Frost & Sullivan: China AI Funding RMB150.4 Billion in 2025, Global AI Investment RMB4.2 Trillion

At the 20th Sullivan Global Growth, Innovation and Leadership Summit and the 5th New Investment Conference held on August 5, Zhu Yibo, co-founder and chief technology officer of StepFun, said the industry has undeniably entered the L3 era of AI. The L3 concept originates from OpenAI's AGI roadmap, which divides the path to artificial general intelligence into five capability levels: L1 chatbots, L2 reasoners, L3 agents—AI systems that not only think but also act—and L4 and L5 as innovators and organisers, respectively. This era began with the launch of Claude Code in the second half of 2025, Anthropic's terminal-based AI coding assistant. According to the newly released '2026 AI Advancement and Leap White Paper' by Frost & Sullivan and Touyan Research Institute, China's AI financing events and amounts rose to 1,579 and RMB150.4 billion respectively in 2025. Capital allocation has shifted to high-growth tracks including generative AI, embodied intelligence, AI chips and industry applications. Global corporate AI investment reached RMB4.2 trillion in 2025, up 129.9% year on year, with private investment of RMB2.4 trillion accounting for nearly 60% of the total. The number of financing deals exceeding RMB1 billion doubled from the previous year, with funds increasingly concentrating on leading enterprises and flowing mainly into foundation models, computing infrastructure and data management. Generative AI attracted RMB1.2 trillion in private investment, representing nearly 50% of global AI private funding, up more than 200% year on year. The white paper forecasts the global AI market will grow from US$1.1 trillion in 2025 to US$6.9 trillion by 2030. Large models remain the core driver of intelligent computing demand, with computing consumption expanding from pre-training to post-training, inference services and agent execution. Surging token usage makes the inference side the main incremental driver of intelligent computing demand in 2026. China's daily token usage stood at approximately 100 billion in early 2024, rose to 100 trillion by the end of 2025, and exceeded 140 trillion in March 2026, a more than 1,000-fold increase in two years. At the model level, from 2025 to the first half of 2026, the United States released 50 AI models and China released 30, with the top positions on mainstream model performance leaderboards occupied entirely by Chinese and American institutions. Since early 2025, models from the two countries have repeatedly swapped the top spot on major evaluation rankings, and the performance gap between Chinese and US frontier models has narrowed substantially. As of September 2025, the number of AI enterprises worldwide surpassed 37,000, with the US accounting for 36.4% and China 15.9%; together, China and the US accounted for 83.3% of global AI unicorns. Li Qing, Executive Director of Frost & Sullivan China, said that while the industry focused on model capabilities in 2024, by 2025 it had fully shifted to agent-driven development, with the core issue becoming how AI moves from question answering to task execution. Competition has now turned to practical applications: whoever can deploy AI stably into complex business processes at a reasonable cost is more likely to build a long-term advantage. The white paper shows that 57.3% of respondents said their companies are running agents in production, but in any single business function, the proportion of companies deploying agents at scale does not exceed 10%. In terms of industry application structure, China's large model deployment is concentrated in five scenarios: internet, finance, industrial manufacturing, consumer retail and education, with case shares of 31.8%, 14.7%, 12.2%, 8.1% and 6.5%, respectively, totalling 73.3%. According to the 'China's Future 50-Year Industry Development Trends White Paper', the core opportunities for the AI industry will revolve around large models, agents, physical AI and AI4S. Large models provide the cognitive foundation, agents handle digital task execution, physical AI pushes intelligent systems into real-world environments, and AI4S reshapes scientific discovery and industrial R&D processes. On August 5, Zhang Ju, chief financial officer and board secretary of UBTech, said humanoid robots are not intended to replace existing automation equipment but to take on work that even automation cannot handle. He cited car super unmanned factories that have achieved a high degree of automation, with painting and welding workshops already fully automated, yet assembly still contains many tasks that cannot be solved by advanced industrial automation equipment—exactly the space humanoid robots are meant to fill. The industry generally identifies three major application scenarios for humanoid robots: industrial manufacturing, where the market size is measured in hundreds of millions of RMB; commercial scenarios; and the home market, which holds the greatest long-term potential.