NVIDIA and Microsoft Extend AI Acceleration to Azure, Edge Devices, and Windows PCs

NVIDIA and Microsoft are broadening their collaboration to enable enterprises to build and deploy AI applications across Azure data centers, edge systems, and Windows PCs. The partnership includes expanding access to Azure virtual machines powered by NVIDIA Vera Rubin NVL72 systems, integrating NVIDIA open models via Microsoft Foundry, and enhancing AI performance with NVIDIA Dynamo inference software and RTX PRO 6000 Blackwell GPUs for Bing. They are also advancing confidential computing and power efficiency through NVIDIA MaxLPS to secure sensitive data and improve performance per watt.
In November 2026, NVIDIA and Microsoft announced an expanded collaboration to integrate AI acceleration across Azure data centers, edge systems, and Windows PCs. The partnership centers on deploying NVIDIA Vera Rubin NVL72 systems in Azure virtual machines, enabling large-scale model training and inference with support for up to 10,240 GPUs in MLPerf Training v6.1. These systems, part of the Blackwell architecture, also delivered over 10% higher interactive throughput in MLPerf Inference v6.1. Microsoft is leveraging NVIDIA Dynamo inference software and RTX PRO 6000 Blackwell GPUs to accelerate Bing search models, while also introducing confidential computing protections across bare-metal and rack-scale infrastructure to secure sensitive data during processing. Power efficiency is being enhanced through NVIDIA MaxLPS, aiming to improve performance per watt in data centers. Enterprises can now customize and deploy AI models via Microsoft Foundry, using NVIDIA’s open Nemotron 3.5 lineup—such as Nemotron 3.5 Super for long-running agents and Nemotron Bolt for low-latency tasks—on both cloud and on-premises environments. EY successfully used Programmed Training to adapt Nemotron models for purchasing record classification, achieving 55% higher accuracy and 15x lower cost than proprietary alternatives. The collaboration extends to edge platforms like NVIDIA Jetson Thor and Orin, and to Windows PCs via RTX Spark, with Direct Stream capabilities coming to GPU-accelerated Windows 365 Cloud PCs in early 2027.
The companies’ move builds on prior codesign efforts and performance optimizations highlighted in MLPerf benchmarks, reflecting a strategic shift toward integrated, scalable AI infrastructure. NVIDIA seeks to dominate the AI chip market by embedding its hardware and software stack across Microsoft’s cloud and client ecosystems, while Microsoft aims to strengthen Azure’s competitive edge by offering end-to-end AI acceleration from cloud to device. The focus on confidential computing and power efficiency addresses growing enterprise concerns around data privacy and sustainability. Unresolved questions remain about the scalability of on-premises deployments, the long-term performance of edge AI in industrial settings, and how widespread adoption of local LLMs on Windows PCs will affect cloud usage patterns. The public preview of Direct Stream in early 2027 will be a key indicator of user demand and technical viability.
This collaboration could significantly reshape how enterprises deploy AI, shifting workloads from centralized cloud environments to distributed edge and endpoint systems. By enabling high-performance AI on Windows PCs and industrial edge devices, NVIDIA and Microsoft may empower organizations in manufacturing, healthcare, and finance to process sensitive data locally—potentially reducing latency and compliance risks. The integration of confidential computing may set a new standard for secure AI processing, particularly in regulated industries. A comparable precedent is the rise of on-device AI in smartphones during the 2010s, which reduced reliance on cloud services for tasks like voice recognition and image filtering. However, unlike mobile AI, this effort spans enterprise infrastructure, raising questions about interoperability, vendor lock-in, and long-term maintenance. The success of Foundry Local and Direct Stream could influence broader trends in hybrid cloud computing and decentralized AI. What remains uncertain is whether the performance gains will justify the complexity of managing AI across diverse hardware environments. Observers should watch for adoption rates in regulated sectors and how developers respond to the open model ecosystem.