Microsoft deploys the first large-scale production clusters of NVIDIA GB300 NVL72 Blackwell Ultra GPUs and launches Azure AI Foundry as the central platform for enterprise AI. What does this mean for organisations serious about AI?
At NVIDIA GTC on 16 March 2026, Microsoft made a series of announcements that raise the bar for enterprise AI infrastructure once again. Most notably: Microsoft is the first hyperscaler to have built a large-scale production cluster with NVIDIA GB300 NVL72 Blackwell Ultra GPUs — more than 4,600 racks specifically configured for OpenAI workloads.
The NVIDIA GB300 NVL72 is the latest generation of data centre GPU. Each rack combines Grace CPUs, Blackwell Ultra GPUs and next-generation InfiniBand networking in an integrated, liquid-cooled unit. Compared to the previous generation, the GB300 delivers significantly higher compute per watt — directly reducing cost-per-token for Azure customers running large language model workloads.
Microsoft reported deploying hundreds of thousands of liquid-cooled NVIDIA Grace Blackwell GPUs across its global data centres within twelve months — an unprecedented pace of hardware deployment at this scale.
Alongside the hardware announcements, Microsoft positioned Azure AI Foundry (formerly Azure AI Studio) as the central platform for building, deploying and managing production-grade AI applications. Azure AI Foundry integrates access to more than a hundred AI models, fine-tuning capabilities, RAG pipelines and agent governance in a single platform.
The model catalogue has been expanded with GPT-image-1, OpenAI's most capable image generation and editing model. GPT-image-1 accepts images as input for editing and inpainting, and generates reliably readable text within images — a capability where earlier models consistently struggled.
The expansion of Azure GPU infrastructure has direct implications for the cost and speed of AI services consumed by Azure customers. Faster inference means shorter response times in production AI applications. Lower cost-per-token makes it economically viable to integrate AI into high-volume processes such as customer service, document processing or real-time data analysis.
GPT-image-1 opens new possibilities for visual applications: automated product photo editing, document digitisation with accurate text recognition, and brand-consistent image generation at scale. For e-commerce companies, marketing agencies or document-intensive sectors, these represent concrete productivity gains.
The Microsoft-NVIDIA partnership positions Azure as the leading cloud for AI workloads. For European organisations serious about deploying AI, it is relevant that Microsoft guarantees European data residency and complies with GDPR requirements, while simultaneously offering the most advanced GPU infrastructure available globally.
Want to explore how your organisation can leverage Azure AI Foundry for concrete AI applications? Contact Zarioh Digital Solutions.
