Azure expands Nvidia Rubin preview as confidential VMs remain in development
Microsoft and Nvidia are expanding Azure’s Vera Rubin virtual-machine preview, with systems already running in its data centers. Microsoft’s separate confidential-AI showcase covers four GPUs, while the announcements leave pricing, regions and general availability unspecified.
- Nvidia says Vera Rubin NVL72 systems are operating in Azure as the companies expand preview access.
- Microsoft describes a confidential-AI showcase on one four-GPU node; Rubin confidential virtual machines remain in development.
- The announcements give no Azure price or general-availability date; Nvidia says Oracle and Google already offer Rubin generally.
Microsoft and AI-chip supplier Nvidia are expanding access to virtual machines powered by Vera Rubin NVL72, Nvidia’s rack-scale AI computing platform, on Microsoft’s Azure cloud. Nvidia says the systems are already running in Azure data centers. Its November 19 Ignite announcement calls the offering a preview and does not identify a virtual-machine series, region, price or general-availability date.
For organizations handling medical records, financial information or proprietary models, access to new infrastructure is only part of the decision. Microsoft’s separate confidential-computing technology preview aims to restrict who can access data and models during processing. It starts with customer-controlled bare-metal instances and a four-GPU showcase. The platform’s security architecture spans a rack; the demonstration covers one node.
Operational hardware, unfinished customer terms
Vera Rubin NVL72 combines 72 Rubin graphics processors with 36 Vera central processors and their interconnects. In its November 17 technical announcement, Microsoft cites Nvidia’s description of a confidential-computing domain spanning that rack. Large models distribute computation across accelerators, so the links between chips also need protection.
This expansion follows months of deployment work. Microsoft said in its November 12 account that it became the first cloud to bring up a Vera Rubin NVL72 system for validation in March, then delivered its first production Vera Rubin systems in its data centers in August. Those are hardware milestones; neither establishes general customer availability.
Azure already has experience operating Nvidia’s preceding Blackwell systems at scale. Nvidia says in its announcement that Azure used 10,240 GPUs in GB200 NVL72 systems for a DeepSeek-V3 run in MLPerf Training v6.1, the largest scale in that round. That is evidence of the partners’ existing infrastructure work, rather than a measurement of Rubin’s Azure performance or cost.
