Microsoft Azure Cobalt 100 VMs Momentum and Adoption - Six Five On The Road

How are enterprises navigating the migration to Microsoft Azure Cobalt 100 VMs, and what does this shift mean for the future of cloud-native workloads?

‍

At Microsoft Ignite 2025, hosts Patrick Moorhead and Daniel Newman are joined by Microsoft’s Brendan Burns, Corporate Vice President, Azure Cloud-native and Management Platform, and Arun Kishan, Technical Fellow, Corporate Vice President and GM, Azure Compute, for a conversation on the adoption of Microsoft Azure Cobalt 100 VMs and what users can expect with Azure’s first custom silicon ARM-based virtual machines. They explore Microsoft’s commitment to ARM-based infrastructure, and share real-world use cases and best practices for migration.

‍

Key Takeaways Include:

‍

🔹Evolution of ARM-Based Infrastructure: Microsoft's progressive approach to ARM-based infrastructure, illustrated by the launch and adoption of Cobalt 100 VMs as a milestone for Azure.

‍

🔹Performance & Efficiency: Customers experience significant improvements in performance and energy efficiency by leveraging Cobalt 100 VMs with Kubernetes and containerized workloads.

🔹Customer Adoption & Impacts: How real-world organizations are migrating cloud-native and data-intensive workloads to Cobalt 100 VMs, and observing tangible impact on cost and scalability.

‍

🔹Enabling Modern DevOps: How the ARM ecosystem strengthens DevOps workflows, boosts developer velocity, and enhances open-source collaboration within Azure environments.

‍

🔹Migration Pathways: Microsoft addresses the challenges of x86-to-ARM migration with tools and guidance, facilitating a seamless transition and underlining future innovation in cloud computing.

‍

Learn more at Microsoft

‍

Watch the full video at sixfivemedia.com, and be sure to subscribe to our YouTube channel, so you never miss an episode.

‍

Or listen to the audio here:

Disclaimer: Six Five On The Road is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

‍

Patrick Moorhead: The Six Five is On The Road here at Microsoft Ignite 2025. Daniel, it's great to see you. We are talking about two of our favorite things and that is cloud and semiconductors and it's really been amazing how architectures have changed over the years. It used to be a primarily x86 shop and now we're looking at ARM based infrastructure. We are looking at GPU and XPU based workloads being driven out there.

‍

Daniel Newman: Yeah, it's been a popular topic for us, Pat, for some time now. We've seen the transition going on. A lot of great relationships of course still exist between the hyperscalers and their merchant silicon providers, but we've also seen vertical integration. We're seeing companies continue to want to have more control over their entire, you know, infrastructure stack, trying to build infrastructure that's more efficient, more productive, that works best for their individual workloads and of course delivers the best performance and economics. So this is a great topic, Pat. I'm really excited to dive into this one.

‍

Patrick Moorhead: Yeah. And one of the companies that has taken a huge stand on this obviously is Microsoft. They announced Cobalt a while back and we thought it'd be a great topic to check in, see how things were going, talk about VM momentum and also adoption with the customer set. So with that I'd love to introduce Brendan and Arun. Welcome to the Six5 first time guest. Great to have you on.

‍

Arun Kishan: Thanks for having us.

‍

Brendan Burns: Yeah, thanks.

‍

Daniel Newman: Yeah, it's. Well, you know, we're going to dive right in here. I mean, you kind of heard the setup. Both Pat and I have been talking, by the way. I'm doing a lot of educating, you know, constantly with both the, with the press, you know, with the industry at large. And of course Microsoft's been leading the way in so many parts of this revolution of AI, of infrastructure, of scale build out. And you know, the Cobalt 100 is something I think a lot of people are very interested in. So why don't you just start there, you know, talk a little bit about how your approach to ARM based infrastructure, how it's evolving and you know, why, why is this cobalt 100 VM such a significant milestone for the company? 

