Mirantis Company Overview

Hey folks. So we're here to talk about mortis and our AI offering today. Uh, I'm Kevin Camel.

I'm on the product side of the house. I've got with me here, Sean O'Mara and Angelica Ambrosio, who will be helping present today. Uh, as part of the presentation, what we're gonna do is first go through a company overview.

We'll then, uh, introduce according ai. We'll go through a case study of what we're doing with one of our customers, Neel, we'll do a product deep dive. We've got a few demos that are mixed in there.

And then we'll talk about what we're gonna be doing next. I'm gonna take a few minutes to talk about the company. Um, for those of you who don't know, Meranti, we're the original private cloud pioneer.

We built and operate some of the largest clouds in the world. Uh, we have a reputation for doing that, and that's the background that we think uniquely qualifies us to basically tackle this new AI infrastructure problem that's in the market. Um, we've been doing this for a long time, and we've been doing this with hundreds of customers over the years.

Um, oh, what did I get ahead of myself here a little bit. Um, for those of you who do know us, uh, you might wonder what we've been up to over the years. Um, since the early OpenStack days, we embraced Cloud native early.

We got involved with Kubernetes. We, uh, shepherded Kubernetes into the CNCF. Uh, in 2019, we acquired Docker Enterprise, and in 2020 we acquired Lens, which is basically a UI for controlling Kubernetes.

Um, around that time, we launched our mosque product, which is, uh, you know, mortis, OpenStack on Kubernetes. And then we also launched MKE, the mortis Kubernetes engine, along with K Zeros, which are our Kubernetes offerings. Um, and now we're focused on enabling the adoption of AI infrastructure for both the enterprise and cloud provider space.

And that's what we're here to talk about today. Um, we have been doing this for some time, so we have customers that are located in something like 85 to 90 countries around the world. We have thousands of customers across these different countries.

We have, uh, teams that are located all around the world. I think we have a physical presence in about 15 different countries, and we have a professional services organization. We're gonna talk more about that in a little bit, um, that are located, uh, in those 15 countries.

There's about a hundred of them out there. Let's talk a bit about accordant ai. So we see three critical challenges that are out in the market today.

Um, the first one is around tech stack complexity. When we talk about tech stack complexity, these ecosystems are single tenant by design. That's the way that they're being offered by the vendors.

And the challenge is, is that for both the cloud providers and for enterprise organizations, they actually need multi-tenant environments. What they need is basically a software stack that lives on top of that hardware ecosystem that enables you to basically offer these multi-tenant environments. Um, the reality is buying the hardware is easy, but operationalizing it through that multi-tenant model is actually pretty difficult.

Um, there's some skill gaps that are out there today. What's ended up happening over the years is that the hyperscalers have effectively vacuumed out all of the talent from the enterprise organizations and from the cloud providers. And, uh, it's made it pretty difficult for people to execute on this stuff themselves.

Mm-hmm. Uh, from an innovation perspective, and, and you know, I know we have some network and storage folks that are in the room here, like network and storage is cool. Again, like there's been some really neat innovations that have happened on that side of the house.

And the problem is from an enterprise and provider perspective is that they don't necessarily have that talent in house anymore to go do that type of work. Um, above and beyond that, you know, people aren't really fluent in what these AI architectures are really about. So A CEO comes in, will basically give some mandate and, you know, maybe this has happened to some of the people in the room here.

Hey, we're gonna go and bring this AI service to market. We want to go to A, B and C that's out there. And the technical team has sort of taken a step back and wondering, well, I mean, great, but like, how exactly do I execute on that?

There's no established playbooks that are out there. It's a largely immature ecosystem. They're really looking for an expert partner who has an experience in this domain who can help them.

Um, from a customer experience perspective, you know, things are different than they were back in like 2010, 2015. There's a context that's out in the market around customer experience and how people consume cloud services. People expect to have, uh, basically a cloud portal where they're gonna come in, they're gonna log in, they'll be able to do IM, and things like this.

They want to be able to come and put their billing information in. They want to be able to come in and basically order different types of services through that particular portal. There's an expectation around having integrated observability.

