NVIDIA and Microsoft extend Azure AI collaboration to on-premises devices

Executive Briefing
“NVIDIA and Microsoft announced an expansion of their AI collaboration across Azure...”
Key Takeaways
- 1.NVIDIA says Vera Rubin NVL72 systems are running in Azure data centers, and the companies are expanding access to Azure virtual machines powered by them.
- 2.Microsoft Foundry is adding options to customize and deploy models. Programmed Training gives developers control over post-training workflows, while Foundry Managed Compute supports deployment of custom model weights.
- 3.NVIDIA reports that EY used Programmed Training to adapt Nemotron models for classifying corporate purchasing records. On EY’s evaluation dataset, it reports 55% higher accuracy and 15-fold lower cost than proprietary frontier models.
- 4.The companies are expanding options for running AI closer to enterprise data. Azure Local will support Jetson Thor and Jetson Orin for industrial applications, while an Ignite demonstration is planned to show language models running directly on Windows PCs powered by NVIDIA RTX Spark.
- 5.For GPU-accelerated Windows 365 Cloud PCs, the companies are bringing Direct Stream capabilities that support up to two 4K streams at 60 frames per second. Sign-ups are open for a limited preview, with a public preview planned for early 2027.
NVIDIA and Microsoft announced an expansion of their AI collaboration across Azure infrastructure, on-premises systems and Windows devices. The initiatives are at different stages, ranging from systems already running to expanded support and demonstrations.
Product Features & Key Highlights
NVIDIA says Vera Rubin NVL72 systems are running in Azure data centers, and the companies are expanding access to Azure virtual machines powered by them.
Microsoft Foundry is adding options to customize and deploy models. Programmed Training gives developers control over post-training workflows, while Foundry Managed Compute supports deployment of custom model weights.
NVIDIA reports that EY used Programmed Training to adapt Nemotron models for classifying corporate purchasing records. On EY’s evaluation dataset, it reports 55% higher accuracy and 15-fold lower cost than proprietary frontier models.
The companies are expanding options for running AI closer to enterprise data. Azure Local will support Jetson Thor and Jetson Orin for industrial applications, while an Ignite demonstration is planned to show language models running directly on Windows PCs powered by NVIDIA RTX Spark.
For GPU-accelerated Windows 365 Cloud PCs, the companies are bringing Direct Stream capabilities that support up to two 4K streams at 60 frames per second. Sign-ups are open for a limited preview, with a public preview planned for early 2027.
Practical Assessment & Evaluation
Bringing models to Azure Local and industrial devices could let organizations process data where it is generated rather than routing every application through the cloud. The practical benefit will depend on supported hardware and each deployment.
EY’s result concerns a specific classification task; it does not establish that customized models will be more accurate or cheaper for other workloads.
Pricing, Availability & Next Steps
The announcement does not give broad-availability dates or pricing for every infrastructure and model offering described.
The supplied account does not provide enough detail about EY’s evaluation method for readers to independently assess its reported figures.







