AI 芯片性能越来越受内存瓶颈限制，做 AI 基础设施或芯片投资的团队需要理解 HBM 的产能节奏——SK hynix 的扩产计划直接关系到 GPU 供应和成本，值得关注。
SK hynix 表示，AI 内存需求巨大，计划在 5 年内将晶圆产能翻倍，但供应紧张预计将持续到 2030 年。晶圆是制造内存芯片的硅基板，翻倍产能意味着扩大物理产出基础，而非仅提高现有产线利用率。AI 供应受限于内存制造的物理节奏，如晶圆、封装、良率和供应协议，其速度远慢于 GPU 路线图。压力主要来自 HBM（高带宽内存），这种堆叠内存用于英伟达 GPU，扩展困难，需要先进 DRAM、堆叠、封装、测试及与 GPU 设计方的紧密协作。SK hynix 正与英伟达和台积电合作开发 HBM4 基础芯片。全球内存市场中，DRAM 前三强（三星、SK hynix、美光）控制约 90% 营收，SK hynix 在 HBM 领域以 58% 份额领先。
SK hynix just said AI memory demand is now so larg…
SK hynix just said AI memory demand is now so large that it will double wafer capacity within 5 years, yet still expects supply to stay tight until 2030.
A wafer is the round silicon starting plate that becomes thousands of memory chips, so doubling wafer capacity basically means SK hynix is trying to expand the physical base of its chip output, not just run current lines harder.
AI supply is increasingly constrained by the physical rhythm of memory manufacturing, where wafers, packaging, yields, and supply agreements move far slower than GPU roadmaps.
The pressure comes from HBM (High-bandwidth memor), the stacked memory used beside Nvidia GPUs.
HBM is hard to scale because it needs advanced DRAM, stacking, packaging, testing, and close work with GPU designers, which is why SK hynix is working with Nvidia and TSMC on HBM4 base dies.
--- The global memory market.
The global memory market has 2 main layers: DRAM, which includes the memory used next to CPUs and AI GPUs, and NAND flash, which is the storage inside SSDs, phones, and data centers.
In DRAM, the market is extremely concentrated, with Samsung at 38.5%, SK hynix at 28.8%, and Micron at 22.4% in 1Q26, meaning the top 3 control about 90% of global DRAM revenue.
In HBM, which is a premium submarket inside DRAM, the AI-specific memory used beside Nvidia GPUs, SK hynix is the market leader, with 58% share in 1Q26, while Samsung and Micron each had 21%.
HBM, or High Bandwidth Memory, is a special form of DRAM built for extreme data movement.
The difference is physical design.
Normal DRAM chips usually sit on memory modules or near the processor, and data moves through relatively narrower connections.
HBM stacks multiple DRAM dies vertically and places them very close to the GPU through advanced packaging, which creates a much wider data path.
That wider path gives AI chips much higher memory bandwidth, meaning the GPU can receive data faster instead of sitting idle.
- IT之家06-02 06:57原文