That's part of it, you know, are the services and the infrastructure that I'm running actually working and doing the things that they want. And people expect that that hyperscaler type experience is gonna be present. Um, they also expect that there's gonna be like a low latency, high availability environment, and that this stuff actually works and does what it's supposed to.

So, you know, there's this context that's out there. And then there's a requirement around operational efficiency. Um, one of the challenges here is that these GPUs, um, that underlie this system, they're pretty expensive.

Um, it, it's not a place where we can basically just take a single customer and go drop it on there, and it's only 5% utilized. There's a loss of efficiency that's here. And so when we're talking to cloud providers, when we're talking to, uh, enterprise organizations, this is like a top concern for them.

They're basically trying to figure out, you know, I've made this huge capital investment in this type of hardware and I want to make sure that I'm actually using it as much as I possibly can. And this means that there needs to be integrated observability as part of this. It means that there needs to be line of sight into like who's consuming what in this particular environment.

Now our solution for this is cord ai, which is effectively turnkey AI infrastructure. Um, the idea here is that this is the platform that public, private, and GPU clouds can use in order to both operationalize and monetize their GPU investments. Um, a key detail here, and Sean's gonna go a lot deeper on this, but one of the primary differentiators for us is that this is a composable architecture.

It's not a static, fully or vertically integrated ecosystem. And, you know, the vendor basically tells you how it's gonna work. One of the requirements that we were talking about is basically that customer experience.

And one of the things that we bring to the table with accordant is this idea of composability. And this extends into a catalog service that exists as part of accordant, where our customers who are then gonna go offer their services to the market can come in and define what their various offerings are gonna be. If somebody basically wants to create some bespoke offerings related to training, they can go ahead and do that, and they can define that, and that could be something proprietary and they can build that in.

And that would basically present itself through that marketplace offering. If somebody wanted to do the same thing for proprietary inference, they could do something similar to that as well. Um, accordant basically provides true multi-tenancy.

Again, Sean will go deeper on how exactly we go about doing that. Um, and we have some interesting features of the platform like configuration reconciliation, whereby we can detect configuration SKU within the ecosystem and then either automatically go fix that for you, um, or notify you that like this configuration SKU has happened. And the idea here again, is to enable you to efficiently operate this, this environment.

Then there's, uh, validated GPU support from the big guys that are out there. And then again, we have this sort of like cloud console type experience where people can come in and do the things that they need to do. And there's also an operator's console where folks can come in and actually stand up the physical infrastructure, um, through bare metal provisioning and things like this where they can onboard GPUs, where they can actually build those different skew offerings such that they can have that, uh, customer experience out there.

Thank you. So do you guys actually offer a catalog of, uh, machine learning capabilities or is that something your, your, your customers provide? Or is that something you're So what we're doing is basically providing a platform, and we do include a canned library, sort of like a starting point of what a marketplace could look like for people to go extend.

Um, but that's really positioned as a starting point. What we expect our customers to do is basically look at that and then say, well actually I would like to offer this particular vector database, or I would like to offer this particular inference platform. And what we're doing is basically providing the ecosystem upon which they can come in and actually like tailor and curate those types of offerings for the market.

And that In, in addition to that, sorry, in addition to that, we also have A number of offerings which we're partnering with some of the top, um, ML machine learning flow tools that are out there. And by building those partnerships, we are able to start to offer our customers a much broader ecosystem that they can choose from. Um, but we simplify that experience for them of getting started.

So we have our own opinionated version of many of these things, but the reality is we're just one vendor and there are hundreds of vendors doing amazing things. So for us it's about how do we bring the whole ecosystem together and give our customers choice, but simplifying the day one and day two operational experience, which is critical because, you know, anyone can install something from helm. It's how do you manage that at scale.

Mm-hmm. Right, right. Thank you.

Is it, is it all based on Kubernetes then? Is that your underlying, um, infrastructure cluster? So we'll touch a little bit on that in the technical depth.

We use Kubernetes as a, as a infrastructure management agent. Mm-hmm. But we do offer virtualization as part of that stack.

