This project is suspended: no new articles or editions will be published. The archive stays available.
HBM (High Bandwidth Memory) is a stacked memory chip design that moves data to and from AI accelerators far faster than conventional DRAM, making it a key ingredient in modern AI GPUs. This hub covers HBM makers, supply constraints, and how memory bandwidth shapes AI chip performance and cost.
SK Hynix will buy back and cancel $28.6 billion in shares, the largest such program in Korean corporate history, after its stock fell more than 50% in two months despite record AI memory demand.
Read more →Shanghai startup Dongfang Suanxin unveiled its DF1000 AI chip on July 13, bypassing US export curbs with 3D memory stacking instead of banned high-bandwidth memory.
Read more →HBM is a specialized memory technology that stacks chips vertically to deliver 10–20× the data throughput of conventional DRAM — making it the critical ingredient in every major AI accelerator today.
Read more →The South Korean chipmaker behind more than half the world's HBM memory—the critical component inside AI accelerators—is seeking $29.4 billion on Nasdaq to fund four new semiconductor plants.
Read more →