Spending on tokens, the measurement unit for artificial intelligence, fell about 20% over the course of a month. Analysts attribute this to individuals and companies reducing their use of top-tier AI models due to cost burdens, or turning to relatively cheaper Chinese AI models. With artificial intelligence transformation (AX) an irreversible trend, the need for cost optimization is expected to grow rather than a reduction in AI dependence.
According to overseas market research firm SiliconData on the 9th, the LLM Token Expenditure Index, which tracks the global average price of tokens, fell 22% in just over a month. After peaking this year at $2.10 per million tokens on May 26, it recorded $1.64 as of the 7th of this month.
Analysts attribute this to the excessive rise in companies' AX-related costs. Companies were initially eager to use frontier AI models from Anthropic and OpenAI, but starting in June, they turned to cost efficiency through lightweight AI models. For simple tasks, the proportion of companies using low-cost Chinese AI models such as DeepSeek, Zhipu AI, and MiniMax—which are about 10 times cheaper—is rising. According to Bloomberg, on OpenRouter, the world's largest AI model brokerage platform, Chinese models held a token share of 48% as of the last week of June, far ahead of U.S. models at 20%.
Furthermore, cases of companies suspending use of high-priced AI models are not uncommon. Nexon decided on the 3rd to temporarily suspend internal use of Anthropic's latest frontier-class AI model service, 'Claude Fable5.' The move came just one day after overseas access restrictions on the model were lifted. As Fable5's corporate billing policy shifted from a subscription model to usage-based billing, the token cost burden was judged to have risen excessively relative to work efficiency. Accordingly, Nexon is reviewing options such as suspending all use of Fable5 for now and then changing its operating method to allow limited use only for some important tasks. "We are currently establishing policies and guidelines for the company's efficient resource management, and we plan to resume the service as soon as the arrangements are complete," a Nexon official said.
For this reason, optimization technology that selects AI models to fit each company's AX project is drawing attention. The telecom industry, which has AI infrastructure requiring large-scale investment while also offering customers suitable pricing, has moved to seize the business of optimizing client companies' token spending. In China, China Mobile and China Telecom have already introduced token pricing plans. KT also recently announced a new business concept to build a "token factory" capable of supporting token generation, brokerage, and billing, and to grow it into a flagship AX business model.
"Among companies with weak information technology (IT) foundations, many still struggle over which AI model to choose for AX amid a rapidly changing AI industry environment," a telecom industry official said. "There will likely be considerable demand for enabling them to select and use AI technology at an appropriate level of billing."
