South Korean Giants Implement AI Token Quotas to Control Rising Costs
Major South Korean conglomerates, including Samsung and SK Group, are implementing strict token quota systems for employees using foreign AI models like Claude, Gemini, and ChatGPT. This measure aims to curb the rapidly rising operational costs associated with high-volume enterprise AI usage. This shift highlights a critical challenge in enterprise AI adoption, where companies must balance the productivity gains of advanced LLMs against soaring API costs. It also underscores the ongoing tension between AI integration and data security, particularly in sensitive sectors like semiconductor manufacturing. Under Samsung's system, employees receive token limits based on their tier and must demonstrate tangible productivity gains to request quota increases. Meanwhile, semiconductor divisions at Samsung and SK Hynix remain highly cautious, largely restricting external LLMs in favor of in-house models due to data leak concerns.
## BACKGROUND
In large language models (LLMs), a "token" is the basic unit of text—such as a word, character, or subword—that the model processes and generates. Because commercial AI providers charge enterprises based on the volume of input and output tokens, heavy usage for complex tasks like software development can quickly lead to substantial operational expenses.