英伟达周三在柏林GTC大会上表示，欧洲基础设施公司正在基于英伟达DSX平台建设人工智能数据中心。这家芯片制造商称，该蓝图可将冷却用水量降至几乎为零。
European infrastructure companies are building AI data centres on Nvidia DSX, Nvidia said on Wednesday at its GTC Berlin conference. The chipmaker says the blueprint can cut cooling water use to almost zero.
这一消息发布的前一天，德国刚在柏林投资峰会上与英伟达签署了人工智能合作伙伴关系。
The news came a day after Germany signed an AI partnership with Nvidia at a Berlin investment summit.
英伟达将这些设施称为“人工智能工厂”：即为大规模训练和运行人工智能模型而建造的数据中心。DSX是其参考设计，一套将芯片、网络、电力和冷却视为一个整体系统的方案与软件。英伟达于2026年5月31日在台北GTC大会上首次展示了该平台。
Nvidia calls these sites AI factories: data centres built to train and run AI models at scale. DSX is its reference design, a set of plans and software that treats chips, networking, power, and cooling as one system. Nvidia first presented the platform at GTC Taipei on May 31, 2026.
节水来自闭环液冷技术：冷却液流过芯片，然后通过干冷器将热量释放到外部空气中，而不是通过蒸发水来散热。
The water savings come from closed-loop liquid cooling: coolant runs past the chips, then releases heat to the outside air through dry coolers instead of evaporating water.
该公司表示，采用这种方式建造的一座2吉瓦数据中心每天可节省5400万升水。该公司补充说，当地天气状况决定在最炎热的日子里是否需要额外冷却。
The company said a two-gigawatt site built this way could save 54 million litres of water a day. Local weather decides whether extra cooling is needed on the hottest days, the company added.
英伟达表示，DSX平台中一款名为MaxLPS的工具可以用最多减少30%的电力产生相同的人工智能输出。总部位于伦敦的Nscale在其冰岛数据中心对四机架的GB300服务器进行了测试，英伟达称，在同样的264千瓦功率预算内，推理吞吐量提升了49%。
Nvidia said a DSX tool called MaxLPS can produce the same AI output with up to 30% less electricity. London-based Nscale tested it on four racks of GB300 servers at its Iceland site, and Nvidia said inference throughput rose 49% within the same 264-kilowatt power budget.
Nscale还在葡萄牙建设一座数据中心，该公司正在使用DSX在施工前为这两个数据中心制作数字模型。其首席执行官乔什·佩恩用一句话概括了理由。
Nscale, which is also building a data centre in Portugal, is using DSX to make digital models of both sites before construction. Its chief executive, Josh Payne, put the case in one line.
“欧洲的人工智能雄心要求我们让每一兆瓦都发挥价值，”佩恩说。
“Europe’s AI ambitions demand that we make every megawatt count,” Payne said.
更广泛的欧洲工程公司群体正在为该平台设计零部件。西门子正在模拟电力、冷却和噪音。
A wider group of European engineering firms is designing parts for the platform. Siemens is simulating power, cooling and noise.
施耐德电气正在发布电力和冷却设计方案，达索系统正在构建维谛技术设备的数字孪生模型。ABB、西门子能源、日立能源和特灵科技也参与其中。
Schneider Electric is publishing power and cooling designs, and Dassault Systèmes is building digital twins of Vertiv equipment. ABB, Siemens Energy, Hitachi Energy and Trane Technologies are also taking part.
在英国，国家电网（National Grid）与云计算公司 Nebius 共同进行了一项试验，测试了由 Nvidia 芯片组成的集群在电网负荷紧张时的响应能力。Nvidia 表示，在模拟的电网故障情况下，该芯片集群的功耗降低了多达 40%，同时关键业务仍能正常运行。
In the UK, a trial with National Grid and cloud firm Nebius tested how a cluster of Nvidia chips responds to grid strain. Nvidia said the cluster cut its power draw by up to 40% during simulated grid events while critical work kept running.
“欧洲的技术领先地位取决于我们能否高效地将电力和基础设施转化为智能资源，”Nvidia 的 DSX 产品管理部门负责人 Warren Barkley 如是说。
“Europe’s technology leadership depends on how efficiently we turn power and infrastructure into intelligence,” said Warren Barkley, who leads DSX product management at Nvidia.
Nvidia 尚未公布这些新功能何时会正式上线。目前，其部分 DSX 操作软件工具仍处于开发者预览阶段。
Nvidia gave no dates for when the new features will be generally available. Several of its DSX operating software tools are in developer preview.