Microsoft announces Azure Cobalt 200, its next-generation ARM CPU for Azure Virtual Machines

Microsoft has unveiled its next-generation, in-house-developed ARM processor, the Cobalt 200, for Azure VMs, claiming up to 50% higher performance than its predecessor.

Back in 2023, Microsoft first revealed that it was working on an ARM processor named Microsoft Cobalt 100 for general-purpose computing workloads on the Microsoft Cloud. Last year, during the Build conference, Microsoft announced the private preview of Cobalt 100-based VMs. In late 2024, Microsoft announced the general availability of Cobalt 100-based VMs, claiming up to 50% better price-performance.

At Ignite 2025 today, Microsoft announced Cobalt 200, the next-generation, in-house-developed ARM processor for Azure VMs. The Cobalt 100 is a 64-bit processor built on Arm's Neoverse N2 architecture, running at 3.4 GHz. The new Cobalt 200 processor features more cores, larger cache, and faster memory. Also, this latest-generation processor is built on the latest Arm architecture and 3nm TSMC process technology for better performance and efficiency.

Microsoft claims that Cobalt 200 can offer up to 50% higher performance than Cobalt 100. Also, Cobalt 200 will be Azure's most power-efficient compute platform. For improved efficiency and security, the Cobalt 200 VMs will have integration with the current generation of Azure Boost and Azure-integrated HSM.

Microsoft team wrote the following regarding the Cobalt 200 announcement:

Cobalt 200 will advance Microsoft’s systems approach to infrastructure, integrating the latest silicon technology with infrastructure enhancements, including Azure Boost and the Azure-integrated hardware security module (HSM).

Surprisingly, Microsoft didn't mention anything about Maia 100. It's a custom AI accelerator designed to handle large-scale AI workloads on Azure's cloud infrastructure. A few months back, The Information reported that Microsoft has delayed mass production of its next AI chip, Braga (expected to launch as Maia 200), by at least six months due to design changes, staffing issues, and instability introduced by OpenAI-requested features. The lack of any news about a successor to Maia 100 indicates that the report from The Information may be accurate.