At the recent annual Ignite conference, Microsoft introduced a successor to Azure Cobalt 100 ARM processors: Cobalt 200, which delivers a 50% performance improvement and integrates with the latest Microsoft security, networking, and storage technologies.
The company claims that since the GA release of the Cobalt 100 VMs, various companies, such as Databricks and Snowflake, have adopted them for their workloads. Hence, the path for a successor was set with Cobalt 200, designed using a portfolio of benchmarks drawn directly from the customers’ Azure usage patterns, including databases, web servers, storage caches, network transactions, and data analytics.
The development of Cobalt 200 utilized a highly advanced, data-driven approach. Microsoft developed a complete digital twin simulation from the silicon up, starting with the CPU core microarchitecture and extending through the entire server design.
Using AI and Azure compute, Microsoft modeled the performance and power consumption of 140 benchmarks against 2,800 combinations of System-on-Chip (SoC) and system design parameters, including core count, cache size, memory speed, and rack configuration.
The authors of an Azure Infrastructure blog post write:
This resulted in evaluating over 350,000 configuration candidates for the Cobalt 200 system during our design process. This extensive modelling and simulation helped us to quickly iterate to find the optimal design point for Cobalt 200, delivering over 50% increased performance compared to Cobalt 100, all while continuing to provide our most power-efficient platform in Azure.
The Cobalt 200 server features the advanced Cobalt 200 SoC, built on Arm’s Neoverse Compute Subsystems V3. Each SoC has 132 active cores, 3MB of L2 cache per core, and 192MB of L3 system cache, providing outstanding performance for customer workloads.
(Source: Microsoft Azure Infrastructure blog post)
The move to the Neoverse V3 architecture provides an inherent advantage in instruction-per-cycle (IPC) efficiency over competitors that rely on Neoverse V2 (such as Amazon Graviton 4 and NVIDIA Grace) or Neoverse N2 (such as its predecessor, Cobalt 100). Cobalt 200 features 132 cores, giving it a lead over the 96 cores in Graviton 4 and the 72 cores in NVIDIA Grace. The jump in both core count and architecture generation suggests that Microsoft is aiming for a major, multi-generational leap in compute density and efficiency, setting a new benchmark for cloud-native workloads.
A comparison shared in a Reddit thread highlights Cobalt 200’s lead:
CPU
# of cores
uArch
Microsoft Cobalt 200
132
Neoverse V3
Microsoft Cobalt 100
128
Neoverse N2
NVIDIA Grace
72
Neoverse V2
Amazon Graviton 4
96
Neoverse V2
Amazon Graviton 3
64
Neoverse V1
Google Axion N4A
64
Neoverse N3
Google Axion C4A
72
Neoverse V2
The company also states that Azure Cobalt is more than just an SoC, as the latest Azure Boost capabilities are built into the new Cobalt 200 system, which enhances networking and remote storage performance by increasing bandwidth and offloading tasks to custom hardware. Moreover, this improvement reduces latency and boosts overall workload efficiency. Additionally, the Cobalt 200 integrates the Azure Hardware Security Module (HSM) for top-tier cryptographic key protection, ensuring the security of sensitive data. It works seamlessly with Azure Key Vault for easier management of encryption keys, while also providing high availability, scalability, and compliance with FIPS 140-3 Level 3 standards.
Lastly, the first production Cobalt 200 servers are now live in Microsoft’s Azure data centers, with a wider rollout and customer availability coming in 2026.
