This chart is one of the strongest pieces of evidence supporting a prolonged memory upcycle, and it helps explain why investors continue to underestimate the earnings power of the DRAM industry.
The key takeaway is simple: demand growth is running materially ahead of supply growth for the next several years. According to the forecast, total DRAM demand rises from 1,921k WSPM in 2025 to 3,775k WSPM by 2030, implying roughly 14% annual growth. Supply, however, is expected to grow only around 12% annually. That 2% gap may sound small, but in a capital-intensive industry operating near full utilization, it creates significant shortages.
The projected shortfall becomes increasingly severe, reaching roughly 507k wafers per month in 2027 and peaking near 795k wafers per month in 2028. Demand-to-supply ratios of 120-130% are historically associated with aggressive pricing power for memory producers.
What is driving this imbalance is not traditional DRAM demand. It is HBM. HBM wafer requirements are projected to increase from 350k WSPM in 2025 to 1.3 million WSPM by 2029-2030. In other words, virtually all incremental wafer demand over the next five years comes from AI memory. HBM’s share of total DRAM wafer starts rises from just 18% today to roughly 35% by the end of the decade.
This matters because HBM consumes disproportionately more wafer capacity, more packaging capacity, and more manufacturing complexity than conventional DRAM. Every wafer allocated to HBM is effectively a wafer unavailable for commodity DRAM production.
The result is a structural tightening across the entire memory ecosystem. Even customers that do not directly buy HBM may experience higher DRAM pricing because industry capacity is increasingly diverted toward AI applications.
The market still tends to think about memory through the lens of previous boom-bust cycles where suppliers aggressively added capacity and eventually destroyed pricing. This cycle looks different. Samsung, SK Hynix, and Micron appear far more disciplined, while HBM manufacturing complexity makes rapid capacity additions significantly harder than in prior generations. If these forecasts prove directionally correct, the industry may be entering a multi-year period where memory becomes one of the largest beneficiaries of AI infrastructure spending.
That is why our preferred positioning remains unchanged: long SK Hynix, Micron, and Samsung, in that order. SK Hynix remains the clear HBM technology leader and should capture the largest share of AI-driven profit growth. Micron continues to close the technology gap and offers significant earnings leverage as HBM volumes scale. Samsung remains the largest long-term strategic player, but execution challenges have delayed its participation relative to peers.
The broader implication is that AI is no longer just a GPU story. Memory is rapidly becoming one of the most important bottlenecks in the entire AI stack. When bottlenecks emerge, pricing power follows. And when pricing power meets constrained supply, earnings revisions tend to move sharply higher.
골드만삭스, 커패시터가 새로운 메모리다⤴️
#삼성전기
골드만삭스는 AI 서버 전력 소비 증가로 커패시터(MLCC)가 HBM 메모리 다음 병목 자원이 될 것으로 분석하며, 이를 ‘새로운 메모리’로 평가했다.
• AI 데이터센터, 전기차, 5G 등으로 커패시터 수요가 급증해 공급 부족이 발생할 전망이며, 전력 관리와 안정화에 필수적인 역할을 한다.
• 무라타, TDK, 야��오 등 주요 공급사 주가가 메모리株처럼 상승할 가능성이 크며, AI-driven MLCC 슈퍼사이클이 2030년까지 이어질 수 있다.