News

NVIDIA and Microsoft expand AI stack across Azure, edge systems and Windows PCs

NVIDIA and Microsoft expanded their AI collaboration across Azure, Microsoft Foundry, edge hardware and Windows PCs, pairing a generally available service with previews and demonstrations.

D
Nov 19, 2026 · 2 min read

NVIDIA and Microsoft expanded their AI collaboration across Azure infrastructure, Microsoft Foundry, edge systems and Windows PCs on November 19. The companies paired a generally available model-deployment service with hardware previews and product demonstrations, extending model customization and deployment options across cloud and local environments.

On Azure, NVIDIA said Vera Rubin NVL72 systems are running in Azure data centers, but neither company’s announcement established general availability for Vera Rubin-powered Azure virtual machines. Microsoft described its confidential-computing work on Vera Rubin as a technology preview beginning with Confidential Bare Metal Instances. Confidential virtual machines, confidential containers in Azure Kubernetes Service and Microsoft Foundry support are listed as future development.

Microsoft’s Ignite demonstration was narrower than a full 72-GPU rack. The company said the single-node demonstration used four Rubin GPUs, vLLM and four-way tensor parallelism to run a planned 397-billion-parameter model workflow.

The companies are also widening model-development options in Microsoft Foundry. Programmed Training is in public preview, letting developers provide Python training recipes while Foundry runs them on serverless GPUs. Foundry Managed Compute, by contrast, is generally available for deploying open, partner and custom models on dedicated Microsoft-managed infrastructure. NVIDIA said Managed Compute supports an expanded Nemotron 3.5 lineup, while Nemotron 3.5 Bolt and Transcribe are coming soon.

NVIDIA cited EY as an early user of the training workflow, saying the firm improved accuracy by 55% on its own evaluation dataset and cut costs by 15 times after adapting Nemotron models. Those figures are vendor-reported customer results and were not independently reproduced in the supplied research.

At the edge, Microsoft put Azure Local version 2610 for small-form-factor hardware and support for NVIDIA’s Jetson AGX Thor Developer Kit into public preview. Microsoft said its Danfoss robot example, which used Foundry Local on Jetson Thor, was a technology demonstration in a Microsoft test lab rather than a deployment on a Danfoss factory floor.

For PCs, NVIDIA presented local large-language-model experiences on RTX Spark Windows machines as Ignite showcases, not a new generally available Windows service. It also opened sign-ups for a limited preview of Direct Stream on GPU-accelerated Windows 365 Cloud PCs and said a public preview is planned for early 2027. Microsoft described the GPU-enabled Cloud PC capability as coming in preview early next year. The companies said the preview technology can support as many as two 4K displays at 60 frames per second with low latency; that remains a vendor claim for a product that has not reached public preview.

More news