NVIDIA DSX POD targets faster Modular AI Data-Center Deployment
36Kr reported that NVIDIA announced DSX POD, a modular reference architecture for the Vera Rubin platform that integrates computing equipment, rack power, networking and liquid cooling for on-site delivery. NVIDIA says deployment can fall from as long as six months to as little as 14 days when external power, cooling and networking are already prepared; partner products are expected from early 2027. The report also reviews modular infrastructure efforts from ZTE and Huawei, a reported 1,024-accelerator Ascend Atlas 950 SuperPoD milestone, and NAND production-line upgrades by Samsung and SK Hynix in China.
DSX POD extends NVIDIA’s role from chips and CUDA into integrated AI-compute delivery, highlighting deployment efficiency as a key part of turning purchased hardware into operational capacity.
From the digest “英伟达的新动作，意外打开了中国AI的另一条路”
资料来源
NVIDIA Newsroom，2026年10月21日，DSX POD AI Factory官方发布。
中兴通讯官网，2026年10月20日，ZXEDC M8000智算微模块产品及项目交付情况。
华为官网，2026年9月16日，3D数据中心架构发布。
IT之家等同期报道，2026年10月15日，1024卡昇腾液冷超节点及Step 5 Preview模型适配进展。
Reuters，2026年10月21日，三星、SK海力士中国NAND工厂技术升级，以及Omdia中国NAND市场份额数据。
本文来自微信公众号“科技旋涡”，作者：王秦州，36氪经授权发布。
#NVIDIA#NVIDIA DSX POD#NVIDIA Vera Rubin#ZTE#ZXEDC M8000#Huawei#Ascend Atlas 950 SuperPoD#Step 5 Preview#Samsung#SK Hynix