The AI Era Reshapes Chip Dynamics: Shortage as the New Normal?

In a significant long-term assessment of the global memory market, SK Hynix CEO Kwak Noh-Jung has projected that the current supply crunch, fueled by the artificial intelligence boom, is likely not a short-term blip but a "new normal" that could extend until the end of 2030. This outlook stems from a fundamental shift in the industry's dynamics driven by AI technology.

From Commodity to Strategic Resource

At the heart of Kwak's argument is the transformed role of memory chips. Historically viewed as standardized, cyclical commodities, their pricing and supply followed predictable boom-and-bust patterns. The rise of AI has changed that. High-performance computing, large language model training, and inference now place unprecedented demands on memory bandwidth, capacity, and efficiency.

Advanced memory technologies, particularly High Bandwidth Memory (HBM), have become critical bottlenecks and strategic components in AI server clusters. They are no longer easily replaceable generic parts but core resources that define the upper limits of AI processing power. This shift in essence reduces the likelihood of a return to the severe oversupply scenarios of the past.

Robust Demand Tempers Cycle Concerns

Despite occasional market chatter about a potential AI investment bubble, Kwak indicates that from direct engagement with customers, he sees no signals of weakening demand. Cloud giants and AI firms continue to demonstrate strong, sustained demand for memory chips, especially high-end products, tied directly to their AI deployment and product roadmaps.

He acknowledges the inevitability of industry cycles. "Challenges could emerge if we pass the peak of AI demand or enter another downturn in the future," he states. However, he quickly adds, "I believe the next downturn will be fundamentally different from those we've experienced over the past decades."

Envisioning a "Flatter" Future Cycle

In Kwak's vision, even if demand growth moderates, the memory market is unlikely to witness the sharp, cliff-like price drops and inventory gluts of history. Key factors contributing to this changed outlook include:

  • Diversified Demand Structure: AI demand layers onto traditional drivers like consumer electronics and PCs, smoothing overall volatility.
  • Higher Technological Barriers: The massive R&D and capital expenditure required for advanced memory chips limit reckless capacity expansion by new entrants.
  • Enhanced Product Value: As a key component in AI systems, its value share has increased, leading to potentially higher customer tolerance for price fluctuations.

"Therefore, even if a downturn arrives," Kwak concludes, "I don't think it will be a sharp decline. It's more likely to be a gradual slowdown in demand, possibly even stabilizing for a period." This "flatter" cycle could become a defining characteristic of the semiconductor industry in the AI age.