The Hidden Story Behind AI's Slowing Pace [Executive Al & Gakdogi Guest Interview] | Infomax Live 260915

The video identifies the core risks of the AI investment cycle not in rising interest rates themselves, but in the enormous cash consumption of data centers and model companies, opaque profitability, supply bottlenecks, and political variables. It notes that OpenAI and Anthropic can continue large-scale investment through funding raised before going public and circular financial support from big tech, but that adjusted earnings excluding stock-based compensation and depreciation alone make it difficult to assess sustainable profitability. It raises the possibility that recent calls to slow AI development reflect not only safety concerns but also interests in easing financing burdens and lowering data center costs. However, it sees limited potential for an immediate, sharp collapse, because power and semiconductor bottlenecks could cause prices to surge before an actual supply glut emerges, and semiconductor demand could remain robust through 2027~2028. If Chinese memory expansion and increases in global production capacity subsequently materialize, the risk of oversupply could grow around 2029. It presents GPU rental prices as the cycle's most direct leading indicator and assesses data center profitability as strong at present, with even older GPUs maintaining their rental rates. It emphasizes that although AI use will be difficult to reverse over the long term, how much the public will pay and which hyperscalers emerge as winners or losers are separate questions, and that a good industry does not necessarily mean good stock returns. For investment in Korea, it takes a relatively favorable view of diversified investment in undervalued Korean stock indexes rather than individual stocks, because even with solid semiconductor earnings, absorbing selling pressure takes time.