A senior KRAFTON official has shared his perspective on the 'AI token' cost issue, which has emerged as a major topic in the industry.

On the 23rd, Lim Gyeong-young, VP and Head of AI Transformation at KRAFTON, assessed that the current token costs for Large Language Models (LLMs) are essentially 'discounted' prices resulting from the intense, capital-intensive infrastructure competition among AI platforms.

He explained, "Once this honeymoon period ends, cost realization and optimization will become the most painful homework for every company," drawing a parallel to the inevitable rise of FinOps—which combines financial management with cloud operations—following the initial adoption phase of cloud computing.

Lim noted that he has held in-depth discussions with business leaders through internal communication channels, identifying two primary causes for the waste of resources driven by indiscriminate AI adoption.

The first is the misuse of AI for deterministic problems. He warned against the mistake of burning expensive tokens on LLMs to solve problems that could be handled more clearly and cheaply with standard code or rule-based systems. He pointed out that companies must rigorously evaluate from the problem-definition stage whether they are overlooking appropriate tools simply for the satisfaction of using the latest technology.

The second is uncontrolled orchestration. Infinite loops—where meaningless prompts are repeated without proper checks or guardrails—were also cited as a major culprit behind uncontrollable infrastructure and API costs.

To minimize these cost-escalation risks, KRAFTON has established and is operating a proactive AI FinOps system.

Lim clarified that true AI Transformation (AX) is not about blindly increasing the number of flashy tools. His stance is that the core lies in accurately defining problems, selecting the optimal means to solve them, and maintaining perfect centralized governance over the AI systems that are implemented.

"In this era of company-wide AI expansion, every organization must move beyond unconditional adoption and prioritize internal checks and pipeline optimization," he emphasized. "Now more than ever, we need the insight to calmly cut through the noise and focus on the essence of the technology."