According to Satya Nadella, Microsoft has deployed a supercomputing cluster utilizing NVIDIA GB300 GPUs, featuring over 4600 GPUs and next-generation InfiniBand connectivity (source: Satya Nadella on Twitter). This marks a significant milestone in AI infrastructure, as Microsoft plans to scale up to hundreds of thousands of GB300s across its data centers. The company is rethinking every layer of its tech stack—spanning silicon, systems, and software—to meet the demands of next-generation AI workloads. This move positions Microsoft to lead in training large-scale AI models and accelerates the adoption of generative AI applications for enterprise and research markets.

According to @satyanadella, Microsoft is building a highly fungible and flexible AI infrastructure to address real-world needs in both inference and training, as discussed by @scottgu with @Kantrowitz (source: Twitter). Microsoft’s infrastructure already powers major AI workloads such as Copilot and ChatGPT, as well as APIs supporting third-party products and enterprise-scale training. This approach allows businesses to deploy and scale AI solutions efficiently, offering significant market advantages in terms of adaptability and reliability for diverse AI applications (source: Twitter).