NVIDIA and Microsoft are deepening their strategic partnership, unveiling a series of new integrations and infrastructure advancements designed to power the next generation of AI systems. The collaboration, highlighted at the Microsoft Ignite conference, centers on the deployment of cutting-edge technologies for AI inference, cybersecurity, and physical AI, with a focus on scalability, performance, and enterprise readiness. At the heart of the effort is the Microsoft Fairwater AI superfactory in Wisconsin, now connected to a new data center in Atlanta, Georgia. This massive-scale infrastructure will host hundreds of thousands of NVIDIA Blackwell GPUs for large-scale model training. Microsoft is also deploying over 100,000 Blackwell Ultra GPUs in NVIDIA GB300 NVL72 systems across the globe, specifically for high-performance inference workloads. To support this vast network, Microsoft is implementing next-generation NVIDIA Spectrum-X Ethernet switches in the Fairwater facility, ensuring the low-latency, high-bandwidth connectivity required to run the most demanding AI models, including those used by OpenAI and Microsoft’s own AI Superintelligence Team. New Azure NCv6 Series virtual machines, now in public preview, are powered by the NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. These VMs offer flexible, right-sized acceleration for a wide range of workloads, including multimodal agentic AI, industrial digitalization using NVIDIA Omniverse, scientific simulation, and real-time visual computing. The integration extends to Azure Local, enabling secure, sovereign AI deployments at the edge and on-premises, bringing low-latency AI to factory floors and other critical environments. A key component of the partnership is the creation of a fungible AI fleet— a dynamic, continuously optimized infrastructure that can efficiently handle any AI workload across multiple NVIDIA architectures. Through full-stack software optimizations, including NVIDIA TensorRT-LLM and the NVIDIA DGX Cloud Benchmarking suite, the collaboration has achieved over 90% of the performance of the reference architecture, making AI more cost-effective and accessible. This has already led to a dramatic 90% reduction in the cost of popular GPT models on Azure over two years. The integration extends to Microsoft 365 Copilot, where the latest models from the Microsoft AI Superintelligence Team—such as MAI-1-preview, MAI-Voice-1, and MAI-Image-1—are now accelerated on the NVIDIA Blackwell platform, delivering richer, multimodal experiences across Bing Image Creator and Copilot. In the enterprise, NVIDIA and Microsoft are bringing AI directly into core business systems. NVIDIA Nemotron models and NIM microservices are now integrated with Microsoft SQL Server 2025, enabling secure, GPU-optimized retrieval-augmented generation for enterprise data, whether in the cloud or on-premises. The partnership also advances the future of agentic AI. The NVIDIA NeMo Agent Toolkit now works with Microsoft Agent 365, allowing developers to build, deploy, and onboard compliant AI agents within Microsoft 365 apps like Outlook, Teams, Word, and SharePoint. Microsoft Foundry now offers NVIDIA Nemotron for digital AI and NVIDIA Cosmos for physical AI, enabling enterprise-grade applications in multilingual reasoning, coding, math, and robotics. Cybersecurity is another focus area. The two companies are developing adversarial learning models using the NVIDIA Dynamo-Triton framework and TensorRT, achieving a 160x speedup over CPU-based methods for real-time threat detection. On the physical AI front, NVIDIA Omniverse libraries on Azure are enabling end-to-end digital transformation in manufacturing and industrial design. Developers are using tools like Synopsys, Sight Machine, and SymphonyAI to streamline workflows. Robotics platforms such as Hexagon’s AEON and Wandelbots NOVA are built on NVIDIA’s full robotics stack, leveraging Isaac Sim and Isaac Lab for simulation and real-world deployment. A standardized OpenUSD approach ensures seamless interoperability across 3D workflows, making digital twins and simulation more accessible in the cloud. The collaboration also expands to include Anthropic, with a new partnership focused on optimizing Anthropic’s models for performance, efficiency, and cost, as well as tailoring future NVIDIA hardware for Anthropic’s workloads. Together, these advancements underscore a unified vision: to deliver a scalable, secure, and intelligent AI infrastructure that powers innovation across industries.