AI 模型和平台
在Ulanqab AI数据中心部署基于SST的800V DC电源
在 Google 上的首选来源中添加 Unite.AI
2026年11月2日的Vision 称，一个整合的800伏直流电(800V DC)电力系统目前正在中国乌兰卡布的Galaxy AI数据中心校园为现场计算工作量提供动力
称系统提供电源终端，从中压AC输入到服务器架，并且它已经在商业上运行
固态变压器和实时协调
新部署的系统整合了三个硬件层：固态变压器(SST)，电池能量存储(BESS)，以及用于计算硬件的调整直流电压的设备。 由Envision的Dubhe Controller(Dubhe Energy Foundation Model)来协调
根据Envision，部署是围绕三个能力构建的：稳定SST直接从10千伏中电压转换为800V的DC总线，跨越运行条件；电池存储既提供二级功率支持，又提供分钟级备份；以及SST与存储之间的毫秒级协调，吸收当地GPU负载波动。
Envision 称的800V DC架构简化了AI数据中心动力系统，引用了更高的功率转换和瓦特对位效率，更紧凑的设备足迹，减少了电缆和铜的要求。 这些优势，称公司可以降低基础设施成本，支持同一空间内更高的计算密度。
人均电力需求上升
公告将部署设定在一个具体的操作问题上：随着AI计算需求加速，操作员必须在受限制的空间内为每个机架提供更多的动力，同时随着计算负荷的剧烈波动而保持供应稳定。
“大赦国际数据中心需要能够协调可再生能源、能源储存和电网的电力系统，以满足计算需求。” 黄 Z称直播系统展示端对端架构，将可再生供应与计算需求连接起来，称这是将功率计算协调从战略愿景转向日常操作实践的关键一步。
银河校园和特派团
位于内蒙古乌兰卡布的银河校区于2026年8月开始运作，当时Envision在8月6日以12万平方公尺的AI超计算设施作为锚地，根据该公司当时的调试公告。 该记录称，该校区设计规模超过2GW容量；11月2日发布时表示，该校区规划容量为3GW和称，旨在支持最多100万台AI加速器在全面建设时平行运行，其80%以上的电力计划来自直接可再生电力连接。
Envision的Galaxy校区网页将该网站作为净零AI数据中心进行市场营销，在10年价格锁、捆绑的电力和绿色证书以及60%的绿色电力直接供电率下，宣传长期固定电价低于电网20%。
它的目标是通过Mission Gobi扩大这一方法。 Mission Gobi是2026年6月在巴黎VivaTech的一个举措，以Galaxy为第一个旗舰项目。 Mission Gobi的目标是到2030年全球沙漠地区绿色AI数据中心容量的5GW，该公司将之作为可复制的蓝图，认为开发甚至1%的全球沙漠和Gobi地区可以以极具竞争力的成本支持千兆瓦规模的计算能力。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
AI Models & Platforms
Envision Deploys SST-Based 800V DC Power at Ulanqab AI Data Center
Add Unite.AI to your preferred sources on Google
Envision said on Nov. 2, 2026, that an integrated 800-volt direct-current (800V DC) power system is now powering live computing workloads at its Galaxy AI data center campus in Ulanqab, China, a deployment the company described in its announcement as the world’s first commercial use of such a system in a large-scale AI data center.
The announcement says the system delivers power end to end, from medium-voltage AC input to the server rack, and that it is already operating commercially ahead of wider industry adoption, providing a reference for AI data centers moving to 800V DC power infrastructure.
Solid-State Transformers and Real-Time Coordination
The newly deployed system integrates three hardware layers: solid-state transformers (SST), battery energy storage (BESS), and equipment that adjusts direct-current voltage for computing hardware. Coordinating them is Envision’s Dubhe Controller, which runs with the company’s Dubhe Energy Foundation Model and adjusts the transformers and batteries in real time so that power supply tracks computing demand, including abrupt swings in load.
According to Envision, the deployment is built around three capabilities: stable SST conversion directly from 10-kilovolt medium voltage to an 800V DC bus across operating conditions; battery storage that provides both second-scale power support and minute-scale backup; and millisecond-level coordination between SST and storage that absorbs GPU load fluctuations locally.
Envision said the 800V DC architecture simplifies AI data center power systems, citing higher power conversion and watt-to-bit efficiency, a more compact equipment footprint, and reduced cabling and copper requirements. Those advantages, the company said, can lower infrastructure costs and support higher computing density within the same space.
Rising Power Demand Per Rack
The announcement frames the deployment against a specific operational problem: as AI computing demand accelerates, operators must deliver more power to each rack within constrained space while keeping supply stable as computing loads swing sharply from moment to moment.
“AI data centers need power systems that coordinate renewable energy, energy storage, and the grid to meet computing demand,” said Kevin Huang, Senior Vice President of Envision Energy and President of Energy Storage and AIDC Power Systems. Huang said the live system demonstrates an end-to-end architecture connecting renewable supply with computing demand, calling it a critical step in moving power-compute coordination from strategic vision to day-to-day operating practice.
Galaxy Campus and Mission Gobi
The Galaxy campus, in Ulanqab, Inner Mongolia, began operations in August 2026, when Envision commissioned the site on Aug. 6 anchored by a 120,000-square-meter AI supercomputing facility, according to the company’s commissioning announcement at the time. That record described the campus as designed to scale to over 2GW of capacity; the Nov. 2 release states a planned capacity of 3GW and says the campus is designed to support up to one million AI accelerators operating in parallel at full build-out, with more than 80% of its power planned to come from direct renewable electricity connections.
Envision’s Galaxy campus page markets the site as a net-zero AI data center, advertising long-term fixed electricity pricing at 20% below grid rates under a 10-year price lock, bundled electricity and green certificates, and a 60% green power direct-supply ratio.
Envision said it aims to scale the approach through Mission Gobi, an initiative unveiled at VivaTech in Paris in June 2026 with Galaxy as its first flagship project. Mission Gobi targets 5GW of green AI data center capacity in desert regions worldwide by 2030, and the company presents it as a replicable blueprint, arguing that developing even 1% of global desert and Gobi regions could support computing capacity at terawatt scale at highly competitive cost.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.