“The UE8M0 FP8 is designed for the upcoming next-generation domestic chip, which is expected to be widely applied.” In August 2025, this announcement made by DeepSeek during the release of DeepSeek-V3.1 dropped a bombshell in the Chinese AI chip sector. Almost simultaneously, NVIDIA announced the cessation of H20 chip production, partly due to Chinese companies pausing procurement until security risks are resolved, and local manufacturers are now able to provide alternatives with comparable or even superior performance. These two events mark the entry of the Chinese chip industry into a brand new development stage. In today’s booming AI and robotics industries, chip autonomy has become the core battlefield of national technology strategy.01 Surging Demand: The Golden Age of AI Chips By 2025, the demand for AI chips in China is expected to reach a staggering $39.5 billion, with ByteDance, Alibaba, Tencent, and Baidu accounting for over 80% of the total demand for H20 chips in 2024. A research report from Bernstein indicates that China’s capital expenditure related to AI is projected to reach 655 billion yuan (a year-on-year increase of 51%) by 2025, with the domestic AI accelerator market expected to reach 276 billion yuan (a year-on-year increase of 85%). The government is actively promoting AI infrastructure construction through venture capital funds and provincial computing vouchers. Internet giants are also increasing their investments: ByteDance is investing 15 billion yuan, and Alibaba plans to invest 38 billion yuan over three years. This surge in demand is not only coming from internet giants; AI companies, automotive manufacturers, third-party data centers, and MaaS providers are all contributing to a diverse demand force. In the first half of 2025, over 40 AI data center bidding projects have emerged domestically, with an average scale exceeding 10 million yuan.02 Insufficient Supply: The Rise and Bottlenecks of Domestic Chips In the face of enormous demand, China’s chip supply capacity is rapidly improving, but significant gaps still exist. By 2025, the total supply of AI chips in China is expected to be around $37 billion, leaving a shortfall of $2.5 billion. The U.S. export control policies have severely impacted China’s chip supply. NVIDIA’s H20 chip once faced a sales ban, with an estimated sales loss of about $16.8 billion in 2025. Although sales have since recovered, supply remains constrained. Domestic chip manufacturers are rapidly rising. It is expected that by 2027, the self-sufficiency rate of China’s AI chip market will increase from 17% in 2023 to 55%. Major manufacturers include Huawei Ascend, Cambricon, and Haiguang Information. The performance of Huawei Ascend 910C is approximately 65-76% of NVIDIA H100, with an expected shipment of 350,000 units in 2025. Cambricon Siyuan 590 focuses on the inference market, with an expected shipment of 150,000 units to ByteDance in 2025. Haiguang DCU3 has received orders from companies like Alibaba due to its compatibility with the CUDA ecosystem.03 Technological Breakthroughs: A Wave of Innovation through Multiple Paths The Chinese chip industry is breaking through process limitations through various technological paths. Advanced packaging technology, such as Chiplet, has become an important choice in the post-Moore era. By modularizing chip integration, domestic companies have achieved efficient interconnection between CPUs and GPUs, reducing the cost of large model inference by 35%. Domestic 5nm Chiplet packaged products have begun mass production, utilizing silicon interlayers to achieve interconnections of tens of billions of transistors, with performance comparable to 4nm single chips. Three-dimensional stacking technology has improved the performance-to-power ratio of 28nm chips to nearly that of 7nm, providing “sufficient and reliable” solutions for scenarios such as smart cockpits and edge computing. The RISC-V open-source architecture has also avoided the politicization of patents due to zero licensing fees, with shipments expected to exceed 10 billion units in 2024, contributing over 50% from China, successfully breaking the instruction set monopoly.04 Ecosystem Construction: From Single Point Breakthroughs to Systematic Collaboration The Chinese semiconductor industry is shifting from single-point breakthroughs to systematic ecosystem construction. In the IP field, the domestic NA900 has become the world’s first RISC-V CPU IP to pass ISO 26262 ASIL-B/D certification, with performance comparable to ARM Cortex-M7. In the equipment sector, domestic companies have begun mass production of nanoscale plasma etching machines, supporting the increase of domestic equipment market share from 