Shanghai Securities has released a research report stating that the structural decline in computing power costs and significant improvements in model efficiency are driving the global AI industry past a critical turning point from a "technology arms race" phase into a stage of "commercial realization." Against the backdrop of U.S. chip export controls, super-nodes have become a stage for domestic chips to prove their capabilities. Domestic manufacturers have achieved large-scale deployment, marking the transition of domestic computing power from "usable" to "effective." Policy-wise, there is a clear emphasis on cultivating emerging and future industries; quantum technology and brain-computer interfaces have been elevated to national strategic levels. The main viewpoints of Shanghai Securities are as follows:
The dual-pole evolution of Chinese and U.S. models sees China advancing from catching up to achieving partial leads. The competition between Chinese and U.S. large models has shifted from a parameter race to a comprehensive contest involving architectural efficiency, reasoning capabilities, and ecosystem penetration. 2026 is identified as a key watershed moment, with core breakthroughs in model capabilities expected in Agentic AI and Physical AI. U.S. models like GPT-5, Gemini 3, and Opus 4.1 remain predominantly closed-source, focusing on breakthroughs in general capabilities and maintaining leadership in long-context reasoning, real-time processing, and autonomous tool use. Concurrently, the performance and ecosystem influence of open-source models have significantly increased. Domestic large models like DeepSeek and Qwen are now comparable to top international models on multiple evaluation benchmarks and have even surpassed them in specific areas such as Chinese language understanding and cost-effectiveness.
The structural decline in computing power costs coupled with enhanced model efficiency is propelling the global AI industry into a crucial phase focused on commercial returns. The token economy is stimulating unprecedented demand for computing power, creating significant opportunities for the domestic computing power chain. Statistics from the National Data Administration indicate that China's average daily token usage has surged from 100 billion at the beginning of 2024 to 140 trillion by March 2026, representing a growth of over a thousand times in two years. Following an 83% price increase for its MaaS platform API in the first quarter, Zhipu AI still experienced a 400% growth in usage, indicating a supply shortage. Lobster-type intelligent agents have evolved from chat tools to autonomous executors, genuinely embedding into enterprise business process scenarios like financial risk control, intelligent customer service, code review, and data analysis, thereby becoming productivity tools. 2026 is anticipated to be the inaugural year for the commercialization of AI Agents, shifting computing power demand from a "training-dominated" phase to a new stage led by "inference and intelligent agents."
Under U.S. chip export restrictions, super-nodes serve as a proving ground for domestic chips. Huawei's Ascend 384 super-node and Sugon's scaleX640 have both achieved large-scale deployment, signifying that domestic computing power has moved from being "usable" to "effective." Policies on new quality productive forces are being intensified, with quantum technology and brain-computer interfaces elevated to national strategies. The 2026 Government Work Report listed "intensifying efforts to cultivate and strengthen new growth drivers" and "accelerating high-level self-reliance and strength in science and technology" among its top ten tasks, explicitly emphasizing the cultivation of emerging and future industries. The 15th Five-Year Plan clearly proposes forward-looking layout for future industries, explicitly listing quantum technology and brain-computer interfaces. Quantum technology is prioritized as the foremost future industry, receiving unprecedented policy support. The view is that quantum technology has transitioned from technical verification to the stage of commercial exploration, with sectors like finance, chemicals, biology, and transportation beginning to validate the value of quantum computing, entering a period of exploration for large-scale applications. The industrialization process of brain-computer interfaces is on the verge of a qualitative leap from "clinical verification" to "large-scale application," driven by policy support, breakthroughs in clinical trials, and capital investment. The next 2-3 years will be critical for regulatory framework refinement and the approval of medical-grade products.
Risk warnings include slower-than-expected policy implementation; delays in technological innovation and commercialization; and intensifying industry competition.
Comments