The true victors of the AI race are the component suppliers. On October 28, 2026, South Korean company SK Hynix reported record Q3 financials: operating profit soared to 83.3 trillion won. The manufacturer explicitly stated that the global shortage of High Bandwidth Memory (HBM) will persist until at least 2030.
This data forms the foundation for a macroeconomic understanding of the industry. The speed of LLM training and inference is bottlenecked not by compute cores, but by memory bandwidth. SK Hynix has consolidated market power, converting its technological edge in HBM into historic windfall profits. Declarations of a deficit lasting until 2030 indicate that the cost of building data centers will remain extremely high. Hyperscalers will not be able to drive down component prices, and consequently, the cost of B2B AI integration for end-user businesses will rise in tandem with hardware amortization.
Sources: Financial Times | Yonhap
Crystallized Gold: SK Hynix Smashes Profit Records, Forecasts HBM Shortage to 2030
