Nvidia and Microsoft are widening their partnership at Microsoft Ignite, from Azure data centers to Windows PCs. The headline item is Azure virtual machines running Nvidia Vera Rubin NVL72 systems, now live in Azure data centers. The Nvidia Microsoft Azure expansion also covers open models in Microsoft Foundry, edge hardware and scientific research tools.
Microsoft is also using Nvidia technology inside its own AI work. It runs Nvidia Dynamo inference software for post-training and RTX PRO 6000 Blackwell GPUs for Bing search models.
Vera Rubin arrives in Azure
The racks are meant for reasoning models, AI agents, and big training and inference jobs. A second project pushes confidential computing down to bare-metal servers and whole racks. With attestation, a customer can check that the environment is trustworthy before handing over a model, a dataset or encryption keys.
A separate effort uses Nvidia MaxLPS to get more work out of the electricity a data center already draws.
Benchmarks give some backing. Azure’s MLPerf Training v6.1 entry used 10,240 Nvidia GPUs in GB200 NVL72 systems on DeepSeek-V3, the biggest in its round. On MLPerf Inference v6.1, Azure’s GB300 NVL72 machines beat other cloud entries on DeepSeek-R1 interactive throughput by over 10%, per Signal65.
Open models and custom training in Foundry
Developers who want to tune a model after pre-training can use Microsoft’s serverless Programmed Training and leave the servers to Foundry. Managed Compute can now host custom weights, so a tuned model goes straight to deployment.
EY is one early user. The firm tuned Nvidia Nemotron open models to sort corporate purchasing records. On its test set, accuracy improved by 55% and cost dropped 15x against proprietary frontier models.
The Nemotron 3.5 lineup on Foundry Managed Compute covers several needs:
- Nemotron 3.5 Super for long-running agents doing coding, research and business tasks
- Lightning for applications that need fast responses and tool use
- Nemotron Embed and Rerank for retrieval, Parse for pulling structure from documents, and Content Safety for detecting harmful content
Nemotron 3.5 Bolt, built for high-throughput, low-latency reasoning, and Transcribe for speech recognition are coming soon.
AI closer to company data
The Nvidia Microsoft Azure expansion reaches beyond the cloud.
Some businesses want to keep data on site. Foundry Local now supports Nemotron 3.5 on Azure Local for reasoning, coding and tool-calling agents, and retrieval models can search text, scanned documents, charts and tables.
Hardware partners will be able to ship Azure Local with RTX PRO 6000 Blackwell GPUs. Jetson Thor and Jetson Orin edge boards join too, for factory vision, robots and autonomous machines. Nvidia’s Cosmos models for physical AI are headed to Foundry Local as well.
Windows PCs and Cloud PCs
At Ignite, the pair will demo large language models running locally on Windows PCs with Nvidia RTX Spark.
GPU-backed Windows 365 Cloud PCs gain Direct Stream, which can push two 4K streams at 60 frames per second with low lag, even on poor connections. Sign-ups for a limited preview are open. A public preview is planned for early 2027.
Science and Microsoft’s own models
The Nvidia BioNeMo Agent Toolkit and ALCHEMI Toolkit are joining the Microsoft Discovery catalog, covering life sciences, drug discovery, and materials and chemical science. With BioNeMo, a scientist can describe a computational biology task in plain language and let agents line up the tools and models. The first integration runs RFD4-Proteina, developed with the University of Washington’s Institute for Protein Design, to propose protein structures and amino acid sequences that researchers then test in the lab.
Microsoft AI relies on Dynamo for rollout inference in reinforcement learning, the stage that produces the experience data used to improve a model. Dynamo spreads that load over a big GPU fleet, handles incoming requests and model updates, and recovers when something fails.
What to watch after Ignite
The Nvidia Microsoft Azure expansion rolls out in stages.
Three dates matter. Nemotron 3.5 Bolt and Transcribe are marked as coming soon, with no date given. The Windows 365 Direct Stream public preview is planned for early 2027. Customers who want the Vera Rubin VMs can already look to Azure.
Nvidia will show a GB300 NVL72 rack and RTX Spark devices at its Ignite booth, alongside sessions and hands-on labs.
Source: NVIDIA Blog