The rapid development of intelligent agents is significantly boosting the demand for computing power services, according to a recent report from the International Data Corporation (IDC) and Inspur Information. The 2026 China Artificial Intelligence Computing Power Development Assessment Report highlights that intelligent agents are driving token consumption through a multiplier effect, causing a substantial upward revision in the growth trajectory of China's intelligent computing power scale. Computing power has now been elevated to the status of a national strategic infrastructure, on par with water conservancy, power grids, and railways.
From a scale perspective, IDC forecasts that the number of active intelligent agents globally will surge from 79.4 million in 2026 to 2.216 billion by 2030, representing a compound annual growth rate of 129.8%. In China, the compound annual growth rate for active intelligent agents is projected to reach 151.2% between 2026 and 2030, while token consumption during the same period is expected to grow at a staggering compound annual rate of 3499.3%. This rapid expansion in both agent numbers and token usage is driving a continuous rise in computing power demand.
"Intelligent agents are primarily inference-driven, involving multi-round chain-of-thought reasoning combined with tool calls. They generate fragmented requests with significant traffic fluctuations, require fine-grained resource partitioning, and demand comprehensive end-to-end auditing and traceability," said Zhan Yuanzhu, head of intelligent computing products at QingCloud Technologies Group, in an interview with Securities Daily.
The sustained increase in computing power demand is also reshaping the form of computing power services. Yu Fenghui, a senior researcher at Pangoal Institution, told Securities Daily that the most notable shift in the industry is the delivery model transitioning from "selling resources" to "selling tokens." The competitive focus has moved from scale to efficiency, with industry benchmarks shifting from "how many GPUs you have" to "token production efficiency per watt." Key capabilities now include integrated software-hardware optimization, inference engine tuning, and intelligent scheduling for peak shaving.
The continuous release of demand is directly reflected in order volumes. Recently, listed companies have been signing large-scale computing power service contracts. For instance, on the evening of September 18, Beijing Parallel Technology Co., Ltd. announced that its wholly-owned subsidiary, Wuhan Parallel Intelligent Computing Technology Co., Ltd., signed a Technical Service Framework Agreement with Client A. Under the agreement, Wuhan Parallel will provide GPU computing resources, supporting IT equipment, and related technical services, with a total contract value not exceeding RMB 1.913 billion (tax inclusive).
On the evening of September 16, Shaanxi Kanghui Pharmaceutical Co., Ltd. announced that its wholly-owned subsidiary, Beijing Kanghui Zhichuang Technology Co., Ltd., signed a Computing Power Service Contract with Client A Company, valued at approximately RMB 1.72 billion (tax inclusive) over a five-year term. To ensure the smooth execution of this contract, Beijing Kanghui Zhichuang also signed a Purchase and Sale Agreement with Supplier G for computing servers, amounting to approximately RMB 1.141 billion (tax inclusive).
Guo Tao, deputy director of the China E-Commerce Expert Service Center, offered advice for computing power service providers: they should avoid homogeneous competition dominated by graphics card resources, build a tiered product matrix, and steer clear of price wars in the general-purpose computing track. Instead, he recommends developing scenario-specific computing power solutions for vertical sectors such as intelligent agents and autonomous driving, thereby establishing delivery barriers in these industries.
This article is for reference only and does not constitute investment advice. Investors should operate at their own risk.