The spread of artificial intelligence (AI) agents is changing companies' AI cost structures. An analysis says AI spending should be viewed through 'tokenomics', weighing token use together with workflow.

Tokenomics is a term combining tokens, the unit of data processed by AI, and economics. The term emerged as the spread of agentic AI made token consumption a key variable for corporate management.

When companies use AI services, tokens are generated in the input and output process. For example, if an AI agent writes a report, input and output tokens keep accumulating as it searches for relevant materials and summarises content. The more an agent repeats verification and revision steps on its own, the larger total token consumption becomes.

Companies' AI costs are set based on this token usage. Industry experts advise companies to prepare for tokenomics as token costs also surge alongside the spread of AI agents. An official at an AI company said, "AI agents are not a structure where you ask once and close it like a simple chatbot," adding, "because they call AI models multiple times, managing token usage is bound to become the core of cost management."

In areas with heavy use such as customer service centres or in-house knowledge search, even small token savings can have a big impact on total costs. Differences may look small over one or two calls, but cost gaps can widen quickly in an environment used daily by more than thousands of people.

Major AI companies have recently released both high-performance models and lightweight models, aligning with this trend. OpenAI, Anthropic and Google are presenting a direction of using high-performance models for tasks requiring complex reasoning, and relatively more cost-efficient lightweight models for simple classification or summarisation and standardised responses.

Another industry official said, "In the early stages of AI adoption, high-performing models came first, but going forward the ability to control costs by task will become important," adding, "the more a company uses AI agents at scale, the more tokenomics will become a key topic."

This trend is also linked to AI infrastructure strategy. That is because reducing AI token costs requires looking together at where data is and which infrastructure processes it. Big Tech companies are also watching the trend closely and using it as a new business opportunity.

For example, computing infrastructure company Dell Technologies recently emphasised tokenomics through its annual event. Dell advised that as AI use increases, companies should go beyond simply adding graphics processing units (GPUs) or servers and design the entire process in which data is generated, stored and processed. AMD is also presenting a computing portfolio strategy that spans CPU-based processing, cost-efficient GPUs and high-performance accelerators to respond to the tokenomics era, showing tokenomics emerging as an industry topic.

Tokenomics, however, is not a problem that can be solved through infrastructure investment alone. It is intertwined with what tasks companies apply AI to, which models they choose, and how many times agents are designed to call models. Demand is also growing for internal governance models that track AI usage by department and token consumption by task.

An industry official said, "AI agents can be a tool that raises productivity and a game changer in corporate operating costs," adding, "a tokenomics perspective that applies the right AI model to the right task is likely to determine the success or failure of AI use."