[REVISION HISTORY]

AI demand drives HBM shortage and structural memory shifts

42 recorded revisions of this situation.

What changed

2026-10-28 12:53 UTC → 2026-11-02 07:52 UTC · added ~~removed~~

The global memory shortage driven by AI demand is intensifying, with industry leaders warning that supply constraints could persist for years. While NVIDIA CEO Jensen Huang previously suggested a supply surplus was unlikely before 2029, ~~recent~~ he reaffirmed this at the GTC Berlin conference, stating that demand for HBM and LPDDR will continue to expand through at least 2029. Recent market indicators show DRAM experiencing its highest level of undersupply in ~~decades. Analysts expect DRAM shortages to persist through at least 2027, with Micron Technology reporting~~ decades, and SK hynix has projected that ~~much of its 2027 production is already committed.~~ memory chip demand will outstrip supply through 2030. In the HBM sector, ~~Samsung is negotiating higher prices for its next-generation HBM4, while~~ TrendForce predicts HBM blended average selling prices could increase 121% year-over-year by 2027. The proportion of HBM within total DRAM wafer input is expected to rise from 18% in ~~2027 due~~ late 2025 to ~~persistent constraints.~~ 30% by the end of 2027. To manage costs, Intel has opted for low-power DRAM for certain AI inference GPUs, while NVIDIA is seeking to standardize its next-generation custom HBM (NVHBM) using TSMC’s 3nm process. Technological competition is accelerating. SK hynix is discussing mass production of its iHBM thermal management technology, which uses an Integrated Cooling Element to reduce thermal resistance by 30% for HBM5 products. Samsung is collaborating with NVIDIA on AMMA technology to reduce latency. Meanwhile, major Chinese firms like Alibaba, ByteDance, and Tencent are reportedly negotiating DRAM supply contracts lasting up to ten years to mitigate geopolitical risks and guarantee supply for AI workloads. In the NAND segment, Yangtze Memory Technologies Co. (YMTC) projects supply shortages could last until 2029. These shortages are beginning to impact consumer electronics; for example, Samsung has increased prices for the Galaxy A57 5G in Indonesia, and global smartphone prices rose by an average of 15% in 2026. Technological competition remains fierce. Samsung has unveiled a roadmap for HBM5 and customized high-bandwidth memory, alongside Compute Express Link (CXL) solutions developed with AMD to address AI data demands. SK hynix is advancing its MARS architecture and 3D Ferroelectric NAND (3D FeNAND) to enable Analog Computing-in-Memory, which could significantly improve throughput and energy efficiency. Micron continues to work toward expanding its 1δ DRAM and G11 NAND processes by late 2027, alongside developing vertical HBM (vHBM) for 2028. Reaffirming his earlier stance, Jensen Huang has dismissed concerns of a memory oversupply, asserting that demand for HBM and LPDDR will continue to expand through at least 2029 as AI industry growth continues. Amidst this demand, Micron Technology has projected fiscal year 2027 first-quarter revenue between $60 billion and $63 billion, with an anticipated adjusted gross margin of 86.25 percent.

Versions

Select a version to see what changed compared to the one before it.