Um, and much more Catalogs are they, um, capable of using our a like limit access to some things. Is that down to like an individual catalog item or like an like groupings or Yeah, it is, we're able to do that Highly configurable to what our customers need. I mean, that's basically the difference, uh, that we're bringing to the table here.

Awesome. Yeah, the end goal is really to enable our customers to operationalize that GPU infrastructure as quickly as possible. And then to monetize that, we're gonna talk more about the catalog, but a key thing about the catalog is that, you know, there's a lot of folks that are out there that basically have these GPU is a service type offerings that are there and they're largely non-differentiated.

So the next sort of wave that's gonna hit the market now is people who are building these sort of proprietary offerings on top of that ecosystem such that people will use their services and then do that at a premium. Sure. Uh, let's talk, uh, just high level about the stack.

So, you know, we have this concept of metal to model. What exactly does that mean? Cordon basically handles everything that you need in order to stand up your infrastructure.

Uh, we have the ability basically to provision bare metal. That's through, uh, you know, common mechanisms like metal cubed, redfish, a couple others that are there. Um, we can also stand up infrastructure using, uh, cloud-based APIs as well.

So if somebody wants to go ahead and basically stand this stuff up within AWS or Azure or something like that, we can do that. And then you can optionally use KVM in order to basically carve up that compute into virtual instances. And then we can also automatically configure the networking.

If that's gonna be ethernet or InfiniBand or something like this, we have the ability to automatically go do that. Um, we have the ability to then layer on an operating system upon that. Um, it's basically whichever operating system our customers want to then offer to the general market as part of that, um, we can onboard GPUs and partition those as needed using, you know, popular techniques like MIG and things that are like that from the Nvidia ecosystem.

We can also stamp out either a cluster or a multi cluster type environment, depending on the particular offering that, uh, that particular vendor wants to offer to the market. And then we can layer on whatever services the sort of value adds that that particular cloud wants to bring to the market. So for example, if somebody subscribed into the NVIDIA ecosystem and things like that, they're gonna need to have some table stakes.

Things that are like the Nvidia GPU operator, they may want to actually come and add higher order ecosystem components to it, like Nvidia NIMS and things like this. And what we're doing here through this catalog interface is basically taking these underlying capabilities and then offering a commercial platform through which people can basically offer that out into the market at large. Um, the idea here again is to help a cloud or an enterprise get up and running very, very quickly so that they don't have these sort of like stranded resources just sitting there that they can't actually go operationalize.

Right. Quick question. Yeah.

Brian Martin signal 65. Um, there's a whole pile of GPUs in that picture. Uh, as you build up from the metal infrastructure, are you also configuring the backend RDMA network and segmenting that for your multi-tenancy?

Sean's gonna go a lot deeper on exactly how that works. Short Answers, yes. Excellent.

Thank you Sean. Yes, I look forward to it. Kevin, separate question 'cause this has been bugging me.

Um, so real familiar with Mirantis since the beginning. It was private cloud, but that really kind of meant on-prem cloud or co cola cloud from the beginning. Mm-hmm.

You've said cloud a number of times here. I'm not sure if you're talking about multi-cloud, hybrid cloud, public cloud. Good Question.

So it's, it's actually all, so we're gonna talk about the different deployment models on one of the slides that's coming up. But this is intended to basically support public, hybrid and private cloud offerings, all of them. So someone coming in with, you know, essentially an all public cloud first approach to all everything, it's not a problem.

Yeah. Okay. In fact, one of the demos we'll show you later as an example of leveraging AI on public cloud, um, as a fast start for certain types of organizations.

Great. Thanks. Yep.

I mean, one of the challenges in sort of today's world is that often these are hybrid type environments, especially for the enterprise. They wanna operationalize all of that regardless of where that compute lives. Sure.

Okay, cool. Talk about the customer experience here. So I started, started to talk about a catalog.

What does that look like? Um, so we have this integrated catalog. The idea here is that you can define the infrastructure, you can define the services that are gonna be offered out into the market.

