Azure Cobalt 200 VMs Boost Agentic AI Performance

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.