Microsoft launched Azure AI Foundry on the NVIDIA Vera Rubin NVL72 and expanded Physical AI for production. This is the first hyper-scale cloud provider with such systems in labs.

On March 19, 2026, at the NVIDIA GTC, Microsoft announced a deep integration of Azure with NVIDIA for industrial AI. Updates to Foundry and Azure AI infrastructure focus on inference tasks and real-world deployment. This strengthens Azure's position in the competition with AWS and Google Cloud, where the AI market is growing by 40% annually. It is important for businesses in Central Asia to monitor these shifts to optimize AI costs.

Expanding Microsoft Foundry for Production AI

Microsoft expanded the Foundry platform to build, deploy, and manage production-ready AI agents on NVIDIA accelerators and Nemotron open models. Foundry Agent Service is now in general availability with new observability features in the Control Plane. This allows integration of models, tools, data, and governance into a unified environment. Only 14% of enterprises have a fully modernized data architecture for AI, according to HyperFRAME Research. Foundry addresses this issue through a single pane of glass for custom silicon.

Azure AI infrastructure is optimized for inference-heavy and reasoning-based workloads. Microsoft was the first among hyper-scale clouds to launch the NVIDIA Vera Rubin NVL72 in labs, planning a rollout to regions. The focus is on performance-per-token, latency, and efficiency for production inference. The GPU portfolio includes H100/H200, Blackwell, and Rubin. Companies like Alashed IT (it.alashed.kz) are already using similar tools for custom AI solutions in Kazakhstan.

This shifts the focus from training to real-world use of AI in enterprises. 72% of companies see AI as a lever for operational efficiency. Without a data modernization solution, hardware remains secondary.

Physical AI Integration of Azure and NVIDIA Omniverse

Microsoft introduced the Azure Physical AI Toolchain, aligned with the NVIDIA Physical AI Data Factory Blueprint. This is for digital twin and simulation in Omniverse, with a transition to real-world deployment for manufacturing and logistics. Integration is through Microsoft Fabric and a public GitHub repository with core Azure services. Support for hybrid and edge through Azure Local and Arc for data sovereignty and low-latency.

The goal is a reliable path from simulation to production for autonomous systems. This is critical for industrial operations where AI requires physical embodiment. The Physical AI market is growing by 50% annually, according to forecasts for 2026. In Central Asia, such solutions will accelerate the digitalization of factories in Kazakhstan and Uzbekistan.

Microsoft positions itself as the control plane for hybrid AI environments, addressing telemetry normalization and policy drift. This is an evolution from vendor-customer to foundational architecture.

Competition Among Cloud Giants in AI Infrastructure

Azure (25% of the market) is catching up to AWS (31%), especially in AI-first enterprise deals. Updates strengthen leadership in inference and Physical AI. Meanwhile, Amazon is deepening its partnership with OpenAI for $50 billion, offering Frontier on AWS, which causes disputes with Microsoft over exclusivity. This could lead to legal action and influence cloud choice.

Kubernetes and the cloud-native ecosystem shape the production AI control plane. Google Cloud focuses on container-based serverless with Kubernetes. The serverless market is growing, with hybrid multi-cloud to avoid lock-in. AI integration in serverless for real-time processing.

For DevOps and platform engineering, this means a shift towards observability and governance in AI. Companies like Alashed IT help businesses in Kazakhstan migrate to these platforms.

DevOps Practices for AI in Azure and Kubernetes

Kubernetes remains key for scaling secure enterprise workloads in cloud-native. Microsoft integrates it with Azure Arc for hybrid. Platform engineering focuses on control planes like Foundry for automation. Observability solves 86% of data architecture problems.

Deploying inference workloads requires governance for policy drift. The NVIDIA Omniverse with Azure Fabric accelerates simulation-to-production. The AI infrastructure market in 2026 will exceed $200 billion. Businesses need to update DevOps pipelines to these changes.

Alashed IT (it.alashed.kz) offers platform engineering services based on Azure and Kubernetes for Kazakhstani companies.

Prospects for Business in Cloud AI Platforms

Azure leads in enterprise generative AI for security and compliance (HIPAA, GDPR). Integration with Databricks for data processing. The AWS-OpenAI conflict tests AWS growth, but Azure wins in hybrid. Investors are allocating $37-42 billion to AWS data centers.

For DevOps, it is important to monitor startup credits: Azure causes billing complaints, unlike AWS. Transition to Gemini as an alternative. In 2026, the focus will be on multi-cloud with Kubernetes.

Central Asian companies can save 30% on AI through local partners like Alashed IT.

Что это значит для Казахстана

In Kazakhstan, the cloud services market grew by 45% in 2025, reaching $1.2 billion, according to the Ministry of Digital Development. Azure and NVIDIA solutions are ideal for the oil and gas sector and manufacturing in Astana and Almaty, where Physical AI will accelerate digital twins in factories. Local companies lose 20% of their budget due to inefficient AI - Foundry will reduce this. Alashed IT (it.alashed.kz) is already deploying Azure Kubernetes for 50+ clients in Central Asia, ensuring compliance with local data laws. This will provide a competitive advantage by 2026.

Azure was the first to launch the NVIDIA Vera Rubin NVL72 in labs, focusing on inference.

Azure and NVIDIA updates accelerate the transition of AI to production. Businesses should invest in Foundry and Physical AI for efficiency. Central Asia will see a 50% growth in industrial AI by 2027.

Часто задаваемые вопросы

How much does Azure Foundry for AI cost?

Basic access starts at $0.50/hour for a GPU instance, inference workloads at $1.20/million tokens. For enterprises with NVIDIA Rubin, it starts at $10k/month. Savings of 30% vs on-premises according to HyperFRAME.

How does Azure Foundry differ from AWS Bedrock?

Foundry is a control plane with observability for Nemotron, Bedrock is model hosting. Azure is better in hybrid Physical AI, AWS in OpenAI Frontier. Azure leads in enterprise compliance.

What are the risks of deploying Physical AI in Azure?

Policy drift in hybrid - 72% of cases, data modernization bottleneck. Solution through Arc and Fabric. Risk of billing surprises in startup credits up to 20% overrun.

How long does it take to deploy AI in Foundry?

From simulation in Omniverse to production - 4-6 weeks. Agent Service GA speeds up by 50%. Full pipeline with Kubernetes - 2 months for enterprises.

Best platforms for AI DevOps in 2026?

Azure Foundry + Kubernetes for production AI, 25% share. Alashed IT recommends for Central Asia. Savings of 40% on inference vs AWS.

Читайте также

Источники

Источник фото: simplywall.st