While OpenAI races to build AI data centers, Nadella reminds us that Microsoft already has them

On Thursday, Microsoft CEO Satya Nadella showcased the company's inaugural large-scale AI system, often referred to as an AI "factory" by Nvidia. He indicated that this installation is just the beginning, with plans for numerous Nvidia AI factories to be rolled out across Microsoft Azure’s extensive global data centers, specifically to handle workloads for OpenAI. Each of these advanced systems comprises over 4,600 Nvidia GB300 rack computers, featuring the highly sought-after Blackwell Ultra GPU chip. These machines are interconnected using Nvidia's cutting-edge InfiniBand networking technology, a strategic acquisition from Mellanox in 2019 for $6.9 billion that has positioned Nvidia favorably in the market. Microsoft is set to deploy “hundreds of thousands” of Blackwell Ultra GPUs globally, marking a significant expansion in its AI capabilities. The scale of these systems is impressive, with Microsoft providing a wealth of technical details for those interested in the intricate hardware. However, the timing of this announcement holds particular significance. It follows closely on the heels of OpenAI, Microsoft’s partner and occasional competitor, securing major data center agreements with Nvidia and AMD. By 2025, OpenAI is projected to have accumulated around $1 trillion in commitments for its own data centers, as its CEO Sam Altman hinted at further developments in the pipeline. In response, Microsoft is eager to highlight that it already possesses a robust network of over 300 data centers across 34 countries, positioning itself as uniquely capable of meeting the current demands of advanced AI technologies. These formidable AI systems are also engineered to support the next generation of models, which are expected to handle “hundreds of trillions of parameters.” More details about Microsoft's strategy to enhance its AI infrastructure are anticipated later this month, with Microsoft CTO Kevin Scott scheduled to speak at TechCrunch Disrupt from October 27 to 29 in San Francisco.