15% in 2020 to 28% in 2024. In the materials sector, domestic photoresists have achieved mass production in the 248nm KrF segment, and wet electronic chemicals have entered the TSMC supply chain. In terms of software ecosystem, Haiguang DCU3 has received orders from companies like Alibaba due to its compatibility with the CUDA ecosystem. This compatibility strategy reduces user migration costs and accelerates the commercialization process of domestic chips.05 Robot Chips: Explosive Growth of Emerging Applications In August 2025, NVIDIA released the next-generation robot-specific chip Jetson Thor, with AI computing power 7.5 times that of the previous generation, a power consumption of 130W, and energy efficiency 3.5 times that of the previous generation. This chip is specifically designed for inference of generative AI models, supporting the next generation of “physical AI” agents, including large transformer models, visual language models, and visual language action models. The domestic market responded quickly, with companies like Yushu Technology and Galaxy General Robotics announcing they would be the first to equip NVIDIA’s latest Jetson Thor chip. Domestic companies such as United Imaging Healthcare, Wanji Technology, UBTECH, Zhongqing Robotics, and Zhiyuan Robotics have also announced they will be among the first to use the new generation of edge-side robotic computing power. In the first half of 2025, the robotics industry chain saw 88 financing events (a year-on-year increase of nearly 80%), with 36 out of 40 listed companies reporting profits; over 83 publicly announced bidding projects in the domestic market, amounting to nearly 330 million yuan, indicating rapid growth in market demand.06 Challenges Remain: The Fragility and Constraints of the Supply Chain The Chinese chip industry still faces severe challenges. Currently, China’s IC self-sufficiency rate is still less than 5%, with production capacity accounting for less than 15%, and a high degree of import dependence. There are many pain points in the transition from laboratory technology to large-scale stable mass production of semiconductor products: immature processes, low yield, difficulties in equipment selection and operation, long construction cycles, lack of talent teams, and low conversion rates of scientific and technological achievements. The supply bottleneck in 2025 mainly lies in wafer foundry capacity and CoWoS production capacity. Domestic 7nm capacity is expected to reach 30-40kwpm by the end of the year, which can only partially alleviate the supply pressure of AI chips. Domestic chips still lag behind global leading products in performance (such as FP16 computing power, memory bandwidth) and software ecosystem (CUDA compatibility). Although Huawei Ascend’s hardware performance is close, its software compatibility still needs improvement.07 Future Prospects: From Replacement to Leadership The Chinese chip industry is undergoing a transformation from replacement to autonomy, and then to leadership. Bernstein predicts that by 2027, the Chinese AI chip industry may shift from supply constraints to demand constraints, with downward pressure on average selling prices. This will prompt companies to shift from price competition to technological innovation competition. The Yangtze River Delta, with Shanghai as the hub, is gathering leading manufacturers like Semiconductor Manufacturing International Corporation (SMIC) and Hua Hong Semiconductor, forming a complete chain from EDA tool design to high-end packaging and testing, contributing 50% of the national chip production capacity. The Pearl River Delta relies on the Shenzhen design industry cluster to incubate the Huawei Ascend AI chip ecosystem; the Bohai Rim region is constructing a “research-production integration” network with a Beijing R&D center and Tianjin materials base. In the first three quarters of 2024, the national integrated circuit output reached 315.6 billion pieces, a year-on-year increase of 26%, with industry sales revenue exceeding 1.5 trillion yuan. This growth is backed by the regional collaborative effect offsetting supply chain fragmentation. Domestic 7nm Chiplet packaged chips have begun mass production, with performance comparable to 4nm single chips; domestic 5G RF chips can now handle high-speed millimeter-wave signals; automotive-grade MCU chips operate stably in extreme cold and high temperatures, with a yield rate of up to 98%. These technological breakthroughs are reshaping the global chip landscape. It is expected that by 2027, the self-sufficiency rate of China’s AI chip market will increase from 17% in 2023 to 55%. The rise of Chinese chips is no longer a hypothesis but a reality that is happening.