Microsoft Azure Unleashes World’s First NVIDIA GB300 NVL72 Supercomputing Cluster for OpenAI, Setting Fresh Standard for AI Compute

Microsoft and NVIDIA Set a New Standard for AI Compute

On October 9, 2025, Microsoft Azure achieved a historic leap in artificial intelligence infrastructure, announcing the launch of the world’s first production-scale supercomputing cluster featuring NVIDIA’s latest GB300 NVL72 systems—purpose-built for the most demanding OpenAI workloads. This achievement marks a bold new era in AI, empowering both Microsoft and OpenAI to break through limitations in large model training, inference, and deployment.

The Announcement: Vision and Partnership

In a blog post by Rani Borkar (President, Azure Hardware Systems and Infrastructure) and Nidhi Chappell (CVP, Azure AI Infrastructure), Microsoft made official the deployment of more than 4,600 NVIDIA GB300 NVL72 GPUs, connected via the new NVIDIA Quantum-X800 InfiniBand network. This is only the beginning of what Microsoft envisions as hundreds of thousands of Blackwell Ultra GPUs deployed across its AI datacenters worldwide.

---

A Closer Look: The Cutting-Edge GB300 NVL72 Cluster

The heart of this breakthrough is the NVIDIA GB300 NVL72—the latest marvel in GPU technology. A single rack integrates 72 Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs, featuring:

• 37 TB of fast memory per VM

• 1.44 exaflops (FP4) of performance per rack

• 130 terabytes per second of NVLink bandwidth within each rack

• 800 gigabits per second per GPU cross-rack networking bandwidth

This tight integration allows the entire rack to function as a single, unified accelerator—ideal for massive AI training and large language model inference on a scale never before possible.

Networking, Cooling, and Software: Engineering for AI at Scale

Microsoft’s new cluster isn’t just about GPUs. Every layer of the data center stack has been reengineered:

• Memory and Bandwidth: Using NVIDIA’s NVLink and NVSwitch within racks supports ultra-large models across tens of thousands of GPUs with minimal latency.

• Next-Gen Networking: NVIDIA’s Quantum-X800 InfiniBand platform scales performance across the global cluster, allowing seamless communication between thousands of GPUs working in concert.

• Advanced Cooling: Standalone heat exchangers and innovative facility cooling support the power and thermal demands of densely packed GPU racks.

• Software Innovation: Azure’s orchestration, scheduling, and storage stacks have been rebuilt to drive unprecedented efficiency and reliability at supercomputing scale.

---

The Blackwell Revolution: Why the GB300 Matters

The Blackwell Ultra GPU, at the core of these racks, is built on an advanced architecture that merges the latest in chip design, memory access, and multi-processor integration. Compared with previous generations, the GB300 delivers:

• Up to 30x faster real-time inference for trillion-parameter models

• 4x faster training for large language models versus NVIDIA H100

• Greater than 25x efficiency improvements in energy use, supporting the sustainability goals of Microsoft’s datacenter investments

The new ND GB300 v6 Virtual Machines (VMs), now available on Azure, will become the backbone for organizations pushing the limits of AI across industries—from healthcare and finance to scientific discovery and advanced robotics.

see also: Microsoft and OpenAI Sign $135 Billion Agreement: Ushering in a New Era for AI Collaboration

---

What Does This Mean for AI?

This launch is about more than new hardware. Microsoft Azure’s GB300 NVL72 cluster is the first system ready to support training and deploying foundation models at scales previously seen as unattainable. The platform enables:

• Training larger, more sophisticated AI models in weeks instead of months

• Running inference tasks for hundreds of trillions of model parameters

• Supporting novel AI workflows such as agentic AI, multimodal generative models, and next-generation reasoning systems

OpenAI stands as the flagship customer for this infrastructure, using the GB300 cluster to train its next wave of multitrillion-parameter models—fueling Copilot, ChatGPT-Next, and as-yet-unannounced frontier AI projects.

---

The New AI Infrastructure Race

Microsoft and NVIDIA’s latest deployment exemplifies the emerging “AI infrastructure race,” where the ability to rapidly scale specialized compute—rather than simply software or algorithms—defines leadership.
Other cloud giants, including Google and Amazon, are expected to respond with their own large-scale deployments and strategic AI hardware partnerships.

NVIDIA’s Blackwell platform continues to be the backbone of the industry, ensuring that AI’s most powerful capabilities are accessible to those with the vision, resources, and partnerships needed for supercomputing at planetary scale.

---

The AI Future

Microsoft is already preparing to ramp up GB300 worldwide deployments, with its ND GB300 v6 VMs setting a new standard for AI infrastructure. The coming months will bring updates on expanded global rollouts, new features for Azure customers, and further demonstrations of unprecedented AI performance.

Stay up-to-date at the Azure Infrastructure Blog tech community. See the full NVIDIA press release here.

---

Important Highlights

• Microsoft Azure is the first cloud to deliver a production-scale NVIDIA GB300 NVL72 cluster, revolutionizing training and deployment for frontier AI.

• The collaboration with NVIDIA and strategic partnerships with “neocloud” providers cement Microsoft’s lead in the AI infrastructure race.

• This supercomputing leap will empower not just OpenAI, but customers across every sector, setting the pace for global AI innovation.

For further details and ongoing coverage, visit msftnewsnow.com for daily updates on Microsoft’s AI advancements and cloud computing leadership.