新型电源模式变革：AI数据中心之UPS

The new power dynamic: AI data centers begin at the UPS

July 3, 2026 Huan Shen

过去，不间断电源（UPS）在数据中心基础设施中扮演着后备电源的角色，但现在它已成为在线运营的中心。人工智能工作负载正在推动数据中心电源动态的深刻变化，而新型智能紧凑型ups旨在解决这一问题。因此，UPS作为设施和电网之间的枢纽，用来缓冲电网的波动和调节，以保护电网和计算层。

The uninterruptible power supply (UPS) used to play a passive role in data center infrastructure, but now it’s at the center of operations. AI workloads are driving profound changes in data center power dynamics, which new intelligent, compact UPSs are designed to address. As such, the UPS acts as a bridge between the facility and the power grid, buffering grid volatility and conditioning power to protect both the grid and the compute layer.

截至目前，数据中心的电力负荷还相对稳定。但人工智能集群引入同步GPU（图形处理单元）活动，触发快速的负载波动、高瞬间峰值和急剧的斜坡率——这些波动都在几秒钟内完成，每年产生数百万个波动周期。

Data center power loads were relatively stable until recently. But AI clusters introduce synchronized GPU (graphics processing unit) activity that triggers rapid load swings, high transient peaks, and sharp ramp rates – all within seconds, resulting in millions of load cycles every year.

在短时间内，电力负荷可以从30%到150%不等。在过去，超过100%的负载被认为是异常的，是引起告警的原因，但今天它被认为是正常波动范围，以适应人工智能负载的变化。这些新的电力动态需要新一代的智能电力系统。

Power loads can swing from 30% to as much as 150% in short bursts. In the past, a load above 100% was considered abnormal and a cause for alarm, but today it is considered normal behavior to accommodate AI load variations. These new power dynamics require a new generation of intelligent power systems.

数据中心负荷的剧烈波动会破坏电网的稳定。当数百兆瓦的变化在几秒钟内发生时，运营商没有能力平衡电力供应。

Dramatic data center load swings can destabilize the electric grid. Operators are not equipped to balance the power supply when variations of hundreds of megawatts occur in just seconds.

这就是UPS的新角色发挥作用的地方。在一些模型中，负荷平坦化和故障波动（FRT）有助于维持电网的稳定性。FRT允许UPS在电压下降或瞬态故障时保持供电并继续运行，而不是立即断开，从而避免了不稳定性因素。这些特性保护了电网与负载，同时也帮助数据中心遵守联邦和州为维护电网稳定性而引入的要求。

This is where UPS’s new role comes in. Load-flattening and fault ride-through (FRT) in some models help maintain grid stability. FRT allows UPS to remain connected and continue operating through voltage dips or transient faults, rather than disconnecting immediately, thereby avoiding additional instability. Such features protect both sides of the grid connection while also helping data centers comply with federal and state requirements being introduced to maintain grid stability.

当数据中心通过燃气轮机或柴油发电机产生自己的电力时，先进的UPS功能也开始出现。由于电力供应的限制，一些人工智能数据中心，比如德克萨斯州阿贝林正在建设的大型星际之门设施，是离网运行的，自己发电。

Advanced UPS capabilities also begin when data centers generate their own power with gas turbines or diesel generators. Because of power supply constraints, some AI data centers, such as the massive Stargate facility under construction in Abeline, Texas, operate off-grid, generating their own electricity.

负荷变化会损坏发电设备。具有负载平滑功能的ups，包括一些新的施耐德电气Galaxy V系列型号，可以切换到保护燃气轮机和柴油发电机。当燃气轮机的负载变化达到每秒1%，柴油发电机的负载变化达到每秒10%时，就会发生这种情况。

Load variability can damage power-generating equipment. UPSs with load-smoothing features, including some new Schneider Electric Galaxy V series models, can switch to protect gas turbines and diesel generators. This occurs when load variations reach 1% per second with gas turbines, and 10% per second with diesel generators.

人工智能数据中心的运营正在改变电力不稳定带来的支出费用。在传统的数据中心，短暂的电力事件通常被视为运营中断。在人工智能环境中，它们可以直接威胁到大规模计算投资的价值。即使是很小的干扰也会中断GPU训练周期，减慢推理操作，降低昂贵的AI集群的利用率，并延迟对关键业务应用程序的洞察时间。

AI data center operations are changing the financial consequences of power instability. In traditional data centers, brief power events were often viewed as operational disruptions. In AI environments, they can directly threaten the value of massive compute investments. Even minor disturbances can interrupt GPU training cycles, slow inference operations, reduce utilization of expensive AI clusters, and delay time to insight for business-critical applications.

