Microsoft made virtual machines built on Cobalt 200, its second-generation Arm server chip, generally available on Azure on 17 November at Ignite, according to a Microsoft blog post. Microsoft says they deliver “up to 50% better performance than Cobalt 100”, and lists OpenAI, Databricks, Snowflake, Amadeus and InMobi among customers.
Our number: the 4-vCPU, 16 GiB Cobalt 200 size (D4ps v7) lists at $0.196 an hour in East US, 26% below the same-size Intel Xeon 6 machine (D4s v7) at $0.265, according to Azure’s public price list read on 18 November, by our arithmetic. It costs 40% more than the Cobalt 100 equivalent at $0.140.
Business Pill · Switching Costs
An evergreen Business Engineer explainer on what it costs a customer to move from one supplier to another. It teaches the general idea only and does not discuss Microsoft, Intel or Arm.
The key insight: Microsoft made its second-generation Arm server chip, Cobalt 200, generally available on Azure; the same-size Cobalt 200 machine lists 26% below the Intel Xeon 6 option in East US by our arithmetic, and Microsoft claims up to 50% more performance than Cobalt 100.
What Microsoft Launched
Cobalt 200 VMs are generally available in 14 Azure regions, including East US, West US 2, UK South and Sweden Central, and Microsoft says availability will expand to 23 regions by the end of 2026. They come in five VM series across general purpose, memory-optimized, high-memory and storage-optimized families, with up to 128 vCPUs per VM. Two families are new for Arm on Azure: a high-memory series with 16 GiB per vCPU and a storage-optimized series with up to 23 TB of local storage.
The chip is built on TSMC’s 3nm process and brings together “132 Arm Neoverse V3 cores”, Microsoft says. Each vCPU is a full physical core, and customers can run cores at a fixed 3.2 GHz or let them boost up to 3.6 GHz. Microsoft’s own benchmarks put Cobalt 200 at up to 1.5 times Cobalt 100 for web serving and 1.9 times for database workloads.
OpenAI is one of the named users. “Cobalt has helped us sustain ChatGPT’s growth”, Vadim Filanovsky, a member of technical staff at OpenAI, said in Microsoft’s post. “We saw higher throughput and lower latency on Cobalt 200”.

The Price Gap
On Azure’s retail price list, the four same-size machines rank as follows per hour in East US on pay-as-you-go terms: Intel Xeon 6 (D4s v7) $0.265, Cobalt 200 (D4ps v7) $0.196, AMD EPYC Genoa (D4as v6) $0.182 and Cobalt 100 (D4ps v6) $0.140. Processor types and sizes are from Microsoft’s size documentation.
So Cobalt 200 lists 26% below the Intel Xeon 6 size and 7.7% above the AMD EPYC Genoa size, by our arithmetic. Against Cobalt 100 the price rises 40%, while Microsoft claims up to 50% more performance; on that best case, performance per dollar improves by up to about 7% (1.5 / 1.4), by our arithmetic. List prices are not performance: Microsoft’s gains vary by workload.
Custom Silicon, Both Ends of the Rack
Microsoft’s Ignite roundup pairs Cobalt with its Maia AI accelerator: “We pair our own Maia and Cobalt chips with the latest from NVIDIA and AMD, so every workload runs on the right silicon for the job”, Arun Ulag wrote in the Ignite post. The same week, Microsoft launched its MAI models at Ignite; see also our update on Microsoft and rival chips.
On Nvidia’s 17 November earnings call, its interim chief executive said Nvidia expects total CPU revenue to more than double in fiscal 2028 (our call piece).
The Business Lever
In our analysis, the lever is owning the general-purpose server. With its own Arm CPU, Microsoft lists a same-size machine below the Intel option on its own cloud, and named customers such as OpenAI run production services on it.
The Structural Read
As we read Microsoft’s price list, Cobalt 200 is positioned between the two x86 options on hourly price: below Intel’s Xeon 6 size, slightly above AMD’s EPYC Genoa size.
As we read the step from Cobalt 100, Microsoft is charging more per hour for the new chip (40% more by our arithmetic) and pointing to performance gains of up to 50% that vary by workload.
Vadim Filanovsky, member of technical staff at OpenAI, in Microsoft’s Cobalt 200 post
“Cobalt has helped us sustain ChatGPT’s growth”
Three Implications
FOR AZURE CUSTOMERS Cobalt 200 VMs are generally available in 14 regions, with 23 planned by the end of 2026, Microsoft says.
FOR INTEL AND AMD On Azure’s list, the same-size Cobalt 200 machine is priced below the Intel Xeon 6 size and above the AMD EPYC Genoa size.
FOR ARM The chip uses 132 Arm Neoverse V3 cores, according to Microsoft.
The Business Engineer Lens
This story maps onto the Business Engineer framework The Nine Layers of AI.
The framework says: “It is a layered industrial stack — and at any given moment, the binding constraint sits at a different layer than it did six months ago.”
As we read it, Cobalt 200 is Microsoft building its own part of the chip layer for general-purpose compute, next to its Maia accelerator. The framework predates the launch and does not discuss Cobalt or Azure pricing.
What Is Not Established
Microsoft does not publish the cost of the chip or the margin on Cobalt VMs. Our comparison uses one size, one region and pay-as-you-go list prices; reserved, savings-plan and negotiated prices differ, and Microsoft says each Cobalt 200 vCPU is a full physical core, while the sources we read do not say the same of the Intel and AMD sizes. The performance figures are Microsoft’s and its customers’ own benchmarks.
The Bottom Line
Cobalt 200 VMs are generally available in 14 Azure regions. Same-size, they list at $0.196 an hour against $0.265 for Intel’s Xeon 6 option in East US, 26% less by our arithmetic, and Microsoft claims up to 50% more performance than Cobalt 100.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Product, performance, customer and region details are from Microsoft’s Azure blog posts of 17 November 2026, which we read in full; processor types are from Microsoft Learn. The hourly prices are from Azure’s public Retail Prices API for East US, pay-as-you-go, read on 18 November 2026. The 26%, 7.7%, 40% and 7% figures are our arithmetic. Nvidia’s CPU outlook is from our transcription of its 17 November earnings call. We first saw the news via a relay post on X. We did not contact Microsoft.
Sources: Microsoft Azure blog, Cobalt 200 GA, 17 Nov 2026 · Microsoft Azure blog, Ignite 2026 roundup, 17 Nov 2026 · Microsoft Azure blog, Cobalt 200 preview (Build 2026) · Microsoft Learn, Dpsv7/Dsv7/Dasv6 size documentation · Azure Retail Prices API, East US, read 18 Nov 2026









