Microsoft's custom-designed processors are now deployed in more than 100 locations spanning 24 countries, according to a new commercial market intelligence report. But the data, compiled by EJL Wireless Research and listed on ResearchAndMarkets, is not an official Microsoft inventory and may include hardware not yet available to Azure customers.

A Global Footprint, Painted by a Third Party

The report, which surfaced via Quantum Zeitgeist, claims to map Microsoft's proprietary silicon deployments down to the city or state level in select cases. It covers four processor families: Maia AI accelerators, Cobalt Arm-based CPUs, Azure Boost DPUs, and the Majorana 1 quantum processor. While the underlying location-by-location dataset isn't public, the report's description promises a granular view of where Microsoft has placed its own silicon.

That granularity is significant because Microsoft's custom-chip strategy is no longer hypothetical. The company has publicly acknowledged Maia 100 and Cobalt 100 chips, and more recently Maia 200 and Cobalt 200. The report suggests these are now spread across 24 countries, hinting at a rapid and geographically diverse rollout. The list includes both expected hyperscale regions and less obvious locations, though without the raw data, we can't verify specifics.

The Chips Involved: Maia, Cobalt, Boost, and Majorana

Microsoft's custom silicon portfolio serves distinct purposes:

  • Maia AI accelerators: Maia 100 debuted first, but the spotlight has shifted to Maia 200. Microsoft officially launched Maia 200 on January 26, 2026, and positioned it as an inference-focused accelerator built on a 3nm process. It delivers over 10 petaFLOPS at FP4 and supports Ethernet-connected clusters of up to 6,144 accelerators. The report mistakenly lists a February 2026 release date—a discrepancy worth noting.
  • Cobalt Arm CPUs: Cobalt 100 and 200 provide general-purpose compute for Azure virtual machines. Microsoft says Cobalt 200 VMs are in preview in West US 3, East US 2, Central US, Sweden Central, Spain Central, and Indonesia Central. These public regions may represent only a fraction of the deployments claimed in the report, since chips can be installed and reserved for internal services without being exposed to customers.
  • Azure Boost DPUs: These offload storage and networking tasks from host CPUs, improving efficiency for cloud workloads. The report includes package and rack-level illustrations and power comparisons for Boost, hinting at its scale.
  • Majorana 1 quantum processor: Introduced in February 2025 as a research milestone, Majorana 1 is a topological qubit prototype—not a commercial product. Including it in a deployment map is misleading if interpreted as Azure capacity. Microsoft has consistently framed it as a step toward fault-tolerant quantum computing, not a service you can spin up today.

What This Means for Azure Users and Administrators

If you're an IT administrator or developer, the immediate impact is limited. Here's how to read the report based on your role:

For Azure administrators: The report is market research, not an operational planning tool. Track public VM SKU availability through official channels like the Azure updates page. The Cobalt 200 preview regions are a concrete starting point. If the report claims installations in other regions, they might be for internal workloads or future expansion. Don't assume you'll get access soon.

For developers: Maia 200 instances are currently deployed in US Azure regions for internal AI workloads, Azure AI Foundry, and Microsoft 365 Copilot. Microsoft hasn't announced general availability for customer AI workloads on Maia 200 yet. If you're building AI applications, monitor Microsoft's AI infrastructure announcements, but don't bank on undocumented regions.

For everyday Windows users: There's no direct impact—you won't install these chips in your PC. However, the expansion signals Microsoft's deepening control over the cloud backend for services like Copilot, Xbox cloud gaming, and Windows 365. Over time, custom silicon could lead to better performance and lower latency for those consumer services.

The caveat about quantum: Don't let Majorana 1 inflate your expectations. That hardware isn't powering Azure Quantum jobs yet. It's a research prototype, and the report's inclusion likely reflects lab installations rather than production data centers.

How Microsoft Got Here: A Brief Timeline of Custom Silicon

Microsoft's custom silicon journey kicked into high gear after hiring chip architects from Intel, AMD, and Qualcomm in the late 2010s. The public milestones tell a story of accelerating ambition:

  • Late 2023: Microsoft unveils Maia 100 and Cobalt 100, initially for internal AI and cloud workloads.
  • 2024: Cobalt 100 VMs become available, offering Arm-based options for general-purpose workloads.
  • February 2025: Majorana 1 is introduced as a research breakthrough in topological qubits, not a production quantum processor.
  • January 26, 2026: Maia 200 launches officially, targeting inference with a massive leap in performance.
  • Mid-2026: Cobalt 200 VMs enter preview in six Azure regions; the third-party deployment report surfaces, claiming over 100 sites.

This timeline highlights a key shift: Microsoft is no longer solely reliant on Nvidia, AMD, and Intel. While those partnerships remain critical, the company is building a vertically integrated stack to optimize costs, power efficiency, and performance for its own services.

What You Should Do Now

  1. Check official sources first. For any Azure service, the Azure updates blog and regional availability matrices are the ground truth. If a region isn't listed there for a custom silicon VM, assume it's not available to you.
  2. Validate third-party claims. The EJL Wireless report may inspire exploratory conversations, but don't make procurement or architecture decisions based on it. Wait for Microsoft's public announcements before planning migrations.
  3. Watch for Maia 200 availability. If your AI workloads could benefit from the accelerator's 10-petaFLOPS performance, start preparing by testing on available Maia 100 instances and familiarizing yourself with the software stack. Microsoft is likely to expand Maia 200 to additional regions in the coming quarters.
  4. Keep Cobalt on your radar. The Cobalt 200 preview regions are a sign of growing Arm-based options in Azure. If you're evaluating cost-efficient compute, benchmark Cobalt VMs against x86 alternatives; the price-performance gap may narrow as the fleet scales.
  5. Ignore Majorana for now. Unless you're a quantum researcher collaborating with Microsoft, Majorana 1 has no bearing on your Azure strategy.

The Road Ahead

Microsoft's custom silicon deployment map will likely widen significantly in the next 12 to 18 months. The Maia 200 launch came just over two years after Maia 100, and that cadence suggests we'll see Maia 300 sooner than later. Cobalt 200's public preview implies general availability is on the horizon, and Azure Boost DPUs will quietly underpin network improvements across more regions.

For IT pros, the takeaway is clear: Microsoft is building its own hardware muscle, and that will eventually trickle down to more VM choices and specialized capabilities. But for now, separate the third-party hype from what's actually available in the Azure portal. The report offers a useful—if incomplete—view of Microsoft's ambitions, but your operational reality is defined by what's online and supported today.