精选理由
华为、中兴这些厂商正在抢建10万GPU级别的超节点，因为Kimi K3和GLM-5这类新模型太吃算力了，看看这场基础设施军备竞赛有多激烈。
华为、中兴、新华三、曙光等厂商正在竞相部署从1024到10万GPU规模的AI超节点。驱动这一竞赛的是Kimi K3和GLM-5等大模型对算力的巨大需求。这场基础设施角逐直接反映模型参数规模竞争对异构计算集群的迫切要求。
原文 · Pandaily
Computing Power Becomes the Lifeline: Tech Giants Scramble for AI Supernodes as Model Parameter Competitions Drive Unprecedented Infrastructure Demand
Huawei, ZTE, H3C, Sugon, and others race to build and deploy AI supernodes from 1024 to 100K GPU scale, as Kimi K3 and GLM-5 demand drives computing capacity to the breaking point.