Microsoft Makes Cobalt 200 Virtual Machines Generally Available in 14 Azure Regions
Microsoft's Cobalt 200 virtual machines go live in 14 regions
Microsoft made its Azure Cobalt 200 virtual machines generally available on Nov. 17, 2026. The launch gives cloud customers access to the company's second-generation custom Arm server processor in 14 Azure regions. A virtual machine is a rented slice of a physical server that customers use to run their own software.
The move takes the instances beyond the early-access preview Microsoft announced on June 2. Microsoft says Cobalt 200 delivers up to 50% better performance than the previous Cobalt 100 generation. The company positions the machines for databases, web services, and CPU-heavy work that supports AI agents, such as code compilation and tool execution.
Key details of the launch
- Cobalt 200 is built on Arm Neoverse CSS V3, according to Arm.
- Microsoft says the chip has 132 cores and assigns a full physical core to each virtual CPU, or vCPU, in the VMs.
- The launch lineup covers general-purpose, memory-optimized, high-memory, and storage-optimized options.
- Most series list sizes from 1 to 128 vCPUs.
- The new storage-optimized Lpsv5 offers up to 23 TB of local NVMe storage for work like data staging, search, and indexing.
- Customers can deploy through the Azure portal, command-line tools, and APIs, including as Azure Kubernetes Service node pools.
What Microsoft announced on launch day
Microsoft's launch announcement, published Nov. 17, lists five VM families with different memory allocations per vCPU. The Dplsv7/Dpldsv7 general-purpose series provides 2 GiB per vCPU, while the Dpsv7/Dpdsv7 general-purpose series provides 4 GiB. The Epsv7/Epdsv7 memory-optimized series provides 8 GiB per vCPU. The new Mpsv4/Mpdsv4 high-memory series provides 16 GiB per vCPU and targets large in-memory databases and caches. The storage-optimized Lpsv5 also provides 8 GiB per vCPU.
The 14 launch 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 also published a customer statement from OpenAI. Vadim Filanovsky, a technical staff member at OpenAI, said ChatGPT services already ran on Arm before the company tested Cobalt 200. He said the new generation delivered higher throughput and lower latency, adding: "That means we can serve more users with the same number of cores."
What the benchmarks actually measure
Microsoft's launch benchmarks reported up to 1.5 times the performance of Cobalt 100 for AI-agent tool execution and up to 1.9 times for database workloads.
Arm ran separate tests comparing Cobalt 200 with Azure instances using AMD EPYC Turin processors. Arm reported up to 139% higher MySQL performance and up to 46% higher SWE-bench task-completion performance at 144 concurrent workers.
The SWE-bench result measures CPU work around an AI agent, not better reasoning by the AI model itself. Arm's testing methodology document states that it supplied prerecorded language-model responses and tool calls to measure CPU-side orchestration, tool execution, and container overhead.
What Microsoft plans next
Microsoft said Cobalt 200 would represent a "significant share" of new Azure Compute capacity beginning in 2027. The company also said regional availability will expand to 23 regions by the end of 2026.