‍

Arun Kishan: So I'll, I'll maybe start there talking about that. We did see quite a bit of momentum in the ARM space kind of across other cloud providers and also just broadly in the industry. Right. When you look at mobile and PC and IoT, there is a lot of momentum around ARM. And as we saw some of the growth there in the server space and cloud space. We were thinking, hey, maybe it makes sense for us to sort of start getting into this area. And so we launched a sort of the V5 army arm based family some time back that was with a 3P silicon solution on the AMPERE platform. And in my career I've spent a lot of time working on, you know, operating systems and cloud and things like that. And we've been always doing software kind of at that silicon boundary, working with external vendors like merchant silicon vendors, intel, amd, et cetera. And a lot of the time they're, you're still kind of tied to what they're providing in silicon in order to provide the value right to the workloads and to the customer. And so here, when we looked at the ARM ecosystem, we saw a bit of a unique opportunity to invest in building silicon where we could start taking advantage of the software learnings that we've had, you know, across the OSes, across workloads and go to work, go towards a vertical integration model and deliver capabilities specifically tailored to the kinds of workloads that we're running. So starting all the way down at the hardware, up through the firmware, the OS and then across the first party and third party workloads, build a compelling solution end to end across that, that we can light up in the software workloads at the top. And cobalt 100 was sort of the first step in that journey. And it was our first party, the first system on chip silicon that we built and designed at Microsoft and we launched it last year and it's been a big step up over the prior generation I mentioned. So we launched the V6 VM family. On top of that, it's about 50% better price per performance than the V5 version and a 40% better performance performance per core than the V5 version. So even though it was our first kind of foray into the custom silicon space, it offered quite a big boost to our customers. We've had a lot of interest in this. Overall, about 60% of Azure customers have actually experimented with Cobalt in one way or the other. And today we actually have it deployed in over 29 regions worldwide. And there's a lot of uptake across both first party and third party scale deployments on the platform as well.

‍

Patrick Moorhead: No, that's great. So Arun, you gave some quick facts and figures out there, but I want to pass it off to Brendan. When it relates specifically to, let's say Kubernetes and AKs, can you talk about the performance and efficiency benefits that people should Expect if. In other words, why should they move, you know, enterprise a lot of times is hey, if it's not broken, don't fix it. If not going to see a definitive benefit, why do the work to go in there?

‍

Brendan Burns: Yeah, I mean there's been a long history actually of Kubernetes and cloud native and arm. Back to some of the early demo days we were building out clusters on top of Raspberry PI's sort of hobbyist class ARM devices. But one of the great things that's happened in the container ecosystem actually is that you can build container images that are architecture agnostic or you know, in the, in the old days we'd sort of call them fat binaries where you have the X86 version and you have an ARM64 version. But what that means is as a developer, when you're configuring your application, you don't have to specify.

‍

Patrick Moorhead: Okay, so my work to do was actually a false premise here, right?

‍

Brendan Burns: Well, it's a little bit of both. You just still have to build both images, right. And so there are some modern languages like Go or Rust or NET Java, where it's really pretty much easy. It's the same thing. Like you just sort of change a couple flags and you get the same thing. If you're in a C binary world, you might have to do a little bit more effort. Although honestly I think a lot of our customers are using higher level languages these days. And so then once you do that and you build one of these multi architecture images, it's actually quite easy to have an application that spans both architectures and go where it makes sense to go based on the price performance. Right. And so I think again, like, you know, as you said, like most business users, most enterprise developers, they have an application to deliver at a certain price point, right? They don't care about the architecture as long as you can make it easy for them to adopt it. And that's what we're really trying to do. So when you get Kubernetes in the AKS service, you can have mixed node pools, you can have a combination of, you know, know x86 and arm and schedule across those and then you know, scale in one, scale out the other, depending on the, the performance you need or the, the cost profile you're trying to hit.

