IDC and Inspur Release 2026 AI Computing Report: 2.2 Billion Agents to Drive Token Consumption Surge

At the 2026 Artificial Intelligence Computing Conference (AICC2026), IDC and Inspur jointly released the '2026 China AI Computing Power Development Assessment Report.' IDC Vice President Zhou Zhengang said during the interpretation that the report took more than half a year to prepare, during which a large number of enterprises were interviewed and relevant data collected. Compared with previous years, this year's report places Agents in a more central position.

The core contradiction the report raises is: enterprise investment is increasing rapidly, yet compute is still insufficient. The report states that 97.3% of enterprises expect to continue expanding AI compute investment, of which 37.7% need to significantly increase inference compute, and 29.3% need sustained large-scale compute resources to support business operations. But the same report expects the global AI compute demand satisfaction rate to continue declining from 79% in 2024, reaching a low of 71% in 2027, and recovering only to about 77% by 2030; the absolute gap widens from $38.9 billion in 2024 to $380.9 billion in 2030.

This combination deserves serious attention because it reframes the question from 'whether to buy cards' to 'whether buying cards can be converted into callable Tokens.' The report's explanation is: demand in the digital world can grow exponentially, but the manufacturing of chips, HBM, and server CPUs is constrained by real production capacity; before data centers can be racked, issues such as machine rooms, power, cooling, and networking must be resolved, and the cycle for grid expansion and delivery of key equipment is typically far longer than the speed at which an Agent product goes from launch to user explosion. Therefore, increased capital expenditure does not immediately translate proportionally into callable Tokens.

The mechanism given in the report is not complicated, but its implications are significant: total Token consumption = number of task executions × Token consumption per task × multi-agent collaboration amplification factor, and it states that these three multipliers are growing simultaneously.

The first multiplier is the number of task executions. Traditional large models usually require users to actively initiate requests; after Agents enter business processes, a large number of calls begin to be automatically triggered by the tasks themselves. The report expects the global number of annual Agent task executions to grow from 280 billion in 2026 to 415.07 trillion in 2030; in China, from 17.8 billion to 39.41 trillion.

The second multiplier is Token consumption per task. The report describes that a simple Q&A may consume only tens to hundreds of Tokens, while complex Agent tasks require loading context, task planning, calling search, databases, APIs, or code execution environments, and then reading tool return results; as long as the task is not completed, it may continue into the next round of reasoning. As context windows expand and model reasoning capabilities strengthen, Token consumption for a single task grows from tens for simple Q&A to tens of thousands for complex document analysis, and then to hundreds of thousands for multi-round decision tasks.

The third multiplier comes from multi-agent collaboration. The report states that a complex task can be split into multiple Agent subtasks, some responsible for retrieval, some for analysis, some for writing code, some for verification and summarization; Agents pass context and exchange results, further lengthening the task chain. When the three multipliers stack, the most direct result is an increased slope in compute demand.

The scale of these numbers needs to be viewed together: IDC expects the number of active enterprise Agents globally to grow from 79.4 million in 2026 to 2.216 billion in 2030, a compound growth rate of 129.8%; over the same period, global Token consumption is expected to grow at a compound rate of 4822.6%, about 37 times the growth rate of active Agent numbers. For China, the report states that the compound growth rate of active intelligent agents from 2026 to 2030 is 151.2%, and the compound growth rate of Token consumption over the same period is 3499.3%. It should be noted that the above are all IDC forecast values, and the basis and calculation method for the 37-fold relationship between 4822.6% and 129.8% are not explained in the report.