The Sino-US AI Chip Competition Escalates: Industry Transformation Under the Intertwined Effects of Technological Breakthroughs and Export Controls

In October 2025, the Sino-US AI chip competition entered a white-hot phase, with the triple tensions of technological breakthroughs, export controls, and market competition forming. The global semiconductor industry, with a market size of $697 billion, is undergoing a new round of restructuring.

US Enhances Controls, Turning Chips into a Strategic Weapon

In May, the US Department of Commerce issued the "AI Chip Export Control Guidelines," adding Huawei's Ascend series and Cambricon to the technology export restriction list, requiring global companies to obtain US licenses for chips produced using US tools. In August, the Trump administration introduced a "percentage-for-license" model, forcing Nvidia and AMD to pay 15% of their chip revenue in China to the US government in exchange for export licenses for models like the H20 and MI308. Despite the controversy surrounding the "generalization of national security" over this move, Nvidia maintained its presence in the Chinese market through a special edition H20 chip. In 2025, H20 orders reached $18 billion, accounting for 78% of its sales in China.

China is accelerating its breakthrough, and domestic chips are on the rise.

Faced with the blockade, China's AI chip industry has developed a dual-engineered strategy of "technology iteration and ecosystem development." Huawei's Ascend 910C has entered mass production, and the Ascend 950PR, featuring its own HBM processor, will be launched in 2026, boasting a 300% increase in computing power compared to its predecessor. Cambricon's Siyuan series chip shipments will increase from 10,000 units in 2024 to 80,000 units in 2025, with ByteDance receiving a pre-order for 200,000 units. Alibaba Cloud's Tongyi Qianwen model has been locally deployed in markets such as Japan and the UAE, driving the performance of the Pingtou Ge PPU chip to surpass Nvidia's A800. TrendForce predicts that the share of Chinese domestic chips in the AI ​​server market will rise from 42% in 2024 to 58% in 2025.

The global industrial chain is being restructured, with the Middle East becoming a new battlefield.

The United States has shifted its strategic focus to the Middle East, signing a five-year agreement through the Saudi sovereign wealth fund Humain to purchase hundreds of thousands of GPUs and deploy the GB300 supercomputer. China has countered with an integrated "chip-cloud-device" ecosystem. Alibaba Cloud's Tongyi Qianwen has become a partner of Apple in China, and Huawei's Ascend chips are being exported to Southeast Asia and Africa. A Deloitte report shows that global investment in AI infrastructure will reach $3 trillion in 2025, with China accounting for 35% and the United States for 40%, creating a bipolar landscape.

Technology paths are diverging, while open source architectures are breaking through.

At the architectural level, the RISC-V instruction set has become a breakthrough for China. Advance Time and Space has launched high-performance AI SoCs covering smart IoT scenarios; Eswin Computing's RISC-V chips have been commercialized in the autonomous driving sector. Nvidia CEO Jensen Huang admitted that China is "only a few nanoseconds behind" in chip research and development. The DeepSeek-R1 model, with a training cost of $294,000, proves that non-Nvidia ecosystems are equally competitive.

This competition has transcended technological boundaries, becoming a microcosm of the reshaping of the global technological order. With the implementation of new Chinese rare earth export controls, the global chip supply chain faces the risk of raw material bottlenecks, while the US's "commission-based" model exposes its anxiety about industrial competitiveness. Over the next three years, the AI ​​chip competition will determine who will lead China and the US in the digital economy.