‍

Daniel Newman: So we've heard a lot from arm. I mean there's been a lot of conversations across the, you know, data center compute ecosystem. Of course, x86 has a lot of legacy and Azure has invested big to make the pivot to arm. And you know, we're hearing now ARM in some cases, in some clouds, you know, maybe up to half the workloads are, are running on ARM based, which has been an incredible transformation. And again it's not a one partnership is bad. But of, of course you're building that vertical stack to a, to your point earlier and you want to make this stuff run well on, on you know, Microsoft Cobalt. I guess I'm kind of curious though, like you know, you, Microsoft, you know, works with hundreds of thousands if not millions actually of cloud clients, right? And so as you're deploying this, as you're seeing customers adopt Cobalt one hundred, you know, what are you sort of hearing and seeing from them, you know, in terms of the impact that it's having on their cloud, native and data intensive workloads? Because there has to be a reason, right, that they're pivoting and in many cases maybe moving to ARM for the first time or to Cobalt for the first time. 

‍

Arun Kishan: So like I was mentioning before, we've actually seen quite a bit of uptake across third parties and of course first parties as well. So some of the you know, big scale workloads we've gotten include like snowflakes, DataBricks, Amadeus and OneTrust. And sort of what they're seeing, you know, there were two, two aspects, right? We have the price for performance and then absolute performance. And so if we, you know, Amadeus did publish a blog with us about it, but they were seeing response time improvements of like over 8% while they still process 2 billion transactions daily. And that's just from the raw CPU performance benefits. Likewise, OneTrust was seeing about a 50% utilization improvement while still maintaining their KPI. And a lot of that is actually coming from some interesting technical design choices we have. So if you're Familiar with most x64 x86 architectures today, they do SMT or simultaneous multi threading where it's very clever idea, but in a processor you'll sort of use, you'll replicate some of the state, but not all of the state and take advantage of like stock or delays in the pipeline to like run an alternate context. And so you get sort of the benefit of having multiple CPUs without fully investing in a dual CPU stack. And we offer these SSMT processors as VCPUs in the cloud today. And when we went to ARM, we sort of made a decision that we'll provide the same CPU kind of density but do it as separate cores.

And so what you get is you get the full power of a core per VCPU, which is giving some of this higher performance. But what it also means is you're able to push the utilization higher on each of those cores than you can in an SMT situation. Because as you go higher in utilization on smt, you start getting some of the side effects of the fact that they're not fully split state wise. And so that's a lot of the reason that you see these CPU benefits customers are getting. And we see that with the first party workloads also. So for example, Teams IC3, which does our video conferencing and transcode, is very CPU heavy. They also are able to take advantage of this much higher CPU capability that we have on the ARM platform and they're able to run at a higher utilization point, which means they deploy fewer VCPUs, which means lower cost overall. And then likewise the Defender workload, you know, when you visit a website, we'll screen the URL that actually runs through the system today on Cobalt. And they're actually able to get better user latency out of that because it's in the critical path of the interaction. So that's kind of where we're seeing the design choices we made sort of translate to real customer benefit.

‍

Patrick Moorhead: Makes sense. So turning back to you, Brendan, we talked a little bit about the efficiency, performance benefits for Kubernetes and AKs. Let's expand the footprint here, expand the conversation to developers, right? People are pretty comfortable with x86 DevOps, obviously, a huge developer community out there, open source. But how does that translate to arm? Do you have the same ecosystem support? Similar. In some circumstances it could be cleaner and better because it's newer.

