THE APEX TIMES
At Microsoft Ignite, the companies expanded access to NVIDIA-powered Azure systems, added more NVIDIA models and customization tools through Microsoft Foundry, and pushed new options for running AI at the edge and on Windows devices.
NVIDIA and Microsoft are extending their “full-stack” collaboration for artificial intelligence, moving beyond cloud infrastructure to cover AI model deployment on premises and on everyday Windows PCs. The expansion was outlined during Microsoft Ignite, as both companies emphasized that enterprises want to build, customize, and run intelligent applications across multiple environments, without losing control of data and compute.
A central part of the update is broader access to Azure virtual machines powered by NVIDIA Vera Rubin NVL72 systems. NVIDIA described the infrastructure as running in Azure data centers and designed for demanding reasoning and agentic workloads, as well as large-scale model training and inference. NVIDIA also said the partnership includes confidential computing enhancements that extend protections from cloud workloads into “bare-metal and rack-scale” AI infrastructure.
Confidential computing is the use of hardware-based safeguards to protect sensitive information while it is being processed. NVIDIA said its Confidential Computing and attestation capabilities help customers verify the computing environment before releasing sensitive models, data, or keys, and that the companies are working to extend those protections across what they described as the AI factory’s interconnected accelerator setup.
On the model-access side, Microsoft is expanding customer availability to NVIDIA open models and customization capabilities through Microsoft Foundry. NVIDIA tied the effort to Microsoft’s serverless Programmed Training offering, in which developers can control post-training workflows while Foundry manages the underlying infrastructure. NVIDIA said Foundry also offers Managed Compute support for deploying custom model weights, linking model customization to managed deployment services.
NVIDIA cited an example of enterprise customization: EY used Programmed Training in Microsoft Foundry to adapt NVIDIA Nemotron open models for classifying corporate purchasing records. NVIDIA said the adaptation improved accuracy on EY’s evaluation dataset by 55% and lowered cost by 15x compared with proprietary frontier models, framing it as evidence that organizations can tailor open models to specialized business tasks rather than relying only on general-purpose deployments.
The companies also expanded the Nemotron 3.5 model lineup available through Foundry Managed Compute, with different options aimed at different latency and runtime needs. NVIDIA described Nemotron 3.5 Super for long-running agents across coding, research, and business workflows, while Nemotron Lightning is positioned for fast responses and tool use. Additional related models include Nemotron Embed and Rerank for retrieving relevant information, Nemotron Parse for extracting structured data from documents, and Nemotron Content Safety for detecting harmful content. NVIDIA added that Nemotron 3.5 Bolt is designed for high-throughput, low-latency reasoning, and that Transcribe for speech recognition is coming soon.
Beyond cloud and model customization, NVIDIA said the partnership is broadening deployment options so that organizations can run AI “closer to their data.” It pointed to expanded Nemotron 3.5 support in Foundry Local for Azure Local environments, which NVIDIA said enables reasoning, coding, and tool-using agents. NVIDIA also said retrieval capabilities in this setting are intended to work across text, document images, charts, and tables.
For edge and on-premises deployments, Microsoft and NVIDIA said Azure Local will support NVIDIA RTX PRO 6000 Blackwell GPUs from hardware partners, and is expanding to NVIDIA Jetson Thor and Jetson Orin edge AI platforms aimed at industrial use cases such as machine vision, robotics, and autonomous operations. NVIDIA further said its physical AI models, including Cosmos, are being brought to Foundry Local, and that the companies will also demonstrate local large language model experiences on Windows PCs powered by NVIDIA RTX Spark.
NVIDIA also described additional connectivity options for users of Microsoft’s Cloud PCs. It said Direct Stream capabilities will be brought to GPU-accelerated Windows 365 Cloud PCs, targeting up to two 4K streams at 60 frames per second with low latency even over challenging networks, with a limited preview open for sign-ups and a public preview planned for early 2027. NVIDIA said Microsoft is using its own accelerated computing stack in parallel, including NVIDIA Dynamo inference software for post-training rollout inference and RTX PRO 6000 Blackwell GPUs for Bing search models.
A new report from Hunterbrook Media highlights concerns about AMD’s footprint in China, renewing attention on how advanced chipmakers navigate export controls and geopolitical risk.
A market report says a short seller has focused on alleged China ties tied to server-firmware updates, raising fresh questions about supply-chain visibility in the semiconductor and infrastructure stack.
A market report tied Meta’s current valuation to a roughly 21-times price-to-earnings multiple while flagging that shareholders may face a potentially large new cost, without detailing what it will be.
In a market commentary, analyst Josh Brown said Netflix has room to outperform Wall Street’s growth expectations, citing Netflix’s own 2026 guidance and a valuation framework tied to forward earnings.
In a debate that mixes star investors and corporate succession, Nvidia’s interim CEO Colette Kress discussed a higher forward growth target after Jensen Huang’s departure, while Cathie Wood questioned whether the interim executive is a deep technology expert.
A report circulating in markets suggested AMD’s chips were not positioned to replace Nvidia, prompting an AMD spokesperson to say the characterization misrepresents the company’s practices and relationships.
A sharp early gain for Anthropic after its public-market debut did not lift broader sentiment on Thursday, with U.S. indexes turning lower and AMD among chip-linked stocks that fell quickly.
A new comparison between Salesforce and Palantir pitches two very different growth stories, but emphasizes that the gap in share-price expectations could be the deciding factor for 2027.
A market-focused analysis highlighted how tightly Nvidia’s fortunes can track key buyers and, in a related twist, how one major Nvidia-related exposure can loom large in portfolio terms.
Traders appeared to react to renewed attention on how Broadcom’s AI infrastructure business intersects with Anthropic, according to a market report published Thursday.
Comments
Community guidelines. Comments may be removed for violations.
Loading comments...