Exponential Growth in AI Compute Consumption as Domestic Models Scale Rapidly
A performance monitoring report for 2026, jointly released on October 25th by the China Academy of Information and Communications Technology among other institutions, reveals an unprecedented surge in foundational resource consumption within China's artificial intelligence sector. The core finding is striking: by June 2026, the daily volume of tokens processed by domestic large language models had soared to approximately 500 trillion. Compared to December 2025, this figure represents a near quadrupling, starkly illustrating the massive computational demand driven by the widespread deployment of AI applications.
Processing Speed Doubles, Yet International Gap Persists
Based on data monitored from January to August 2026, the report provides a comparative analysis of service performance across several mainstream models. A key efficiency metric—Tokens Per Second (TPS)—reached an average of 78.23 in August. This marks a significant doubling of performance from the 2025 annual average of 38.25 TPS, highlighting effective underlying optimizations and technological iteration in domestic models.
However, the same metric for typical international models during the period stood at 112.82 TPS. This indicates that a performance gap of roughly 44% still exists between domestic and top-tier global models in terms of pure processing efficiency, underscoring a continued path of追赶.
Structural Rise in Service Costs, But Value Advantage Remains
Accompanying the performance gains and demand explosion is a shift in market pricing for model services. The report notes that the normalized average price for domestic large model services has risen to 13.46 yuan since the start of 2026. Representing a near doubling from prior levels, this increase is driven by a combination of computing power costs, R&D investment, and market demand.
Despite this rise, the cost of domestic AI services remains substantially lower than the international average of 117.8 yuan, maintaining a strong competitive edge in terms of value. This combination of "higher performance at a relatively lower cost" provides favorable conditions for the broad adoption of AI applications and corporate efficiency gains within China.
Industry Outlook: Balancing Efficiency and Cost is Key
The milestone figure of 500 trillion tokens consumed daily signals that China's AI industry has entered a phase of large-scale, high-intensity practical application. Looking ahead, a core challenge for the industry will be to continue enhancing model processing performance (TPS) while effectively managing or even reducing service costs. For enterprise users, model selection will increasingly involve finding the optimal balance between performance, price, and suitability for specific use cases.