DeepSeek-V4-Flash has claimed the top position on OpenRouter’s global AI model usage rankings, as worldwide token consumption continues to rise amid growing demand from AI Agents and other generative AI applications.

According to data from OpenRouter, global AI model usage reached 28.9 trillion tokens during the week of May 18–24, marking a 7.4% increase from the previous week. Token consumption has now risen for five consecutive weeks, signaling sustained demand for large language model services.

Chinese AI models consumed 9.22 trillion tokens during the period, up 19.89% week-over-week, while U.S. models processed 4.93 trillion tokens, an increase of 16.27%. Chinese models have maintained the global lead for four consecutive weeks.

OpenRouter is an AI model aggregation and API platform that provides transparent token-level monitoring and billing systems. The platform primarily serves overseas developers, with Chinese developers accounting for approximately 6% of its user base.

The surge in token consumption is even more pronounced in China’s domestic market.

According to China’s National Bureau of Statistics, daily token usage in China exceeded 140 trillion in March 2026. Doubao, ByteDance’s AI chatbot, saw its daily token usage double to 120 trillion within three months.

China International Capital Corporation (CICC) estimates that under moderate adoption scenarios, once AI Agent penetration reaches 8%, total token consumption from Agents will equal that of traditional chatbots.

The widespread adoption of AI Agents is expected to create a multiplier effect on token demand, with daily token usage potentially increasing fivefold as task complexity, usage duration, and user penetration rise simultaneously.

As token consumption accelerates, China’s telecommunications carriers have begun commercializing token-based services.

China Mobile launched token computing services for individual users on April 21, supporting mainstream models including DeepSeek and Qwen, with packages starting at RMB 5.99 (US$0.83).

China Telecom launched trial commercial token packages on May 17, pricing its entry-level SME package at RMB 39.9 (US$5.53) per month.

China Unicom’s Shanghai branch began offering token services to local enterprise customers on May 16.

China Telecom has also issued a tender for centralized procurement of “Token Factory” generation-capacity services, signaling the industry’s shift toward standardized token-production infrastructure.

Tianfeng Securities noted that AI data centers are increasingly evolving into “Token Factories,” emphasizing the importance of large-scale data processing and token-generation capabilities.

CITIC Securities stated that the emergence of Token Factories and token operators marks a transition of token generation from an auxiliary function into a standardized, billable service.

The brokerage expects this shift to transform the computing-power rental market from the current fixed monthly leasing model—based on bare-metal server rental duration—to a usage-based pricing system tied directly to actual token consumption.

CITIC Securities further noted that once tokens become the core pricing unit for computing power, infrastructure providers will be able to fully capture the upside from continuously expanding token demand across AI applications.

The firm argued that the current rapid growth of China’s computing-power rental industry is largely driven by persistent supply-demand imbalances in the domestic AI infrastructure market. This dynamic, combined with ongoing industry consolidation toward major players, strengthens the competitive positioning of leading providers with access to high-end AI chips and computing resources.

As a result, CITIC Securities remains bullish on the long-term growth prospects of China’s top computing-power rental companies amid the latest wave of token consumption expansion.

Related Coverage：

DeepSeek Weaponizes Compute Costs to Force Global AI Consolidation in 2026