HBM Demand Explodes! UBS: Global Memory Chip Sales Hit Record $74.6 Billion in June, Supply Gap Expected to Last Until 2028

UBS has recently released its 'Monthly Memory Chip Report,' indicating that the global memory chip market achieved its strongest monthly performance in history last month, with global memory sales reaching $74.6 billion, a 31.7% month-on-month increase.

UBS assesses that as AI demand continues to grow and long-term agreements (LTAs) expand, the memory cycle will 'further strengthen,' with structural supply-demand imbalances expected to persist at least until mid-2028.

Detailed data shows DRAM dominates the memory market, with June sales reaching approximately $48 billion, up 27.7% month-on-month. NAND Flash performed even more strongly, surging 40.7% to $25.8 billion, setting a new record, primarily driven by explosive demand for data center SSDs due to soaring AI training and inference workloads.

UBS forecasts the memory industry's total revenue this year will reach $992 billion, nearly doubling to $1.76 trillion by 2027.

The core driver of growth is HBM. UBS predicts HBM demand will increase 90% year-on-year this year to approximately 331 billion Gb, and rise another 77% to 587 billion Gb by 2027.

UBS has also revised its contract price estimates upward: DDR memory contract prices are expected to increase 32% in Q3 and another 18% in Q4 of this year. NAND prices are projected to rise 30% in Q3 and 12% in Q4. The DRAM industry will maintain a structural shortage until at least Q2 2028, with demand growing 36.2% annually in 2027, far outpacing supply growth of 19.3%.

UBS highlights Micron, Samsung Electronics, and SK Hynix as the main beneficiaries of this pricing cycle, with the three collectively dominating global DRAM and NAND supply.

However, the investment bank warns that the sustainability of AI capital expenditures by hyperscale cloud providers and customer price tolerance represent the primary downside risks to this unprecedented upcycle. 'Demand-side, not supply-side, is the variable that cannot be fully modeled,' the report states.