Inside Microsoft Ignite 2025: How Microsoft and NVIDIA Are Building a Full-Stack AI Platform

At Microsoft Ignite 2025 in San Francisco, Microsoft and NVIDIA used the conference floor to make a clear argument to cloud architects and IT leaders: the next phase of enterprise AI will be delivered as a tightly integrated stack, from GPUs and libraries through to cloud services, AI agents, and enterprise data platforms. Their joint announcements and technical sessions positioned Azure plus NVIDIA as a unified platform for building and operating AI agents, physical AI, and industrial digital twins at scale.

Across more than 40 sessions that highlighted NVIDIA technologies on Azure—15 of which are now available on demand for a limited time—the two companies detailed new GPU infrastructure, AI agent tooling, and database integrations aimed at making AI deployments more performant, governable, and easier to place where the data lives.

Ignite 2025: AI agents, physical AI, and digital twins take center stage

Ignite 2025 was framed around a single theme: accelerating success with AI. For Azure customers, that translated to practical guidance on how to move from pilots to production with agentic AI, physical AI systems, and digital twins.

NVIDIA and Microsoft used their presence to emphasize the breadth of their collaboration “from silicon to services.” The joint story cut across:

• GPU infrastructure on Azure for training, inference, simulation, and visualization.
• Agent frameworks and model services for enterprise-grade AI assistants.
• Support for digital twins and industrial workflows via NVIDIA Omniverse on Azure.
• Database integrations that bring retrieval‑augmented generation (RAG) directly to SQL Server 2025.

For cloud architects and infrastructure engineers, the message was that the individual pieces—GPUs, models, toolkits, and databases—are being delivered as cohesive building blocks in Azure, with NVIDIA’s platform elements embedded at multiple layers of the stack.

NVIDIA + Azure: a full-stack approach from silicon to services

One of the recurring themes across the Ignite content catalog was the degree of integration between Azure and the NVIDIA platform. Sessions focused on how to combine Azure’s global cloud footprint and hybrid capabilities with NVIDIA’s GPUs, SDKs, and model microservices.

More than 40 sessions showcased NVIDIA solutions and capabilities on Azure, pointing practitioners to:

• New generations of NVIDIA GPUs available in Azure virtual machine families.
• CUDA-accelerated libraries exposed through Azure-hosted services.
• NVIDIA microservices—branded as NVIDIA NIM—that package models such as Nemotron for use in enterprise AI workloads.
• Omniverse-based simulation and digital twin workflows operating on Azure infrastructure.

Fifteen of these sessions are now available to watch on demand for a limited time, giving Azure-focused teams a way to review architectural patterns, cost and energy efficiency considerations, and real-world usage examples at their own pace.

Blackwell on Azure NCv6: right-sizing converged AI and visual workloads

The most directly infrastructure-focused announcement was the public preview of the Azure NCv6 Series Virtual Machines powered by NVIDIA RTX PRO 6000 Blackwell GPUs. This marks a new expansion of NVIDIA’s Blackwell platform into Azure’s VM portfolio.

According to the companies, NCv6 with Blackwell is designed to “right-size” acceleration for converged AI and visual computing workloads. The targeted scenarios span:

• Digital twins and industrial simulation, including workflows built on NVIDIA Omniverse.
• Agentic AI workloads that require both compute and graphics capabilities.
• 3D rendering tasks and visualization pipelines.
• LLM inference and RAG on small-to-medium models.
• Scientific visualization and analysis.

For cloud architects, the positioning is that NCv6 offers a seamless upgrade path: existing GPU-based workloads can move to Blackwell-based VMs to gain performance and efficiency improvements without a wholesale re-architecture. The companies highlighted benefits in cost and energy efficiency in the related Ignite session “Power next-generation AI workloads with NVIDIA Blackwell on Azure,” alongside case studies showing how customers combine Azure orchestration with NVIDIA’s GPU stack.

Ignite also underscored that NVIDIA Omniverse libraries are now available on Microsoft Azure. Paired with Azure Local, this provides a consistent environment from edge to cloud for teams building and running industrial AI and simulation workloads. A joint session with NVIDIA, Microsoft, and Ansys—“Transform manufacturing with digital twins and real-time simulation”—explored how this stack can be used to optimize manufacturing processes and accelerate time to insight.

For hybrid and edge strategies, Microsoft’s “What’s new in Azure Local” session covered how NVIDIA RTX PRO 6000 GPUs integrate into local deployments, giving organizations options to run converged AI and visual computing workloads close to where data is generated while still aligning with the same platform patterns used in the public cloud.

Omniverse and digital twins: industrial workflows on Azure

Digital twins and “physical AI” were prominent topics across the joint content. NVIDIA Omniverse technology on Azure was featured as a way to create and operate high-fidelity digital representations of industrial systems, with Azure providing the underlying compute, storage, and management layers.

