Microsoft Azure Expands Circular Centers: Secure Server Recycling, Cobalt 200 and 92% Hardware Recovery

That sounds like a sustainability-report item, and partly it is. It also touches three things IT pros already care about: capacity, cost and the secure handling of data-bearing hardware.

Two jobs with every hardware refresh​

Microsoft's argument is simple. Each new generation of cloud infrastructure creates two tasks: get more capable systems deployed fast, and deal responsibly with the hardware they replace. Microsoft says the process starts early in the lifecycle, with the design of our silicon, servers, networks, and datacenters. It continues through years of operation. When equipment leaves active service, Microsoft Circular Centers help determine its next useful life.

According to the post, the Circular Centers help us recover components, prepare space for new infrastructure, and get more value from the hardware Microsoft already owns. The middle point matters more than it looks. In a datacenter, floor space and power are the real limits. Old racks have to leave before new ones can go in.

Section summary: Microsoft treats buying new hardware and retiring old hardware as one connected process. The Newport and Sydney openings add Europe and Asia Pacific capacity to that process.

More compute per rack and per watt​

The first half of the post is about density. Microsoft makes these claims:

• About 13 times more cores per rack since 2014
• Roughly 90% less power needed to complete the same task over that period
• A code-compilation example that used about 100 watts on early systems and now runs on less than 10 watts

A caveat belongs next to those numbers. Microsoft attributes them to its own internal analysis, using averaged data across its general-compute infrastructure partners and its Cobalt chips, with performance based on industry benchmarks such as SPEC CPU and what it calls an industry-standard utilization markdown. The post doesn't include workload configurations, test methodology or benchmark results, so nobody outside Microsoft can reproduce the comparison. Read the figures as a fleet-level average, not a promise about your own VM.

The direction makes sense, though. Packing more work into the same power budget is what every hyperscaler is trying to do now that AI demand is pushing datacenter power grids to their limits.

Where Cobalt fits​

Microsoft uses its Azure Cobalt Arm processors as the example of its "systems approach." Because it designs the chip, it can tune the silicon, servers, networking, storage and software together. The lifecycle post repeats that Cobalt 200 delivers up to 50% more performance than Cobalt 100.

Microsoft's separate Build 2026 announcement adds detail:

• Cobalt 200 VMs launched in early access preview. They are aimed at scale-out, cloud-native, Linux-based workloads.
• The chip is Microsoft's second-generation Arm design. It is built on Arm Neoverse V3 Compute Subsystems and made on TSMC's 3nm (N3P) process.
• VMs scale to 128 vCPUs, and memory encryption is on by default through a custom memory controller.
• The 50% CPU figure is a maximum. Microsoft says improvements vary by workload.
• Cobalt 100 is deployed in 32 Azure regions.

For planning, the practical takeaway is that "up to 50%" is a ceiling, not a standard result. If you're considering Arm-based VMs, benchmark your own workloads.

Section summary: By Microsoft's own unaudited figures, Azure hardware has become far denser and more power-efficient since 2014. Cobalt 200 is the latest step, and its headline gains are best-case numbers.

What happens when a server retires​

This is where the post gets interesting for anyone who has decommissioned hardware themselves. Microsoft says it achieved a 92% reuse and recycling rate for decommissioned servers and components "last year." It also admits that at its scale, retiring a server takes more than tossing parts in the right bin.

The process Microsoft describes:

1. Secure decommissioning. Every server is formally taken out of service.
2. Wiping data-bearing components. This happens before anything else.
3. Deciding where the value is. Can the whole system serve another purpose? Can processors, memory and other parts keep working elsewhere? Which materials can go back into the supply chain?

Possible next lives include:

• Microsoft labs and training environments
• Schools and universities teaching technical skills
• Sale to qualified buyers for continued use
• Spare parts for other systems
• Recycling when reuse isn't possible

The post leaves out several things security-minded readers will want to know. It doesn't name the sanitization standard, the verification process for wiped media, or how buyers are qualified. None of that is unusual for a corporate blog post, but it means you should read Microsoft's compliance documentation, not this post, if media disposal is part of your risk assessment.

About that 92%​

The headline number needs context. Microsoft's corporate sustainability page gives the timeframe the blog post doesn't: in 2024, Microsoft reused or recycled 92% of decommissioned datacenter hardware, beating its 90% goal a year early. The same page says more than 3.2 million components were reused that year, which increased the value Microsoft gets from existing equipment by more than 30%.

Two cautions:

• Reuse and recycling are different. A 92% combined rate does not mean 92% of servers got a second life. Some share went back into service and some was recycled, and neither the blog post nor the sustainability page breaks out that split.
• The facility counts tell a timeline. The sustainability page says the Circular Centers program, launched in 2020, had grown to seven facilities. The new post says there are now eight in operation after the Wales and Australia launches. These aren't contradictory; they are snapshots from different dates. The 2024 recovery figures come from before the latest expansion.

Section summary: Retirement comes down to security first, then value. The 92% figure covers 2024 and combines reuse with recycling, so don't read it as a reuse rate.

The network: eight centers, San Antonio next​

Microsoft describes the Circular Centers as a connected global network. With the Newport and Sydney launches, it says eight centers now operate across North America, Europe and Asia Pacific, and a San Antonio, Texas center is planned.

The post gives no launch dates for Newport or Sydney and no opening date for San Antonio. It also mentions automation work "with AI-powered capabilities," including robotic disassembly and autonomous material handling, but gives no numbers on how widely these are deployed or how much they improve recovery.

Why would a company invest in robots that take servers apart? Because at hyperscale, manual teardown doesn't keep up. Refresh cycles are speeding up as AI clusters arrive, so the volume of retired hardware grows too. Microsoft hasn't shown that automation is paying off yet, only that it is investing in it.

What this means for Azure customers and IT admins​

Is this just a sustainability press release? Partly. The post was written to present Microsoft well, and it ends with links to Azure migration content. Still, there are some real takeaways, which are my own analysis rather than Microsoft's claims:

• You don't control any of this directly. Customers can't choose how Circular Centers handle hardware. What you can do is ask your account team for documentation on media sanitization and disposal if your compliance framework requires it.
• Efficiency doesn't guarantee lower prices. Microsoft says circular operations support "cost efficiency and reliability." That is plausible: reusing parts internally can reduce reliance on new procurement for spares. But the post makes no pricing commitments, and you shouldn't assume any.
• Arm is now a core part of Azure's roadmap. Microsoft picked Cobalt as its flagship example of efficiency gains, which shows where it expects future capacity to come from. If your .NET, Java, Python or container workloads haven't been tested on Arm64, the Cobalt 200 preview is a good reason to start.
• Use this as a model for your own hardware retirement. The sequence of decommission, wipe, reuse whole systems, harvest parts and recycle the rest applies just as well to a 40-server on-prem environment as to a hyperscale fleet. Many organizations still skip straight from "old server" to "e-waste pallet."

The bottom line​

Microsoft is presenting refresh and retirement as two halves of one plan: get more compute from each rack and each watt, and get more value from each retired server and component. The Newport and Sydney openings, the planned San Antonio site and the 92% recovery figure for 2024 are the concrete facts here. The density claims are plausible but come from Microsoft's own unpublished analysis.

For Windows and Azure admins, the useful part isn't the recycling percentage. It's the reminder that a server's lifecycle ends with verified data destruction and a documented disposal decision, and that applies to your hardware as much as Microsoft's.