We’re saying the early entry preview for Azure Cobalt 200 Arm-based Digital Machines (VMs), designed for Linux-based agentic AI workloads.
On this article
As organizations more and more embrace the agentic period for AI, buyer calls for for compute are reshaping the structure of cloud infrastructure as we all know it. As we speak at Microsoft Construct 2026, we’re saying the early entry preview for Azure Cobalt 200 Arm-based Digital Machines (VMs), designed from the bottom up for scale-out, cloud-native, and Linux-based agentic AI workloads, with as much as 50% higher generational efficiency over Cobalt 100.
Cobalt 200 is purpose-built from silicon to servers to providers—integrating Microsoft’s newest improvements in safety, networking, storage, and offload to outperform conventional Arm-based compute. That hardware-software co-optimization lets us push the boundaries of scale, safety, and price for AI inferencing, information pipelines, and the net and API tiers that energy fashionable providers. Brokers are distinctive from conventional workloads: they cause, make sequential selections, and run repeatedly at scale—demanding a essentially completely different computational profile. Cobalt 200 is constructed for precisely this atmosphere, delivering 50% efficiency positive factors for the workloads defining the following period of enterprise AI, making brokers sooner, extra succesful, and economically viable at scale.
Constructing on the success of Cobalt 100 VMs
Microsoft Azure Cobalt 100, our first custom-built processor for cloud-native workloads is now deployed in 32 Azure datacenter areas world wide, with cloud analytics leaders like Databricks and Snowflake adopting Cobalt 100 to optimize their cloud footprint, whereas prospects like Amadeus, OneTrust, Siemens, Sprinklr, and Temenos have achieved important real-world efficiency and effectivity positive factors.
Inside Microsoft’s personal cloud providers, Azure Cobalt 100 VMs are delivering as much as 45% higher efficiency whereas utilizing 35% fewer compute cores in comparison with its earlier compute platform. Microsoft Defender for Endpoint (MDE) noticed 40% higher efficiency in its cyber information curator, enabling sooner risk response at large scale.
With Cobalt 200 VMs, we’re making use of every little thing we discovered in the direction of supporting these demanding providers to lift the bar even additional.
What’s new in Cobalt 200 Arm-based VMs
The Cobalt 200 CPU represents a major leap ahead in efficiency, safety, and effectivity. On the coronary heart of each Cobalt 200 VM is the Cobalt 200 System-on-Chip (SoC), our second-generation Arm processor constructed across the Arm Neoverse V3 Compute Subsystems, the highest-performance V-series core in Arm’s portfolio. Fabricated on TSMC’s 3nm (N3P) course of, Cobalt 200 incorporates a fashionable chiplet structure, {custom} accelerators, and a {custom} reminiscence controller.
Key improvements in Cobalt 200 VMs embody:
Generational efficiency positive factors throughout compute, storage, and networking: In comparison with previous-generation Cobalt 100 VMs, new Cobalt 200 VMs ship as much as 50% higher CPU efficiency, 20% increased distant storage IOPS with NVMe, 10% higher distant storage throughput with NVMe, and 15% increased community bandwidth, with enhancements various by workload.
As much as 128 vCPUs: Cobalt 200 VMs scale as much as 128 vCPUs, delivering extra compute capability for demanding scale-out, cloud-native, agentic AI, and data-intensive workloads.
Sooner distant storage and networking: Azure Increase integration helps enhance distant storage IOPS and throughput with NVMe whereas rising community bandwidth, benefiting distributed purposes, storage-intensive providers, and high-throughput information pipelines.
Superior cache and scalable system design: A contemporary chiplet structure with a bigger cache hierarchy—together with 3 MB of L2 cache per core and 192 MB of system-level L3 cache—retains extra energetic information nearer to the workload, serving to scale back latency and enhance responsiveness for databases, in-memory caches, analytics engines, and different data-intensive providers.
Stronger safety by default: Reminiscence encryption is enabled by default by a custom-designed reminiscence controller, serving to elevate the baseline safety posture for each workload with negligible efficiency impression.
Efficiency for agentic AI workloads
Cobalt 200 delivers the per-core efficiency and scalability wanted to energy fashionable agentic AI workloads. Every Cobalt 200 core is a full bodily core, paired with devoted 3 MB of L2 cache, and main reminiscence bandwidth per core. These design factors allow increased isolation and sustained efficiency beneath load, which permit agentic workloads to pack extra agent sandboxes per VM whereas assembly latency and throughput necessities.
