AI is transitioning from the "tool era" to the "partner era": Mainland research institutions review the top ten trends in AI

The Quantum Bit Think Tank of Chinese Research Institutions released the "Top 10 AI Trends Report for 2025," indicating that AI is transitioning from the "tool era" to the "partner era" by 2025, profoundly reshaping economic structures, social forms, and human lifestyles. The report lists ten key trends in AI development, including infrastructure for computing power, chip AI, AI reshaping traffic entry points, and the emergence of the Agentic Internet. Infrastructure for computing power. The demand for data centers is skyrocketing, and the computing power economy is the largest engine of the smart industry. In the future, more "computing factories" (data centers) will need to be built to meet the enormous demand for computing power from AI. Chip AI. The native demand for AI is reshaping chip innovation, with an increasing number of chips designed specifically for AI to meet the special needs of processing large amounts of data and performing complex calculations. Pre-training determines the hierarchy of large models, and architectural innovation determines the level of pre-training. This means that the quality of large models largely depends on the quality of early training, and training methods and technologies will be innovated to influence the training effectiveness of models. Large models are entering the inference phase, with inference demands driving model innovation. AI models are beginning to transition from the "learning phase" to the "application phase." AI models will improve through processing specific tasks. The application period of information AI, the research and development period of physical AI, and embodied intelligence becoming a converging trend. Currently, AI mainly processes digital information such as text and images, but in the future, it will develop the ability to understand and manipulate the physical world, such as robots perceiving and acting like humans. AI reshaping traffic entry points. Innovating the traditional PC Internet and mobile Internet, opening up the Agentic Internet, where future AI Agents will directly help users find the information and services they need. Multimodal becomes the key to AI application implementation, with videos, 3D, and code sequentially demonstrating productivity. In the future, AI will not only understand text and images but also process videos, 3D images, and various types of information. AI hardware flourishing across various platforms. In the future, artificial intelligence will be equipped in laptops, mobile phones, cars, glasses, toys, and other devices. AI4S breakthroughs accelerate the realization of AGI (Artificial General Intelligence), with AI mathematics reaching doctoral levels; enabling AI to possess comprehensive abilities similar to humans. Open-source AI enters China, with AGI (Artificial General Intelligence) following a Chinese route. It is expected that China will see greater development in the field of AI open source