AI Bot Traffic Surpasses Humans by Five Times; Low-Cost Chinese Models Gain Competitive Edge
nashnova research
AI agents now consume more than five times the compute of human users — the resulting surge in token costs is pushing model pricing to the center of enterprise decisions, opening a competitive window for low-cost Chinese models.
How fast is agent traffic growing?
Data from model aggregation platform OpenRouter shows agent requests consume roughly 15 times the tokens of a typical human query.
By early August, daily agent token consumption on the platform hit 7.3 trillion — up roughly 14-fold in six months.
This means → the compute demand generated by AI systems "talking to themselves" has far outpaced the growth curve of human users.
Why do agents burn so many tokens?
A conventional chatbot handles one question and one answer. An agent — an AI program that executes multi-step tasks on its own — automatically triggers code writing, database lookups, tool calls, and self-review within a single task.
In plain terms = a chatbot is like sending a text message; an agent is like handing AI an entire workflow to run solo — every intermediate step burns tokens.
This "one task, many calls" pattern multiplies per-unit compute consumption.
When did machine traffic overtake human traffic?
Agent traffic first surpassed human users in February this year.
By August, agents were generating more than five times the 1.4 trillion tokens produced by human users.
This reflects a structural shift: AI's primary workload is moving from "human talks to machine" to "machine talks to machine" — the economics of compute have fundamentally changed.
What does this have to do with Chinese models?
When token consumption grows exponentially, model pricing becomes the most sensitive variable in enterprise procurement.
This means → Chinese models entering the market on a low-price strategy hold a natural advantage in this cost competition.
Analysts note, however, that whether this pricing edge can translate into sustained market-share gains remains the key question — attracting trials with low prices is easy; retaining high-value customers is another matter.
市场有风险，内容仅供研究参考，不构成投资建议。