Actual cloud workload efficiency
Cobalt 200 VMs present a broad generational uplift over Cobalt 100 on the workloads that matter most in manufacturing.
With cloud workloads, we observe:
As much as 135% higher efficiency for cloud database workloads.
As much as 40% higher efficiency for net serving workloads.
As much as 45% higher efficiency for communication encryption workloads.
And as much as 80% higher efficiency for caching workloads.
The uplift measured is seen in actual hyperscale providers. Taken collectively, these outcomes present that Cobalt 200 raises the ground on per-core efficiency throughout cloud-native purposes, databases, analytics, caches, and communications workloads.
Business companions and buyer adoption
We now have been working intently with our know-how companions and prospects in the course of the preview interval to make sure Cobalt 200 Arm-based VMs ship robust outcomes throughout a variety of real-world workloads. Main software program improvement corporations and enterprises are already evaluating and adopting these VMs for his or her most demanding purposes.
Teradata
Teradata is happy to be an early preview companion for Microsoft’s Cobalt 200 VMs. Microsoft has been a robust companion, and we worth the chance to assist form the design and specs to raised meet the wants of our joint prospects. Early testing has been encouraging, and we sit up for continued collaboration.
—Brandon Mincey, Engineering Fellow, Teradata
Elastic
Elastic is dedicated to delivering best-in-class efficiency with the Search AI Platform that powers observability, safety, search, and AI options. Efficiency and price effectivity are essential for groups operating these workloads at scale and our preliminary testing of the brand new Cobalt 200 VMs exhibits promise for additional enhancements in these areas. We sit up for bringing these advantages to Elasticsearch customers on Azure by our continued collaboration with Microsoft.
—Yuvraj Gupta, Director, Product Administration, Elastic
Arm
Agentic AI is reshaping the cloud and creating demand for infrastructure that may effectively orchestrate and scale thousands and thousands of clever interactions in actual time. Our collaboration with Microsoft on Cobalt 200, constructed on Arm Neoverse CSS V3, displays how purpose-built Arm-based compute is enabling the following era of AI-driven providers whereas persevering with to ship distinctive efficiency for cloud-native purposes.
—Eddie Ramirez, Vice President of go-to-market, Cloud AI Enterprise Unit, Arm
Canonical
Cobalt 200 delivers key advances for manufacturing Linux workloads on Arm, together with reminiscence encryption enabled by default, built-in acceleration for compression and encryption, and better throughput for data-intensive cloud providers. Ubuntu offers organizations a constant platform for cloud-native and agentic AI workloads throughout Azure, and Ubuntu Professional retains it production-ready with long-term safety upkeep and Livepatch, which now brings rebootless kernel updates to Arm so always-on providers keep obtainable.
—Jehudi Castro-Sierra, Public Cloud Alliance Director, Canonical
Developer ecosystem and Arm compatibility
The Arm developer ecosystem continues to thrive. Cobalt 200 Arm-based VMs ship full compatibility for workloads at the moment operating on Cobalt 100 VMs, making migration seamless. Main developer platforms and languages—together with C++, .NET, Java, Python, and Rust—present Arm-native variations with optimizations that absolutely leverage the capabilities of the Arm structure.
The broader ecosystem has embraced Arm with native assist throughout common infrastructure and deployment options. GitHub Actions helps Arm by each self-hosted and GitHub-hosted runners. Azure Kubernetes Service (AKS) helps Arm agent nodes in addition to combined x86 and Arm structure clusters. Containerized workloads profit from the rising availability of Arm-native container pictures throughout the ecosystem.
Microsoft providers powered by Cobalt 200 VMs
Microsoft’s personal cloud providers are among the many first to undertake Cobalt 200 Arm-based VMs, constructing on the success of Cobalt 100 adoption throughout our most demanding, mission-critical workloads.
Dataverse
Dataverse is the core utility and information platform underpinning Dynamics and Energy Platform workloads, requiring excessive efficiency, scalability, and low latency to ship a responsive buyer expertise.
We deployed the Energy Apps platform on Cobalt 100 for its improved efficiency and energy effectivity. We’re validating Cobalt 200 now and excited by the efficiency positive factors we’re seeing–as much as 60% higher efficiency for our base workload over Cobalt 100.