Brendan Burns: Yeah, I think interestingly, you know, the same kind of transformation is happening out in the consumer laptop market also, right. So we're seeing an increasing number of ARM 64 based laptops from a variety of different providers. And so actually, I think, you know, developers actually may find it easier to build an ARM 64 container than to build a x86 container because they can build it natively on the hardware that they're using for development. And, and then also I think when you look at more modern languages, right, if you look at, you know, Golang as an example or.net or you know, even TypeScript, you know, those are all languages that make it super easy to target any architecture that you want or at least of the popular architectures. It's quite easy. Net to flip a flag and generate binaries for Both or obviously TypeScript. Is an interpreted language and Node JS runs on both of those different platforms. I think for a lot of developers, it's not really something that they notice and for many developers who've bought a laptop in the last few years may even make their lives a lot easier. And obviously the full container stack that people use for development is all compiled and runs on Native Arm 64 as well. And so I think really most people, it's going to be pretty transparent, they're not going to necessarily notice the difference because both because their development platform has changed, maybe, and also because the languages that they're using at this point expect to have to support both x86 and ARM.

‍

Daniel Newman: So let's kind of round this off with something I'd like to hear from both of you a little bit about. And that's basically the migration. You know, it seems like it's been some years that people are talking about this, but it's still a thing, especially because of the large volume of enterprise customers that you serve. You know, refactoring is a consideration, of course. Microsoft, you know, I think I've made this out a few times. In Azure, you serve all the clients and their needs across the site. Silicon Stack, which is. It is wonderful, but you're also investing heavily not to not offer this to people like, you know, you believe this is probably the best mousetrap in many cases, you know. So when you talk about the opportunities and challenges as it pertains to, you know, this x86 to arm migration, you know, how are you guys making this seamless and not really just for the cobalt 100, but for, you know, really for everything you build going forward. Because this is really not just about one part, it's about what you're trying to build from an infrastructure standpoint. Brendan, I'll start with you.

‍

Brendan Burns: I mean, I think we're invested, you know, as I mentioned, like a lot of them. Even our own hardware for developers to use or tools like Visual Studio, you know, everything that we do, we're targeting it at both, at both platforms. NET is doing a ton of work to make it an easy platform for people to target. Our own cli, Azure cli. I think, broadly speaking, at this point, like, if we're building tools, we're going to build it, we're going to build it for both platforms and have it work exactly the same. And then I think the other thing is, I mentioned that a lot of other languages are relatively easy to bring to arm, but not every piece of code is, and some, especially enterprises Oftentimes have that legacy C or C code that's just kind of been sticking around in the application for a really long time. And there, I think one of the real benefits that's happened in the last couple years has been things like GitHub Copilot. So people who are able to use AI driven tooling to both identify potential problem areas in their code so you can get that C code to compile, but it might not run right because the armament instructions are just different. But you can use AI to identify potential problems and use, and use GitHub Copilot to fix those problems in your source code before you migrate it off to an R64 infrastructure. And then of course we're also really invested in hybrid models where in AKs you can actually have a cluster where you mix and match. So if it's really impossible for you to move one particular piece of your application, you can easily run that in an x86 machine in the same cluster where you're running a lot of your other workloads on ARM 64. And so that means, you know, you don't have to be an all or nothing. It can be a gradual migration and even if there's some stuff you're going to leave behind forever, it still lives in the same kind of environment and the same monitoring infrastructure, the same development and deployment infrastructure. So I think all those things make it easier.

