Core Insights - Microsoft has launched a new data center in Atlanta, referred to as an "AI super factory," which connects to other data centers and provides computational power from hundreds of thousands of NVIDIA GPUs to support AI workloads [1][2] - The Atlanta facility is the second in the "Fairwater" series and integrates with the first site in Wisconsin, creating the world's first truly cross-state AI computing cluster [1] - The concept of the "AI super factory" differs from traditional data centers by merging geographically dispersed centers into a single virtual supercomputer, allowing for a unified approach to AI model training [1] Group 1 - Microsoft CEO Satya Nadella emphasized the vision of interchangeable infrastructure that can operate any workload with maximum performance and efficiency [2] - Each Fairwater data center can integrate hundreds of thousands of the latest NVIDIA GPUs into a unified cluster, ensuring no GPU is unnecessarily idle [2] - Over 100,000 GB300 GPUs have been deployed this quarter for inference tasks across other infrastructure clusters [2] Group 2 - The scale of infrastructure required for training AI models is now several times larger than just one or two data centers [2] - The data centers are interconnected through a dedicated AI wide-area network using specialized fiber optics [2] - The future of AI will be shaped by connecting data centers into a unified distributed system [2]
微软(MSFT.US)“AI超级工厂“启动!整合数十万GPU,可实现多个数据中心互联