As OpenAI bets big on data centers, Satya Nadella says Microsoft already runs them
The announcement comes just days after OpenAI, Microsoft’s close partner and occasional rival, struck high-profile data center deals with Nvidia and AMD.
ADVERTISEMENT
Microsoft CEO Satya Nadella unveiled the company’s first large-scale AI system — what Nvidia calls an AI “factory” and said it would be the “first of many” deployed across Microsoft Azure’s global data centers. The systems are designed to power workloads for OpenAI.
Each installation consists of more than 4,600 Nvidia GB300 rack computers equipped with the new Blackwell Ultra GPUs and linked through Nvidia’s high-speed InfiniBand networking technology. Nvidia cornered the market on InfiniBand in 2019 when it acquired Mellanox for $6.9 billion, a move that has since proved strategic as demand for AI infrastructure has surged.
Microsoft said it plans to roll out “hundreds of thousands of Blackwell Ultra GPUs” worldwide as it builds these massive clusters. The company emphasized that the scale of the systems makes them capable of training and running the next wave of AI models with “hundreds of trillions of parameters.”
The timing is significant. The announcement comes just days after OpenAI, Microsoft’s close partner and occasional rival, struck high-profile data center deals with Nvidia and AMD. OpenAI has already committed an estimated $1 trillion to building out its own data center footprint, with CEO Sam Altman promising even more to come.
By contrast, Microsoft highlighted that it already operates more than 300 data centers in 34 countries, and said Azure is “uniquely positioned” to “meet the demands of frontier AI today.”
