US-China AI gap
13 articles · Updated · Fortune · May 6
In China, more than 600 million people were using generative AI by December, up 142% year on year, while OpenRouter says Chinese models' weekly token share has recently overtaken US models.
The model's partial use of Huawei chips reduces reliance on Nvidia, as Tencent, Alibaba and others embed agentic AI into consumer apps and business workflows across sectors from healthcare to courts.
Stanford researchers said the top-model performance gap has effectively closed, though US export controls still constrain advanced chips even as they push China to strengthen its domestic AI supply chain.
The U.S. AI buildout depends on Chinese parts. Who holds the true advantage in the global tech race?
As the U.S. sells advanced AI chips to China, has its strategy to contain Beijing's tech ambitions failed?
DeepSeek V4’s 1-Million Token Context and Huawei Ascend 950 Power China’s AI Sovereignty Amid U.S. Export Controls
Overview
In response to strict U.S. export controls on advanced chips, DeepSeek launched its V4 models in April 2026, featuring a novel Hybrid Attention Architecture that drastically reduces computational demands. This breakthrough enabled disruptive pricing, making AI more affordable globally, especially in the Global South. Meanwhile, Huawei's Ascend 950 chip, built with innovative domestic manufacturing techniques, powers DeepSeek's models and supports China's push for AI hardware independence. These advances have fueled rapid adoption but also triggered U.S. accusations of intellectual property theft, escalating geopolitical tensions. As both nations invest heavily in AI, the race is shaping a divided global ecosystem focused on sovereignty, cost efficiency, and strategic influence.