沙利文：中国企业级大模型日均调用量提升至37.0万亿tokens 阿里千问领先优势扩大占比第一

Core Insights - The report by Frost & Sullivan and Toubao Research Institute indicates a significant divergence in strategies among global AI vendors by the second half of 2025, with Chinese companies leading in the open-source ecosystem while overseas firms focus on closed-source models [1][2][4] Group 1: Market Dynamics - By the second half of 2025, the daily usage of enterprise-level large models in China is projected to reach 37 trillion tokens, a 263% increase from 10.19 trillion tokens in the first half of 2025, marking a significant transition in AI's role within enterprises [4] - The growth in daily usage reflects a shift from sporadic assistance to deep integration into key processes, driven by increased frequency of AI calls within individual business processes rather than just an expansion of user base [4][9] Group 2: Model Deployment Trends - Open-source models are expected to surpass closed-source models in terms of usage, becoming the mainstream deployment mode for enterprise-level large models [7] - The increase in usage is attributed to two main demand types: expansion for core systems and external services, which prefer closed-source models for stability, and internal efficiency tools that are more cost-sensitive, favoring open-source models [7][9] Group 3: User Migration Patterns - There is a growing willingness among enterprises to migrate from closed-source to open-source models, with the intent to switch increasing from 22.6% to 48.5% for closed-source users, while the intent to switch from open-source to closed-source remains low at 7.5% [9] - As enterprises scale their AI usage, the cost pressures associated with closed-source models are prompting a shift towards open-source solutions for standardized and replaceable scenarios [9][14] Group 4: Application Areas - Text content creation remains the most significant application area, with multi-modal content creation showing the fastest growth at an 11.9% increase, outpacing AI search and intelligent customer service [11] - The report highlights that large models are now widely used across core business functions, including content production, knowledge acquisition, data analysis, and research support [11] Group 5: Market Concentration - The enterprise-level large model market is becoming increasingly concentrated among leading vendors, as companies prefer to streamline their supplier choices to reduce operational burdens [14] - As daily usage scales reach trillion-level tokens, new traffic is likely to be directed towards established vendors with proven stability and capability, such as Qianwen and Doubao, which offer advantages in computational management and cost efficiency [14]