Microsoft has made Azure Cobalt 200-based virtual machines generally available, announcing the release at Microsoft Ignite 2026. The company says the new VMs deliver up to 50% better performance than Cobalt 100 and are available in 14 Azure regions, with availability expected to expand to 23 regions by the end of 2026.
Microsoft designed the Arm-based instances for cloud-native and data-intensive applications, as well as agentic AI workloads such as agent sandboxes, file search, code compilation and tool execution. Higher per-core performance can help reduce latency in these tasks, according to the company. Microsoft expects Cobalt 200 to account for a significant share of new Azure compute capacity beginning in 2027.
Performance claims and customer testing
Microsoft reports benchmark gains over Cobalt 100 of up to 1.5 times for web serving, 1.6 times for Java workloads, 1.5 times for AI agent tool execution and 1.9 times for database workloads. Those are company-reported results; customer experiences vary by application and configuration.
Several customers described results from their own evaluations. OpenAI’s Vadim Filanovsky said the company observed higher throughput and lower latency when comparing the same ChatGPT workloads on Cobalt 200 and Cobalt 100. Databricks said its serverless compute on Cobalt starts faster and runs more efficiently, while Snowflake described performance and efficiency improvements across successive Azure Arm VM generations.
Amadeus reported more than 30% higher flight-search throughput in its benchmark lab compared with Cobalt 100. InMobi said its ad-serving tests showed performance and efficiency improvements; the company also described local neural-network inference on 32-core machines, reporting approximately 50,000 inferences per second. These figures reflect the specific workloads and tests cited by the respective companies.
Silicon, networking and processor modes
At the hardware level, the Cobalt 200 system-on-chip uses TSMC’s 3nm process and combines 132 Arm Neoverse V3 cores in a multi-chiplet architecture. Each vCPU corresponds to a full physical core. Microsoft specifies 3MB of private L2 cache per core and 192MB of system-level cache. The chip’s memory controller supports 12 channels of DDR5, and Microsoft says memory encryption is enabled by default with negligible latency.
Every Cobalt 200 VM includes Azure Boost, which offloads networking and storage processing from the host. Compared with Cobalt 100, Microsoft cites up to 15% more network bandwidth, 20% higher remote-storage IOPS and 10% higher remote-storage throughput. The servers also include an Azure Integrated HSM for hardware protection and encryption-key storage, plus compression and cryptography accelerators. Microsoft says on-chip accelerators can reduce Azure SQL backup times by up to 2.5 times while minimizing disruption to live workloads.
Customers can select between two processor modes. Deterministic Mode keeps cores at a consistent 3.2 GHz. Opportunistic Mode can boost them to 3.6 GHz when power and thermal headroom permit; Microsoft says this can provide up to 12% more performance, with some frequency variability. The ProcessorMode setting can be changed during VM creation or later through the Azure portal, CLI, PowerShell, Azure Resource Manager or Bicep templates.
VM families and deployment
Cobalt 200 spans five series across general-purpose, memory-optimized, high-memory and storage-optimized options. Microsoft says sizes range from 1 to 128 vCPUs in most families, with Mpsv4 and Mpdsv4 high-memory sizes reaching up to 84 vCPUs. The new Mpsv4 family provides 16 GiB of memory per vCPU for workloads such as in-memory databases and analytics. The storage-optimized Lpsv5 family offers local NVMe storage of up to 23 TB per VM.
Available regions are Australia East, Canada Central, Central US, East US, East US 2, Indonesia Central, Italy North, Korea Central, South Central US, Spain Central, Sweden Central, UK South, West US 2 and West US 3. Microsoft says the instances support popular Linux distributions, Windows 11 Client, Azure Kubernetes Service node pools and Virtual Machine Scale Sets. Deployment is available through the Azure portal, SDKs, APIs, PowerShell and Azure CLI. The company also says workloads already running on Cobalt 100 can move to Cobalt 200 where they need additional performance or the newer VM families.