The development

What happened

NVIDIA’s 19 November Ignite announcement describes expanded Microsoft Foundry access to its open AI models and customization tools, alongside Azure and local-deployment work. The concrete AI connection is model training, inference and tool-using agents across managed cloud services and customer infrastructure. The package mixes available capabilities with previews, demonstrations and future support: Vera Rubin systems are running in Azure data centers under a preview, some Nemotron models remain forthcoming, and several local-hardware and Cloud PC integrations are not presented as generally available. This is a partner-reported integration update, not an independent performance or confidentiality validation.

Key facts

  • NVIDIA says Vera Rubin NVL72 systems are running in Azure data centers; the post calls the Azure offering a preview. Rack-scale confidential-computing protections and power-efficiency work are described as ongoing collaboration.

  • Foundry Managed Compute adds custom-weight deployment and expanded Nemotron 3.5 support, including Super and Lightning. Related models cover retrieval, reranking, document extraction and content-safety checks.

  • Nemotron 3.5 Bolt and the Transcribe speech-recognition model are described as coming soon. The announcement does not make the entire lineup generally available at once.

  • Foundry’s serverless Programmed Training gives developers post-training control while Microsoft manages infrastructure. NVIDIA describes EY adapting Nemotron for purchasing-record classification; this is a reported customer example, not an independent trial.

  • The partners describe expanded Nemotron support through Foundry Local for Azure Local, planned RTX PRO 6000 Blackwell support, and Azure Local expansion to Jetson Thor and Orin. These local integrations have differing delivery states.

  • For GPU-accelerated Windows 365 Cloud PCs, Direct Stream supports up to two 4K streams at 60fps. NVIDIA says limited-preview sign-ups are open, with a public preview planned for early 2027.

The context

Why it matters

TideReach assessment: the useful distinction is what teams can deploy now versus what they can only preview or plan around. Managed customization and local inference serve different data and operating requirements; a broad partnership announcement does not settle workload compatibility or confidentiality guarantees. Buyers should check each named service’s actual availability and validate it on their own tasks.