Skip to main content
NVIDIA and Microsoft logos on a dark background, symbolizing their expanded Blackwell accelerated computing partnership for W

Editorial illustration for NVIDIA and Microsoft Expand Blackwell Accelerated Computing to Windows PCs

NVIDIA Blackwell Comes to Windows PCs With Microsoft

• 4 min read

Microsoft put NVIDIA's Vera Rubin NVL72 systems into live Azure data centers this week, timed to announcements at Microsoft Ignite. The two companies are also pushing NVIDIA's reach past the data center and onto Windows PCs, part of what they're calling a full-stack push that stretches from cloud infrastructure down to edge devices where businesses actually generate their data.

The scope goes beyond raw hardware. Microsoft Foundry is getting NVIDIA's open models and customization tools, giving customers a way to build and tune applications without leaving Microsoft's platform. Microsoft is turning NVIDIA's own technology back on its products, too, running Dynamo inference software to speed up post-training work and using RTX PRO 6000 Blackwell GPUs to power Bing's search models.

Security is part of the pitch. NVIDIA and Microsoft are extending confidential computing to bare-metal and rack-scale AI infrastructure, aimed at companies handling proprietary or regulated data that need to verify their computing environment before exposing sensitive models or keys. There's also a power angle: NVIDIA's MaxLPS technology is meant to squeeze more performance out of the electricity data centers already have, a problem that's only grown as AI workloads scale up.

NVIDIA and Microsoft are expanding deployment options beyond the cloud, helping organizations run models and agents closer to their data while maintaining control over their infrastructure and operations.

Why this matters

For developers and founders building on Azure, this is a signal that the Blackwell-to-Rubin pipeline isn't theoretical anymore, it's showing up in procurement decisions and MLPerf benchmarks you can actually check. The 10,240-GPU DeepSeek-V3 run matters less as a vanity number and more as proof that NVIDIA and Microsoft are willing to put real infrastructure behind the "codesign" language they keep using. If you're training large models, that scale claim is worth testing against your own workloads before you take it at face value.

The Windows PC angle is the quieter story here: pushing accelerated computing down to local machines through Foundry and open models changes the calculus for teams who've been cloud-only by default. Researchers should watch whether "customization capabilities" actually means usable fine-tuning tools or just more marketing around existing APIs. We'd treat Ignite announcements like this as a starting point for due diligence, not a reason to switch stacks.

The real test comes when independent benchmarks on Rubin NVL72 systems start circulating outside NVIDIA's own MLPerf submissions.

Common Questions Answered

What is the Vera Rubin NVL72 system and how is Microsoft deploying it?

The Vera Rubin NVL72 is NVIDIA's accelerated computing system that Microsoft has deployed into live Azure data centers this week. This deployment was timed to announcements at Microsoft Ignite and represents a key step in expanding NVIDIA's Blackwell accelerated computing capabilities to enterprise cloud infrastructure.

How does the NVIDIA and Microsoft full-stack push extend beyond data centers?

Beyond deploying systems in Azure data centers, NVIDIA and Microsoft are expanding Blackwell accelerated computing to Windows PCs, creating a full-stack approach that stretches from cloud infrastructure down to edge devices where businesses generate their data. This strategy allows organizations to run models and agents closer to their data while maintaining control over their infrastructure and operations.

What role does Microsoft Foundry play in the NVIDIA partnership?

Microsoft Foundry is receiving NVIDIA's open models and customization tools as part of the expanded partnership. This integration enables developers and organizations to leverage NVIDIA's accelerated computing capabilities with greater flexibility and customization options within the Microsoft ecosystem.

Why is the 10,240-GPU DeepSeek-V3 run significant for this partnership?

The 10,240-GPU DeepSeek-V3 run serves as proof that NVIDIA and Microsoft are willing to put real infrastructure behind their 'codesign' language and demonstrates the practical viability of their Blackwell-to-Rubin pipeline. This scale achievement matters less as a vanity metric and more as evidence that the partnership's infrastructure claims are backed by actual procurement decisions and verifiable performance benchmarks.

What does the expansion to Windows PCs mean for developers building on Azure?

For developers and founders building on Azure, the expansion to Windows PCs signals that the Blackwell accelerated computing pipeline is no longer theoretical but is showing up in real procurement decisions and MLPerf benchmarks. This gives developers confidence that they can test and deploy large model training at scale using the infrastructure that NVIDIA and Microsoft are actively investing in.

LIVE01:21Apple seeks engineers to build AI agent that browses web and writes code