Nvidia cuts HBM specifications on flagship GPUs, trimming memory from 384GB to 192GB
Nvidia reduced the HBM specification on its flagship GPUs from 384GB to 192GB, cutting global HBM demand by about 10% and hitting SK Hynix, which holds roughly 70% of the HBM market. Multiple sources attribute the cut to insufficient HBM4E supply rather than reduced demand.
Detected & updated continuously · Source: Nebula
Story subjects
Sources
@WangNextDoor2
本月老黄先是一个消息带崩了存储SK海力士的股价，又拉了一把光通信的盘，国内的长鑫却稳如老狗，笑看风云变化，想想还是有点意思的。 1. 核心撕裂：HBM 与传统 DRAM 的冰火两重天 海力士遭重创： 英伟达削减旗舰 GPU 的 HBM 规格（384GB 降至 192GB），致全球需求降一成；占 HBM 七成份额的 SK https://t.co/O7mUw8EzgD
@A_Flying_Narwhal
Nvidia is trimming memory because it can’t source enough HBM4E in time, not because it needs less. It actually boosts \*total\* demand: AI companies running models on lower-memory chips, still need to deploy more chips ultimately, and I don’t think SNDK / MU / Samsung will do themselves a disservice on pricing, as the whole ecosystem boosts pricing— some of it being locked in, some of it not. Stock has already fallen on this stale news— and likewise, SNDK makes NAND, not HBM. Still a degen bet with a 10-bagger setup, no doubt there.
@imrickjamesbioch
NVDA raising prices by 15%, HBM is sold out for the year and folks think AI is slowing down? Problem with NVDA is regardless of their beat, insiders will make some bullshit reason to tank the stock so it can squeeze all the call option bag holders out if their money.
@R_and_Invest
The funniest thing ever. Moving from 12Hi to 8Hi drives more GPU and ASICS shipment which would keep demand for HBM (same or slightly higher) & even after this $DRAM HBM is so sold out can’t meet 1/2 the DC demand. But lowering multiple and PT to $1300 on $MU LOL!!😂 https://t.co/fnmDwvYIQ9