AI Fuels DRAM Supercycle Following Years of Stagnation, Propelled by Surge in HBM Adoption and Big Tech’s Custom ASIC Developments

Artificial Intelligence (AI) has made significant inroads into various aspects of the supply chain, particularly with regard to Dynamic Random Access Memory (DRAM).Recent analyses indicate a dramatic escalation in DRAM demand, influenced largely by the expanding AI sector.

Projected DRAM Production Cannot Meet Rising Demand from the AI Sector

The surge in demand for DRAM is driven by the necessities of AI advancement. Each AI cluster operates with a requirement for High Bandwidth Memory (HBM) per chip, resulting in notable spikes in DRAM requirements. Major players in the AI domain are now focusing on the creation of custom Application-Specific Integrated Circuits (ASICs) for their internal infrastructures, which is playing a pivotal role in pushing these DRAM demand forecasts higher. Furthermore, NVIDIA’s soaring demand for its AI-related products has rendered DRAM almost as crucial as chip nodes themselves. A recent report from Chosun Biz and insights from UBS (via Jukan) indicate that DRAM demand is set to experience exponential growth in the near future.

Very interesting. UBS expects that OpenAI’s ASIC will use 12-high HBM3E, and in particular, it projects that Samsung will expand its market share.pic.twitter.com/h7LKPBJ8T4

— Jukan (@Jukanlosreve) October 2, 2025

Diving deeper into UBS’s analysis, it is suggested that OpenAI’s forthcoming ASIC will incorporate the advanced 12-Hi HBM3E technology. This innovation could yield a staggering 500, 000 to 600, 000 DRAM WPM (Writes Per Minute) from 2026 to 2029, representing a substantial segment of the overall DRAM output. This single ASIC competitor underscores the enormous potential for DRAM manufacturers, particularly given the pressing need for increased production capacities. Despite forecasts indicating that the DRAM industry could achieve 1.955 million WPM by 2026, it may still fall short of meeting overall demand.

According to TrendForce, the inventory levels for global DRAM suppliers have plummeted to just 3.3 weeks—marking the lowest level in seven years. Typically, this inventory hovers around 10 weeks, highlighting a concerning tightness in supply. It’s important to recognize that the demand for DRAM isn’t exclusively tied to custom AI chips; it also plays a critical role in data centers and other applications, which we will explore further.

DRAM Memory Supply To Experience Shortages As Manufacturers Shift Focus Towards HBM Production 1 — Samsung DRAM

OpenAI’s Stargate initiative is anticipated to account for a considerable portion of the global DRAM supply, reportedly utilizing around 900, 000 DRAM WPM—representing roughly 40% of the current total supply. Such a level of demand is historically unprecedented, and with the bulk of DRAM manufacturing concentrated in South Korea, the ability of these companies to meet this surge will be a crucial area to monitor.

The future landscape of the DRAM industry is poised for significant changes, particularly with the emergence of HBM4 technology which promises to scale up production even further. The insatiable appetite for HBM from major technology firms necessitates a strategic increase in DRAM supply to keep pace with demand.

Source & Images