Azure

Azure Cobalt 200 VMs Boost Agentic AI Performance

3 min read

Summary

Microsoft has announced early access preview for Azure Cobalt 200 Arm-based VMs, delivering up to 50% better generational CPU performance than Cobalt 100 for cloud-native, Linux-based, and agentic AI workloads. The new VMs add higher storage and networking performance, scale to 128 vCPUs, and enable memory encryption by default, making them important for organizations optimizing AI inferencing, data pipelines, and modern web services.

Need help with Azure?Talk to an Expert

Introduction

Microsoft has introduced Azure Cobalt 200 VMs in early access preview at Build 2026, targeting the growing demand for infrastructure that can handle modern agentic AI workloads. For Azure customers running scale-out, Linux-based, and cloud-native services, this update matters because it combines higher performance, stronger default security, and better efficiency in a new Arm-based VM generation.

What’s new in Azure Cobalt 200 VMs

Azure Cobalt 200 is Microsoft’s second-generation custom Arm processor platform and is designed for AI inferencing, data-intensive applications, and web/API tiers.

Key improvements include:

  • Up to 50% better CPU performance compared to Cobalt 100
  • Up to 20% higher remote storage IOPS with NVMe
  • Up to 10% better remote storage throughput with NVMe
  • Up to 15% higher network bandwidth
  • Scaling up to 128 vCPUs for larger workloads
  • Memory encryption enabled by default with minimal performance impact
  • Larger cache design, including 3 MB L2 cache per core and 192 MB system-level L3 cache

Microsoft says these enhancements are especially suited for workloads that run continuously and at scale, such as AI agents, distributed applications, analytics engines, and databases.

Performance highlights

Beyond the headline 50% generational gain, Microsoft shared additional workload-specific improvements versus Cobalt 100:

  • Up to 135% better performance for cloud database workloads
  • Up to 40% better performance for web serving workloads
  • Up to 45% better performance for communication encryption workloads
  • Up to 80% better performance for caching workloads

Azure Boost integration also helps improve remote storage and networking performance, which is relevant for high-throughput services and storage-heavy architectures.

Why this matters for IT administrators

For Azure administrators and platform teams, Cobalt 200 could offer a more efficient option for:

  • Linux-based application hosting
  • AI inferencing and agent orchestration
  • Data pipelines and analytics platforms
  • High-scale web and API services
  • Security-sensitive workloads that benefit from default memory encryption

Organizations already standardizing on Arm-compatible software stacks may see improved price-performance and density for production services. The preview also signals Microsoft’s continued investment in custom silicon for Azure infrastructure.

Next steps

Admins and architects should:

  • Evaluate whether current workloads are Arm-compatible
  • Identify cloud-native and scale-out applications that could benefit from Cobalt 200
  • Review early access preview availability for relevant Azure regions and VM families
  • Test performance for AI, database, caching, and API workloads before broader rollout

As Azure expands its custom silicon roadmap, Cobalt 200 looks positioned to become a strong option for enterprises building modern AI and cloud-native platforms.

Need help with Azure?

Our experts can help you implement and optimize your Microsoft solutions.

Talk to an Expert

Stay updated on Microsoft technologies

AzureArm VMsagentic AIcloud-nativeAzure Cobalt 200

Related Posts

Azure

Azure Foundry IQ Adds Serverless Retrieval and MCP

Microsoft has expanded Azure Foundry IQ with serverless retrieval in public preview, new multi-source knowledge connectors, and generally available knowledge bases for production agent workloads. The updates help developers build and scale grounded AI agents faster while improving security, retrieval quality, and access to both enterprise and web data.

Azure

Microsoft Discovery GA: R&D AI Platform and App Preview

Microsoft has made Microsoft Discovery generally available as a production-ready platform for building and governing agentic AI workflows in scientific and engineering research. It also introduced the Microsoft Discovery app in preview, giving researchers and academic teams a simpler local entry point before moving to enterprise-scale deployments.

Azure

Azure AI Agent Platform: Microsoft’s Enterprise Vision

Microsoft outlined its broader Azure-led strategy for enterprise AI, arguing that successful adoption depends on a governed, integrated system around agents rather than standalone models or chatbots. The company is positioning Azure, GitHub, Microsoft IQ, Foundry, Microsoft 365, and Security tools as a unified platform to build, run, govern, and continuously improve AI agents at scale.

Azure

Microsoft Foundry: Managing AI Models, Cost, Quality

Microsoft Foundry is expanding its model ecosystem and operational tooling to help developers manage AI applications across selection, evaluation, optimization, and production operations. The update includes general availability of Fireworks AI on Microsoft Foundry, giving teams more model choice through a single Azure endpoint while improving cost control, governance, and lifecycle management.

Azure

Microsoft Build 2026: Fabric and Databases for AI Apps

At Microsoft Build 2026, Microsoft introduced new data platform capabilities aimed at helping developers move AI and agent-based apps from prototype to production. Key announcements include Rayfin for building Fabric-backed app backends, Azure HorizonDB in public preview for AI-ready PostgreSQL workloads, and new security and migration tools for Azure Database for PostgreSQL.

Azure

Claude Opus 4.8 in Microsoft Foundry Now Available

Microsoft Foundry now includes Anthropic Claude Opus 4.8, giving developers and enterprises access to a stronger model for coding, agentic workflows, and document-heavy analysis. The release matters because it expands model choice in Foundry while helping teams build and evaluate advanced AI applications with enterprise controls.