—Mauktik Gandhi, VP Engineering, Agent 365 Platform
Azure Databases
Azure SQL Database is a pure match for Cobalt 200 Arm-based VMs. The built-in compression and cryptography accelerators are notably impactful for database workloads—by offloading compression and encryption duties to the Cobalt 200 accelerator, Azure SQL is ready to scale back using essential compute sources, prioritizing them for buyer queries and transactions.
Our groups collaborated from the start to make sure that the structure of Cobalt 200 is optimized for internet hosting databases in Azure. We’re excited concerning the efficiency positive factors we’re seeing over Cobalt 100 and searching ahead to broader availability later this 12 months.
—Shireesh Thota, Company Vice President, Azure Databases
VM households and specs
Cobalt 200 VMs considerably broaden our Arm VM portfolio based mostly on buyer suggestions to assist a broader set of workloads. Whereas Cobalt 100 supplied Basic Goal (Dp,Dpl), and Reminiscence Optimized (Ep) VM households, Cobalt 200 provides two extra VM households: the Excessive-Reminiscence Optimized Mpsv4 VMs and Dense Native Storage Lpsv5 VMs, bringing extra alternative throughout compute, reminiscence, and storage profiles.
All VM sequence ship as much as 85 Gbps of community bandwidth and 70 Gbps of distant storage throughput, besides Mpsv4/Mpdsv4, which give as much as 70 Gbps of community bandwidth and 46 Gbps of distant storage throughput. Most sequence can be found with or with out native NVMe disks, whereas Lpsv5 contains native NVMe storage throughout all sizes, giving prospects flexibility to optimize for value or efficiency.
Most non-memory-intensive and scale-out workloads corresponding to microservices, small databases, caches, gaming servers, and extra.
Basic Goal Dpsv7/Dpdsv7 1–1284:1Yes – As much as 7 TiB
Most scale-out enterprise workloads corresponding to net and utility servers, small to medium databases, caches, and extra.
Reminiscence Optimized Epsv7/Epdsv7 1–1288:1Yes – As much as 7 TiB
Massive relational and NoSQL databases, in-memory caches corresponding to Redis and Memcached, and real-time analytics.
(New) Excessive Reminiscence Optimized Mpsv4/Mpdsv41–8416:1Yes – As much as 4.4 TiB
Massive in-memory databases, ERP techniques, large-scale caching layers, and memory-intensive analytics workloads.
(New) Storage Optimized Lpsv5 1–1288:1Yes – As much as 23 TB
Information pre-processing and staging, relational and NoSQL databases with native storage necessities, large information analytics, and search/index engines.
All Cobalt 200 VMs will assist distant disk sorts together with Normal SSD, Normal HDD, Premium SSD, and Extremely Disk storage. You’ll be able to deploy these new VMs utilizing current strategies, together with the Azure portal, SDKs, APIs, PowerShell, and the command-line interface (CLI).
Availability
Cobalt 200 Arm-based VMs at the moment are obtainable in preview. The brand new VMs can be obtainable in preview within the following areas: West US3, East US2, Central US, Sweden Central, East US, West US2, Spain Central, and Indonesia Central, with extra areas to be introduced.
The following chapter in Azure’s {custom} silicon journey
The announcement of Cobalt 200 Arm-based Digital Machines marks the following chapter in Azure’s purpose-built infrastructure journey. With our end-to-end techniques method—designing the CPU in-house and optimizing it for Azure’s infrastructure—we’re delivering a tightly built-in platform that gives efficiency, energy effectivity, and safety for our prospects.
Whether or not you’re accelerating product improvement, scaling analytics platforms, operating mission-critical databases, or bettering consumer experiences, Cobalt 200 Arm-based VMs supply a compelling alternative for contemporary cloud workloads. We’re excited to see how our prospects and companions create breakthrough services and products on this new platform.
Strive the Cobalt 200 Arm-based VMs immediately in early entry preview and expertise the following era of Azure’s {custom} infrastructure innovation.
Thanks for becoming a member of us on this thrilling journey.
Get early entry to Azure Cobalt 200 VMs.
Extra sources
Run next-generation AI workloads with Azure Cobalt 200
Uncover how Azure Cobalt 200 Arm-based VMs ship high-performance, scalable compute for agentic AI, cloud-native, and data-intensive workloads.