由于人工智能工作负载依赖于在极端性能水平下运行的紧密同步、高密度系统，因此风险尤其高。随着人工智能部署的规模和紧迫性不断增长，电力弹性不再仅仅是一个设施问题。它已经成为直接影响AI计划的成功和盈利能力的业务连续性、投资保护和ROI需求。

The stakes are especially high because AI workloads depend on tightly synchronized, high-density systems operating at extreme performance levels. As AI deployments grow in scale and urgency, power resilience is no longer simply a facilities concern. It has become a business continuity, investment protection, and ROI requirement that directly affects the success and profitability of AI initiatives.

AI数据中心的规划应该在服务器设计之初就已经做好电源架构决策。传统的“UPS-last”设计方法将备用电源视为后备型组件，这与人工智能部署相比日益脱节。

Planning for AI-ready data centers should include power architecture decisions made long before servers are installed. Traditional “UPS-last” design approaches, where backup power is treated as a downstream utility component, are increasingly out of step with the realities of AI deployments.

AI数据中心负载具备快速负载波动、高机架密度和电网敏感功率动态，要求UPS从设计过程一开始就发挥更具战略性的作用。通过将UPS与数据计算、设备冷却和设施规划一起视为基础设施，运营商可以构建更高效的基础设施，提高弹性，并更好地支持未来的电网交互功能。

AI workloads introduce rapid load swings, high rack densities, and grid-sensitive power dynamics, requiring the UPS to play a more strategic role from the outset of the design process. By treating the UPS as a foundational element alongside computing, cooling, and facility planning, operators can build infrastructure that scales more efficiently, improves resilience, and better supports future grid-interactive capabilities.

部署适应AI数据中心的ups，如施耐德电气的Galaxy VXL，带来的好处远远超出了备用电源。AI负载平滑和FRT等未来功能将帮助运营商稳定动态AI工作负载，同时减少对峰值需求过度配置基础设施的需求。这使得更高效的电源架构能够降低资本成本，提高利用率，并随着人工智能环境的密度和复杂性不断增长而支持长期可扩展性。

Deploying AI-tolerant UPSs, such as Schneider Electric’s Galaxy VXL, delivers benefits that extend far beyond backup power. Future features such as AI load smoothing and FRT will help operators stabilize dynamic AI workloads while reducing the need to overprovision infrastructure for peak demand. This enables a more efficient power architecture that lowers capital costs, improves utilization, and supports long-term scalability as AI environments continue to grow in density and complexity.

Galaxy VXL UPS还解决了人工智能数据中心最大的限制之一：空间。最新型号的密度是前几代的两倍，在高密度环境中占用的空间减少了52%，为产生收入的计算基础设施腾出了宝贵的空间。此外，施耐德电气将UPS创新与NVIDIA GPU路线图紧密结合，以确保电力系统跟上快速增长的人工智能性能需求

The Galaxy VXL UPS also addresses one of the biggest constraints in AI data centers: space. The latest models are up to two times denser than previous generations and occupy 52% less footprint in high-density environments, freeing valuable floor space for revenue-generating compute infrastructure. In addition, Schneider Electric closely aligns UPS innovation with NVIDIA GPU roadmaps to ensure power systems keep pace with rapidly increasing AI performance demands.

随着人工智能部署的扩大，UPS技术将发挥越来越重要的战略作用。除了保护关键基础设施外，支持人工智能的UPS还有助于优化人工智能投资，增强运营弹性，并减轻大规模人工智能工作负载对电网的影响。Galaxy VXL可以帮助稳定AI工作负载并优化电源性能。

As AI deployments expand, UPS technology will play an increasingly strategic role. In addition to protecting critical infrastructure, AI-ready UPSs also help optimize AI investments, enhance operational resilience, and mitigate the impact of large-scale AI workloads on the electric grid. Learn how the Galaxy VXL helps stabilize AI workloads and optimize power performance.

DKC交流群邀请

深知社创立于2017年，是数字能源行业工程技术人员学习、交流、合作、创新的虚拟社区平台。深知社秉承互联网精神，践行工程师文化，提倡全球视野、终身学习、乐于分享的博学深知理念，感兴趣的同学可以扫描下方二维码申请进群交流学习。

翻译：

谢亚成

天津市天河计算机技术有限公司 暖通工程师

DKI（DeepKnowledge Institute）高级研究员

公众号声明：

本文并非官方认可的中文版本，仅供读者学习参考，不得用于任何商业用途，文章内容请以英文原版为准，本文不代表深知社观点。文中内容来自互联网，如有侵权，将在24小时内删除。中文版未经公众号DeepKnowledge文字授权，请勿转载。

推荐阅读：