J.P. Morgan: China's AI Enters Token Consumption Expansion Phase, Low-Cost Models Reshaping Competitive Landscape

BlockBeats News, June 23rd - Jufui released an AI series report on June 22nd, stating that as code, agents, office, and content generation scenarios accelerate in popularity, AI usage is shifting from "chatting" to "workflow," driving rapid Token consumption growth. According to OpenRouter data, as of the week before June 22nd, platform Token consumption increased by 4.7% compared to the previous week, reaching 46.7 trillion.

In this trend, the share of Chinese models is expanding. The report indicates that under the OpenRouter criteria, Chinese models consumed 18.8 trillion Tokens during the week, surpassing the 5.8 trillion of American models. DeepSeek V4 Flash ranked first with 4.94 trillion Tokens, followed by Xiaomi MiMo-V2.5, MiniMax M3, and Tencent Hy3 preview.

Behind this change is the Chinese models achieving a more competitive balance between performance and cost. Jufui stated that the gap in intelligence between Chinese and American models is narrowing, with API call costs only a fraction of the American models. The report attributes this cost advantage to MoE, attention mechanism optimization, and higher model compute utilization.

However, the increase in Token usage does not necessarily mean a synchronous increase in expenditure. Jufui's Silicon Data LLM Token Expenditure Index remained between 1.64 and 1.68 from June 14th to 19th, lower than the 2.04 on May 31st, indicating that enterprises and developers are expanding their calls while shifting towards cheaper, more efficient models.

Jufui believes that this combination of "more calls, lower unit price" will benefit companies like Baidu, Alibaba, Tencent, Kingsoft Cloud, AI Labs, and Kuaishou Kelin. Cloud service providers will benefit from the growth in model calls and MaaS demand, while AI applications are expected to find a clearer path to commercialization in code, agents, office software, and mid-long content generation.

The report specifically mentioned that the Volcano Engine is about to hold the FORCE Summit, and the market will focus on Douyin's large-scale model daily Token consumption, subscription package pricing, Seedance 2.0, TRAE, Arkclaw, Coze, and other code and agent product developments, as well as MaaS revenue targets. On the Tencent side, WeChat AI "Xiaowei" is currently in small-scale testing, and Jufui expects it to be officially launched in Q4 2026.

The user base on the application side is also expanding. In the May Global AI Mobile App Monthly Active Users ranking, ChatGPT still leads with approximately 9.58 billion active users, but Chinese apps have taken multiple top positions, including Douyin, Qwen, Quark, DeepSeek, and Yuanbao. The report suggests that the next key stage for Chinese AI applications will transition from user growth to high-frequency scenario penetration and commercialization.

In the long term, JFrog cited data from F&S and CAICT, stating that the China Big Model Market is expected to have a compound annual growth rate of about 64% from 2024 to 2030, with a scale exceeding 100 billion RMB by 2030. Enterprise-side and on-premises deployment are expected to be the main revenue sources.