Microsoft has made its Azure Cobalt 200 virtual machines generally available, running on in-house Arm-based processors that deliver up to 50% better CPU performance than the previous Cobalt 100 generation.
The chips are built on Arm's Neoverse Compute Subsystems V3 design.
Each virtual CPU (vCPU) gets a complete physical core rather than a smaller virtualised slice.
The chips run at up to 3.6 GHz in Opportunistic mode or 3.2 GHz in Deterministic mode.
Cobalt 200 is aimed at cloud-native workloads that scale out across many machines, such as web services, databases, caches and microservices.
It is available across Microsoft's D-series, E-series, M-series and L-series instance families, with memory ranging from 2 to 16 GiB per vCPU.
The virtual machines are live in 14 Azure regions, and Microsoft plans to expand to 23 by the end of 2026.
Cobalt 100, Microsoft's first in-house Arm chip, had already shown up to 1.9 times the performance and 2.8 times the price-performance of comparable x86 instances on some AI inference workloads.
The launch is part of a wider move by the hyperscalers, the biggest cloud providers, towards custom Arm chips.
Amazon Web Services (AWS) has Graviton and Google has Axion.
Arm has said Graviton now powers more than 50% of AWS's recent capacity, while Google has cited up to twice the performance per watt for Axion compared with x86 alternatives.
Arm-based processors are expected to account for more than 60% of server deployments at the leading hyperscalers by the end of 2026.
Cloud operators are tailoring their infrastructure for more demanding workloads, including AI agents that write and run code and coordinate tasks across applications.
Your draft attributes the 60% figure to "the primary source" without naming it, so I've left a placeholder. If it's from Arm's site, like the other links, credit Arm.
Read more