During its Ignite conference in San Francisco on Tuesday, Microsoft announced general availability of Cobalt 200 virtual machines, which run on the company's second-generation Arm-based processors. The VMs are now deployed across 14 Azure regions, with plans to expand to 23 by year's end. According to Microsoft's performance data, the new chips deliver up to 50% better performance than Cobalt 100, with benchmarks showing up to 1.5 times the performance specifically for AI agent tool execution—operations such as file searching, code compilation, and sandbox task execution.

The Cobalt 200 announcement comes roughly a year after Microsoft first introduced the chip and approximately six months following its entry into preview status. Pat Stemen, Microsoft's vice president for Azure Cobalt, explained to Kernelwire that the engineering team devoted significant effort over the past year to optimizing the infrastructure for these particular operations. According to Stemen, the chip's architecture—featuring substantial caches, strong memory performance, and multiple physical cores—was already well-suited to workloads involving sandbox initialization. Microsoft subsequently developed "a set of benchmarks inside the company for those common operations" and refined the product based on those internal measurements.

Two processor modes for different workload requirements

Cobalt 200 offers developers a choice between two operational modes. Deterministic mode maintains cores at a fixed 3.2 GHz frequency, whereas opportunistic mode permits cores to reach 3.6 GHz when power and thermal conditions permit. Microsoft indicates that opportunistic mode can yield up to 12% additional performance, though clock speeds will fluctuate depending on system conditions.

Stemen noted that "Customers love fixed frequency, by the way. They love the predictability that comes with that." Microsoft recommends deterministic mode for distributed data and analytics clusters—environments where the slowest node determines overall throughput—and for services with strict performance requirements. Web and API services, transaction processing, and media workloads may gain more benefit from opportunistic mode's higher speeds.

Developers can select their preferred mode when creating each VM or adjust it afterward using the Azure portal, command-line interfaces, or infrastructure-as-code templates. Stemen emphasized the flexibility available: "So if you buy four cores from us, you can choose all four of them to run flexibly or deterministically." He added that the performance advantage typically remains available until the machine's cores reach heavy utilization.

Hardware design tailored to Azure requirements

The Cobalt 200 processor contains 132 Arm Neoverse V3 cores manufactured using TSMC's 3nm process technology, with VMs scaling up to 128 vCPUs. Each vCPU maps to a full physical core and includes 3 MB of dedicated L2 cache. The chip incorporates 192 MB of system-level cache and 12 channels of DDR5 memory.

The transition from Cobalt 100 involved shifting from a cost-optimized Arm core design to a performance-optimized architecture, accompanied by expanded cache capacity and enhanced memory bandwidth. Every Cobalt 200 VM includes Azure Boost, a feature that offloads networking and storage processing from the primary host. Microsoft reports achieving up to 15% higher network bandwidth and 20% higher remote storage IOPS relative to Cobalt 100.

Stemen described the design methodology: the team segments Microsoft's internal workloads into discrete components, models their behavior before hardware fabrication, and leverages those simulations to inform decisions regarding cache architecture, memory configuration, and even server rack cooling systems. Product teams participate in early planning phases to shape the product direction. "They're influencing the beginning of the product from when it's just in PowerPoint," Stemen said.

Real-world performance examples demonstrate the improvements. Microsoft's Teams media-processing service achieved 1.65 times the performance on Cobalt 200 compared with Cobalt 100, with the service demonstrating linear scaling across physical cores and reaching 90% vCPU utilization while handling live traffic. Outside Microsoft, travel technology company Amadeus reported more than 30% higher flight-search throughput in its benchmark testing compared with Cobalt 100.

Expanding Arm adoption across Microsoft services

Multiple Microsoft services already operate on Cobalt infrastructure, including Azure Database for PostgreSQL, Azure DevOps, Azure SQL, Microsoft 365, and Xbox. Stemen observed that customer concerns about migration complexity have diminished as evidence of successful deployments has accumulated. "Customers had real concern about the lift," he noted. "Since then, there's more evidence of that working out, and the tools have improved, and the products have gotten better all at the same time."

Beyond general-purpose and memory-optimized VM families, Cobalt 200 introduces two additional VM families to general availability: the high-memory Mpsv4 family, offering 16 GiB of memory per vCPU, and the storage-optimized Lpsv5 family, providing up to 23 TB of local NVMe storage. Stemen indicated that customer interest has shifted toward large in-memory databases and substantial existing enterprise workloads. "The ability to port them now being easier has helped," he said.

While Microsoft will maintain investment in x86 processor development, the company projects that Cobalt 200 will constitute "a significant portion of our new Azure Compute capacity in 2027 and moving forward." Developers can provision these VMs through existing Azure tooling and integrate them into Virtual Machine Scale Sets or Azure Kubernetes Service node pools.