Nvidia and Microsoft to Redefine Data Centre Supercomputers

As hyperscale cloud providers race to build the next generation of AI-ready data centres, Microsoft Azure and Nvidia are setting a new benchmark for infrastructure performance and scale. 

Their latest collaboration represents a fusion of cutting-edge hardware and cloud engineering designed to meet the most demanding AI workloads ever seen in the data centre industry.

Microsoft Azure and Nvidia have partnered to create a new era of supercomputer clusters. The debut, using Nvidia’s GB300 NVL72 hardware, will be the industry’s first supercomputing-scale production cluster. The move by the two companies is to support OpenAI’s most demanding AI inference workloads. 

The cluster features more than 4,600 Nvidia Blackwell Ultra GPUs which are all connected by the Nvidia Quantum-X800 InfiniBand networking platform. 

These are all running at an extremely fast pace and require a significant amount of memory. To withstand these demands, Microsoft Azure utilises its specialised engineering techniques to improve both memory and networking. 

Doing so allows the supercomputers to manage great amounts of computing required to run advanced AI models and act like models and agentic AI systems.

Nidhi Chappell, corporate vice president of Microsoft Azure AI Infrastructure, says:

“Delivering the industry’s first at-scale Nvidia GB300 NVL72 production cluster for frontier AI is an achievement that goes beyond powerful silicon — it reflects Microsoft Azure and Nvidia’s shared commitment to optimising all parts of the modern AI data centre. 

“Our collaboration helps ensure customers like OpenAI can deploy next-generation infrastructure at unprecedented scale and speed.”

The relationship between Nvidia and Microsoft Azure has been nurtured over many years and has created the foundation which has made space for the creation of the supercomputer possible. The two companies have worked together to create purpose-built digital infrastructure for the most demanding AI workloads.

What does the Nvidia GB300 look like?

The new Azure AI computers bring together a wealth of advanced technological dynamics, all of which play their part to create a cutting-edge supercomputer. 

The main components are the 72 GPUs and 36 CPUs, which combine and are cooled by the liquid-cooled, rack-scale Nvidia GB300 NVL72 system. 

Keeping information stored are the 37 terabytes of fast memory and 1.44 exaflops of FP4 Tensor Core performance per VM. This provides a substantial amount of memory space for models, agentic AI systems and complex multimodal generative AI.

Built to train and run advanced AI extremely quickly, Nvidia Blackwell Ultra is assisted by the full-stack Nvidia AI Platform. The tools and formats from Nvidia, like the NVFP4 and Nvidia Dynamo have been developed to boost training and AI performance. 

Recent tests showed the MLPerf Inference v5.1 benchmarks, Nvidia GB300 NVL72 systems produced recording-setting performance using the NVFP4 with results being up to 5 times faster than previous systems when running large AI models like Deep-Seek-R1 and Llama 3.1.

“Microsoft Azure’s launch of the Nvidia GB300 NVL72 supercluster is an exciting step in the advancement of frontier AI,” says Ian Buck, Vice President of Hyperscale and High-performance Computing at Nvidia. “This co-engineered system delivers the world’s first at-scale GB300 production cluster, providing the supercomputing engine needed for OpenAI to serve multitrillion-parameter models. This sets the definitive new standard for accelerated computing.” 

Connecting the Supercomputer

As there are over 4,600 Blackwell Ultra GPUs in a single cohesive supercomputer, connection is crucial. This is where the two-tiered Nvidia networking architecture is utilised to support Microsoft Azure’s cluster. The two-tiered supports both scale-up performance inside the rack and the scale-out performance for the entire cluster.

The rack becomes a unified accelerator with a shared memory pool but this can only happen with the fifth-generation Nvidia NVLink Switch fabric which supplies 130 TB/s of direct, all-to-all bandwidth between the 72 Blackwell Ultra GPUs.

Connecting all of the AI systems together into a giant supercomputer, Microsoft Azure uses Nvidia Quantum-X800 network InfiniBand platform, technology which is specifically designed to handle huge AI models with trillions of parameters. This works by giving each GPU incredibly fast connections, 800 gigabits per second, so all 4,608 GPUscan talk to other efficiently.

For data centre operators, this collaboration offers a glimpse into what’s next for large-scale compute infrastructure. Liquid cooling, ultra-high bandwidth networking and advanced GPU-to-GPU interconnects are no longer niche capabilities – they are fast becoming the baseline for AI-era facilities.

The partnership between Nvidia and Microsoft Azure underscores how the world’s leading technology firms are reimagining the physical and digital layers of the data centre. As AI models grow ever larger, supercomputing-grade clusters like this one will define the next decade of hyperscale design and deployment – shaping not just cloud performance, but the very architecture of tomorrow’s intelligent data centres.