Market Focus Shifts from AI Spending Scale to Investment Productivity

According to Hana Securities analyst Lee Young-joo on July 25, investor focus on hyperscaler AI investments is shifting from capital expenditure scale to productivity efficiency. The critical question has changed from 'how much will be spent' to 'is the money yielding returns.'

Two data points signal progress. The LLM token expense index, tracking the average cost to process one million inference tokens, has declined from its May-June peak to March-April levels, reflecting optimization across model selection, inference processes, and pricing. Additionally, AI-related revenue has surpassed depreciation costs of operating data centers and semiconductor infrastructure for two consecutive quarters, indicating that deployed infrastructure is beginning to generate sufficient returns to justify continued investment.