Microsoft CEO Satya Nadella on Thursday tweeted a video of the company’s first large-scale AI system, or as Nvidia likes to call it, an AI “factory.” He promised that this would be the “first of its kind” Nvidia AI factories to be deployed across Microsoft Azure’s global data centers to run OpenAI workloads.
Each system is a cluster of more than 4,600 Nvidia GB300 rack computers equipped with highly sought-after Blackwell Ultra GPU chips and connected through Nvidia’s ultra-fast networking technology called InfiniBand. (Beyond AI chips, Nvidia CEO Jensen Huang had the foresight to corner the market with InfiniBand when he acquired Mellanox for $6.9 billion in 2019.)
Microsoft promises to deploy “hundreds of thousands of Blackwell Ultra GPUs” as it rolls out these systems globally. While the scale of these systems is impressive (and the company has shared even more technical details for hardware enthusiasts to peruse), the timing of this announcement is also noteworthy.
This comes on the heels of its partner and well-documented frenemy, OpenAI, inking two high-profile data center deals with Nvidia and AMD. Some estimates suggest that in 2025, OpenAI will reach $1 trillion in commitments to build its own data centers. And CEO Sam Altman said this week there’s more to come.
Microsoft clearly wants the world to know that it already has more than 300 data centers in 34 countries, which are “uniquely positioned” to “meet today’s frontier AI demands,” the company said. These monster AI systems can also run next-generation models with “hundreds of trillions of parameters,” the company says.
Expect more information later this month about how Microsoft is increasing its support for AI workloads. Microsoft CTO Kevin Scott will be speaking at TechCrunch Disrupt in San Francisco from October 27th to October 29th.
