SK Group Chairman Chey Tae-won recently made a significant statement at Nvidia's GTC conference, suggesting the global chip wafer shortage could last until 2030.
This isn't just another market forecast; it points to a deep, structural problem driven by the insatiable demand of the AI era. We can see clear evidence of this demand in Nvidia's recent earnings, where they reported record-breaking revenue. This shows that the main issue isn't a lack of customers, but a fundamental lack of supply. Chairman Chey’s comment reframes the conversation from a short-term cycle to a long-term structural challenge.
The problem can be understood as a chain of three critical bottlenecks. First is the supply of silicon wafers, the foundational material for all chips. The wafer market is an oligopoly, controlled by just a handful of companies. Building new factories, like GlobalWafers' new plant in Texas, is a slow process that takes years. This means we can't quickly increase the fundamental supply of raw materials.
Second is the shortage of High-Bandwidth Memory (HBM), the specialized memory essential for training and running large AI models. This is a direct consequence of the wafer shortage. Companies like SK hynix, a leader in HBM, have already sold out their entire advanced chip production for 2026. Despite accelerating plans for new factories, the industry simply can't keep up with the demand from AI giants.
Finally, there's a hidden but crucial bottleneck: energy. AI data centers consume vast amounts of electricity, with demand projected to more than double by 2026. This puts immense strain on existing power grids. This is why Chairman Chey also emphasized an “energy strategy.” SK Group's investments in new power facilities and next-generation energy sources like Small Modular Reactors (SMRs) are a direct response to this looming crisis. Without enough power, even having enough chips becomes irrelevant.
In conclusion, these interconnected challenges—wafer supply, HBM capacity, and energy availability—create a complex, long-term hurdle for the AI industry's growth. The market has already started to recognize this, with the stock prices of memory suppliers like SK hynix rising significantly. The race is now on, not just to make more chips, but to build the entire infrastructure needed to power the future of AI.
- Wafer: A thin slice of semiconductor material, such as silicon crystal, upon which microcircuits are built.
- HBM (High-Bandwidth Memory): A high-performance RAM interface for 3D-stacked memory, designed to provide high throughput for data-intensive applications like AI.
- SMR (Small Modular Reactor): A type of advanced nuclear fission reactor that is smaller than conventional reactors, designed to be built in a factory and shipped to a site for installation.