As OpenAI actively plans to build its own data centers, Microsoft CEO Satya Nadella showcased its operational large-scale AI system to the industry via a tweet. This system, referred to by Nvidia as an AI “factory,” has now become the first instance within Microsoft Azure’s global data center network to run OpenAI workloads. Nadella explicitly stated that this is merely “the first of many similar systems”.

The AI factory cluster consists of over 4,600 rack-mounted Nvidia GB300 computing units, equipped with the highly demanded Blackwell Ultra GPU chips and interconnected via Nvidia’s InfiniBand ultra-high-speed networking technology. Notably, Nvidia CEO Jensen Huang acquired Mellanox for $6.9 billion in 2019, a visionary move that solidified Nvidia’s dominant position in the InfiniBand technology market.

Microsoft has pledged to deploy “hundreds of thousands of Blackwell Ultra GPUs” worldwide to expand these systems. Beyond the impressive scale of the systems, the company also shared extensive technical details for hardware enthusiasts to delve into. The timing of this announcement is equally noteworthy—just recently, its partner and competitor OpenAI had signed two highly anticipated data center agreements with Nvidia and AMD.

According to industry estimates, OpenAI has secured commitments totaling approximately $1 trillion for data center construction by 2025. This week, CEO Sam Altman further announced plans to continue expanding construction efforts. Microsoft’s timing in releasing this news clearly aims to demonstrate to the world that the company already operates over 300 data centers across 34 countries, with these facilities being “uniquely advantaged” and capable of “meeting the demands of cutting-edge AI today.” Microsoft emphasized that these massive AI systems also possess the capability to run next-generation models with parameter scales reaching “trillions.”.