With Omniverse libraries now accessible on Azure, enterprises can:

• Develop and run digital twins that mirror complex physical environments, such as factories or supply chains.
• Use real-time simulation to test process changes and optimization strategies before applying them in the real world.
• Combine simulation workloads with agentic AI and 3D visualization on common GPU infrastructure.

By linking Omniverse and Azure Local, organizations gain a path to deploy these capabilities at the edge—for example, in on-premises industrial sites—while still managing them as part of a unified Azure-based AI and simulation strategy.

Agentic AI in the enterprise: Microsoft Agent 365 meets NVIDIA NeMo

On the application side of the stack, Microsoft and NVIDIA focused heavily on agentic AI: systems that can plan, reason, and act across tools and data sources. Their key announcement in this area was the integration of the NVIDIA NeMo Agent Toolkit with Microsoft Agent 365.

Microsoft Agent 365 is designed to bring AI agents into the core Microsoft 365 productivity apps—Outlook, Teams, Word, SharePoint, and others. With the NeMo Agent Toolkit integration, developers can:

• Create workplace agents that are tailored to specific organizational workflows.
• Build agents that comply with enterprise security and governance requirements.
• Deploy these agents across Microsoft 365 while leveraging NVIDIA’s tools for agent behavior and orchestration.

On the model side, Microsoft Foundry has made NVIDIA Nemotron models available as secure NVIDIA NIM microservices to power these enterprise agents and other digital AI use cases. NVIDIA Cosmos models are positioned to power physical AI workloads, complementing the agentic capabilities with models focused on interacting with and understanding the physical world.

The Nemotron models are described as supporting multimodal intelligence and multilingual reasoning, as well as math, coding, and additional capabilities that are relevant to enterprise agents. For teams building internal AI assistants, the combination of Agent 365, NeMo Agent Toolkit, and Nemotron-based NIM microservices offers a path to assemble and deploy agents without having to stitch together disparate model and orchestration components on their own.

NVIDIA’s Ignite session “From prompt to production: Scale agentic AI with NVIDIA and Azure” expanded on this, covering the broader enterprise strategy and walking through how components such as NVIDIA NIM microservices, the NeMo Agent Toolkit, CUDA libraries, and the Nemotron model family can be composed into production-grade agentic AI architectures.

SQL Server 2025 and Nemotron RAG: bringing AI to enterprise data

The data layer was another focus area, with Microsoft and NVIDIA framing new integrations as a way to build scalable AI directly on enterprise data without re-platforming core databases. The key announcement here: SQL Server 2025 can now connect to NVIDIA Nemotron RAG models deployed as NVIDIA NIM microservices.

For infrastructure and database teams, this integration is intended to simplify how retrieval‑augmented generation is applied to existing SQL Server data. By connecting SQL Server 2025 directly with Nemotron RAG models:

• Organizations can streamline AI deployment on top of their structured data.
• GPU-accelerated RAG workflows can run on Azure Cloud or Azure Local, avoiding CPU-bound bottlenecks.
• Data privacy and security controls stay anchored in the database platform rather than in ad hoc pipelines.

A key architectural promise is that AI can be “brought to the data” rather than forcing teams to move data into separate AI-specific stores. With GPU-accelerated RAG running wherever the data resides—on-premises or in the cloud—organizations can maintain full data sovereignty while still taking advantage of high-performance AI inference.

This approach is meant to remove the operational overhead of building and maintaining complex data pipelines just to feed AI systems. Instead, SQL Server 2025 serves as the authoritative data source, with NIM-hosted Nemotron RAG models providing the retrieval and generation layer on top.

How to go deeper: on-demand sessions and next steps for Azure teams

For Azure-focused practitioners who were not on-site—or who want to revisit the technical content—15 NVIDIA-related Ignite sessions are now available on demand for a limited time. These cover topics such as:

• Running next-generation AI workloads with NVIDIA Blackwell on Azure.
• Using Omniverse and Azure to build and operate digital twins and industrial simulations.
• Designing and scaling agentic AI with NVIDIA and Azure tools and microservices.
• Robotics simulation and infrastructure patterns for AI-heavy workloads.

The sessions collectively illustrate how NVIDIA’s hardware, libraries, and models are being surfaced as first-class capabilities within Azure services, from VMs and edge deployments to productivity tools and databases. For cloud architects, the practical takeaway is that the AI stack on Azure is becoming more vertically integrated—offering pre-aligned options rather than requiring custom integration at every layer.

Organizations evaluating or standardizing on Azure can use these sessions to inform decisions about GPU selection, hybrid architectures with Azure Local, strategies for deploying enterprise AI agents, and approaches to AI over existing SQL Server data estates.

More details and additional sessions can be found in the Microsoft Ignite catalog under NVIDIA-related content. The on-demand availability is time-limited, so teams interested in these integrations may want to prioritize viewing the sessions that align most closely with their near-term AI projects.