像 Microsoft、Anthropic 和 Meta 这样的大型人工智能公司越来越多地在没有备用电源（如发电机或中央 UPS 系统）的情况下建造数据中心。这一趋势由 SemiAnalysis 跟踪，涉及超过 15GW 的容量。此决定的驱动因素是节省成本、加快部署时间以及人工智能训练工作负载对因持续检查点而导致的停机的容忍度。 AI
影响 数据中心基础设施设计的这一转变可以通过降低大规模计算资源的成本和部署时间来加速人工智能的发展。
排序理由 来自专业分析公司的重要行业趋势报告。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
This is not a small-deployment thing, top dogs are doing it. Microsoft Fairwater runs the GPU hall fully unbacked, gensets only on the network core. Anthropic's datacenter design also strips the genset. Meta's AI datacenters (the two that look like a Rectangle and Tent) show zero…
But why run a GPU cluster with zero backup power? 🟠 training doesn't care, constant checkpointing tolerates downtime 🟠 build speed and cost, realizing capex savings and cutting lead times, permits, commissioning, and multi-vendor contracts. Today, bringing compute online f
SemiAnalysis Datacenter team tracks all dc sites. Datacenter team started noticing new builds missing the gensets. Datacenter team checked the Industrials Model and now tracks 15GW+ of capacity with no gensets, no central UPS, or neither. (1/3)🧵 https://t.co/IU5wVkoQqI