Analysis · AI Weekly ·

SK Group chair warns of 60-100% jump in AI memory demand as capacity lags

Speaking at the Korea Chamber of Commerce and Industry's Jeju Forum on July 19, SK Group Chair Chey Tae-won said foreign governments now treat AI memory access as a matter of economic security, and warned that customers are asking for 60-100% more AI memory capacity in 2027 while, in his words, no company currently has meaningful new capacity coming online to meet it.

Based on reporting by AI Weekly — analysis by dalili

SK Hynix, the SK Group subsidiary that supplies high-bandwidth memory (HBM) to Nvidia and other AI hardware makers, sits at the center of a supply chain that has become as strategically sensitive as the GPUs themselves. Chey's remarks frame AI memory not as a component shortage in the traditional sense but as a geopolitical chokepoint: governments negotiating for guaranteed allocation, treating access the way they might treat energy or rare-earth supply.

The 60-100% demand increase Chey cited for 2027 reflects the compounding effect of frontier model training runs growing larger and inference workloads scaling with adoption — both of which are memory-bound in ways that pure compute capacity additions don't fully address. His characterization that no company currently has meaningful new capacity coming online points to the multi-year lead times required to bring new memory fabrication capacity to market, meaning any shortfall identified today will not be resolved by 2027 regardless of investment decisions made now.

The comments arrive as memory pricing and allocation have already become a recurring flashpoint in AI hardware supply chains through 2026, with hyperscalers reportedly securing long-term supply commitments well in advance of need.

Key takeaways

  • SK Group Chair Chey Tae-won: foreign governments now treat AI memory access as economic security
  • Customers are requesting 60-100% more AI memory capacity for 2027
  • No company currently has meaningful new capacity coming online to meet that demand, per Chey

Why it matters

If AI memory becomes a governmental allocation issue the way advanced chips already have, it adds another layer of supply-chain fragility that every AI-dependent economy — including those building sovereign compute capacity — will need to plan around.