NVIDIA, Microsoft expand Azure-to-PC AI infrastructure and enterprise model tools
Enterprise demand for running artificial intelligence across cloud, on-premises systems and end-user devices is pushing technology suppliers to broaden deployment options and model customization tools. NVIDIA and Microsoft say they are extending their collaboration from Azure data centers to Windows PCs and industrial edge systems, with new enterprise offerings spanning infrastructure, software and research applications.
Highlights
- NVIDIA and Microsoft are expanding Azure access to NVIDIA Vera Rubin NVL72 systems and open models via Microsoft Foundry for broader enterprise AI deployment.
- In MLPerf Training v6.1, Azure scaled to 10,240 NVIDIA GPUs with GB200 NVL72 systems on DeepSeek-V3, while GB300 NVL72 led cloud inference submissions by over 10% on DeepSeek-R1.
- EY improved accuracy by 55% and reduced costs by 15 times on corporate purchasing record classification using NVIDIA Nemotron open models with Microsoft Foundry Programmed Training.
Broader enterprise AI rollout across Azure and local systems
As reported by NVIDIA Blog, citing NVIDIA Newsroom, NVIDIA and Microsoft are expanding access to Azure virtual machines powered by NVIDIA Vera Rubin NVL72 systems, while adding customer availability for NVIDIA open models and customization capabilities through Microsoft Foundry. The companies are also widening options for running AI workloads at the edge, where enterprises generate and use their data.Vera Rubin NVL72 systems are running in Azure data centers and are aimed at demanding reasoning workloads, agentic applications, and large-scale model training and inference. The companies are also working to extend confidential computing to bare-metal and rack-scale AI infrastructure, a step intended to protect sensitive models and regulated data during processing and to let customers verify computing environments before releasing data, models or keys.
NVIDIA and Microsoft say they are also collaborating on power efficiency through NVIDIA MaxLPS. In the latest MLPerf results cited by the companies, Azure scales to 10,240 NVIDIA GPUs with GB200 NVL72 systems on the DeepSeek-V3 model in MLPerf Training v6.1, while Azure's NVIDIA GB300 NVL72 systems lead cloud submissions on DeepSeek-R1 interactive throughput by more than 10% in MLPerf Inference v6.1.
Through Microsoft Foundry, the companies are expanding model and service options for enterprise AI agents that need to reason through tasks, retrieve company information and interpret business documents. Microsoft says its serverless Programmed Training offering gives developers control over post-training workflows, while Foundry Managed Compute adds support for deploying custom model weights through Microsoft's managed services.
EY uses Programmed Training in Microsoft Foundry to adapt NVIDIA Nemotron open models for classifying corporate purchasing records, improving accuracy on its evaluation dataset by 55% and lowering cost by 15 times compared with proprietary frontier models, according to the companies. The expanded Nemotron 3.5 lineup on Foundry Managed Compute includes Super for long-running agents and Lightning for faster-response applications, while other models cover retrieval, document parsing and content safety.
Earlier, we discussed S&P Global’s AI chip financing criteria as investors increasingly treat GPUs and related hardware separately from data center funding.