‍

Arun Kishan: Yeah, no, I think aligned with what Brendan was talking about here. I've done a bunch of these architecture ports over time and what we found is the most native code, even CC that's written, you know, to compile on a 64 bit code base generally translates pretty easily over to ARM 64. The compiler and toolchain front ends kind of take care of that for you. There's complexity when it comes to specific micro architectural aspects, like if you're using specific instruction set extensions or low level assembly or memory ordering or things like that, there's small pieces of code that need to be tailored and performance is always going to be a little bit different because of the microarchitectural differences. I think the AI powered tools Brendan was talking about, GitHub Copilot and other transformation tools can really help close the gap on that last mile of code that's difficult to move over. And then a lot of the cloud native ready kind of workloads which are either dynamically translated, like JIT or like. NET or interpreted, there's no effort required at all. There at least from a code porting perspective. Yeah, you'll have to maybe tune and make sure it works properly on ARM, but it's ready to some extent. And also because there has been so much momentum in the industry in the open source ecosystem, a lot of the packages and things on Linux and open source have been moved to ARM and optimized for ARM already. So it's kind of easier to adopt these from an ecosystem perspective. So overall I think this kind of has paved a good path for migrating workloads over to arm on cobalt 100. And then as you've heard about the new platforms coming out, like cobalt 100 was sort of the first step in our journey of doing custom silicon and, and we'll continue to sort of double down on that and innovate to provide the better vertical integration and capabilities for workloads. So you've heard now about the Cobalt 200 at Ignite, which is another step function improvement. It's another 50% better in performance. It has new acceleration capabilities for cryptography and compression which were designed again based on how we know the workloads take advantage of these capabilities so that we can get that better end to end benefit. Now it's still a continuation of the ARM platform. Right. So the workloads that people invest in porting over to Cobalt 100 can be carried over pretty much as is, without significant re engineering or effort required over to this next generation. That will continue to be true as we move forward to future generations of the platform.

‍

Daniel Newman: It sounds really exciting and it sounds like it's going to keep happening fast. I mean, you know, quick last question, just kind of curious, but how, you know, how much, how much acceleration do you guys expect to see now with a client sort of moving? Because it sounds like you've got a high rate of success for your customers trying it. Like, you know, as you move from 100 to 200, I mean, is this the inflection, you know, where you're going to really start to see this pick up? Because it's a big bet. It's not, it's not inexpensive or trivial to be investing in building your own, your own silicon.

‍

Brendan Burns: Yeah, I think it's really going to, I mean, I think it is a great moment and I think the other thing that's happening is, you know, it's not just us, it's, it's an industry moment. So when you look at the latest graphics chips for Nvidia that are coming out, you know, they're running on arm 64 also, right. It's their own. It's, it's not, you know, our silicone, it's their silicone. But, you know, I think there's just momentum in general in the industry. That means that the investments that people make, obviously they'll help them, help them with Microsoft and help them with Azure, but it'll help them in general.

‍

Arun Kishan: Yeah, absolutely. I think any investment customers are making in ARM is leveraged accrues to these AI platforms which are trending to be ARM based. But then, you know, also our general purpose platforms on Cobalt and elsewhere in the ecosystem like we talked about. And also I do see us getting to an escape velocity. Like we've already had a lot of good uptake on 100. People have seen the potential both on the performance and what kind of pricing and efficiency we can offer there. And I think it's only accelerating from there. That was the first in the generation with 200 and beyond. I think you're going to keep seeing that and hopefully we continue to follow through on the promise by showing the differentiation and the capabilities that give that vertical integration benefit and workload light up. And I just expect it to continue to grow from here.

‍

Daniel Newman: Well, listen, congratulations to you on the progress. We continue to be watching this space very closely. It's a bit of a race heating up as everyone is building and then of course everyone is competing and then also collaborating. That's kind of been one of the funnest things to watch the kind of whole ecosystem evolve is, you know, you are in some ways competing and you're continuing to build these giant, you know, these clusters and these clouds where you're running applications on so many different variations of silicon and Microsoft getting, you know, generation to generation. We know from history that these custom chips, each generation leads to better outcomes almost in all cases and continued growth and continued adoption. So, you know, I want to thank you both Brendan and Arun for being part of the show. We'd love to get you back to do another update sometime in the near future, you know, and have a really, really awesome event. It's, you know, this is such a big week every year for Microsoft. So thanks a lot.

‍

Patrick Moorhead: Thank you.

‍

Arun Kishan: Thank you. Thanks for having us.

‍

Patrick Moorhead: Thank you.

‍

Daniel Newman: And thank you everybody for tuning in to this episode of the Six Five. We are on the road at Microsoft Ignite Virtual. We appreciate you tuning in. Be part of our community. Hit subscribe. Check out all of our content from the event and of course all the Six Five content at large for the show for this episode. Time to say goodbye.

‍