And our customers, basically the operators of this particular service can easily come in and configure this particular catalog. Um, and that's all done through, uh, the underlying interface is effectively yaml, but we have a builder and we're gonna actually demo that for you folks today as to what exactly that looks like. And that builder lives as part of an operator console.

And the idea behind the operator console is that you have this sort of singular interface whereby you can come in and you can stand up that bare metal and it's all with a GUI and things like that. You can decide if you want to virtualize those instances or not. You can define these particular market offerings for what that physical infrastructure is gonna be, and you can also figure out what those services are gonna be that you're gonna offer on top of all of that through a customer facing cloud console.

Um, but again, this end idea is that we're bringing all of the parts necessary in a turnkey solution so that if you have made these types of hardware investments that you basically can install accordant and get to a place where you operationalize that and monetize that. Very, very quickly talk about these deployment models that we talked about it a minute ago. So what we're seeing today is basically two main use cases and then we'll talk about the third one.

Um, so there's basically these sovereign AI use cases. There's also a neo cloud use case where people are basically buying this compute infrastructure, buying the GPUs, and then they're creating these offerings and bringing them out into the market. Um, the United States, it ends up being a little bit more of the neo cloud stuff overseas.

We see a little bit more of the sovereign use case that's there today. Now for some folks for the enterprise, they don't necessarily want to take this and actually offer it to the general market. The idea here is that they're gonna offer this to an internal group of folks that are there.

So, uh, you know, I don't want to name any customers or things like that, but like if I'm a financial services company and you know, I process credit cards and things like this, like, you know, maybe I would buy these GPUs and then I would, you know, operationalize that with something ag agentic or something to this nature. Um, that wouldn't be a general market offering. It would be for my internal, uh, practitioners to use in order to build out value added services to help my credit card customers that would be there.

Now there's a hybrid use case that we've run into, um, and the hybrid use case is basically where folks essentially want to use, uh, the public cloud to host our control plane. And Sean's gonna talk a bit about control plane, which is a very unique thing about what we're bringing to the table. But that control plane will live up in the public cloud in something like AWS or Google.

And the idea here is that's gonna allow you to control and remote all of the stuff that's living inside of your physical data center. And by doing that, we don't need to have any footprint that lives inside of that data center whatsoever. We're just basically gonna run all of of that stuff up in the cloud for the control plane stuff, and then we'll just need access into the data center and the bare metal, uh, provisioning and through the network and things like this so that we can stand all of that stuff up.

Um, in terms of pricing, so two key ways that we're looking at this today. Um, it's still early, we're still learning from our customers, but, um, you know, an OPEX type model, a sort of pay as you go type model. Pretty interesting for folks.

They go ahead, they buy all of these GPUs, they've made this pretty large investment here, they don't have customers yet, and this basically allows them to come in and as they're sort of selling these capabilities to the market and getting it out there, now they've got some revenue coming in. And we would take a, a small piece of that. Um, and the idea here is that, you know, as your business grows, our footprint is basically growing with you over time.

Um, you know, traditional pay as you go type thing. Um, another model, um, that we've run into is basically a CapEx type model. And the idea here is that if I'm making this large expense on these GPUs on this compute infrastructure, what I'll do is basically attach the sale accordant to that.

And that basically is pretty interesting for like telcos and things like this who basically want to depreciate their assets and things like that. They can do all of that upfront and basically account for that at the same time. Is there, is there a migration plan or capability to move between the two as you and your capabilities grow?

I'd imagine there's probably a tipping point where the CapEx becomes better, quote unquote, or more financially doable than the opex. Um, yes. So one of the tricks here is that right now it's still early and people aren't really sure about what their business models are going to be.

Mm-hmm. And this is kind of one of the really unique things about mortis. Again, like the OpenStack pedigree and things like this gives us a lot of insight into how this market is gonna emerge.

Right? Um, our view here is basically that people are gonna start to operationalize this, and then as they start to make money, as the ecosystem is evolving, then yes, they'll probably change what their own commitment into the ecosystem is. In some cases, we may actually be offering fully managed services for them.

They may not even have that expertise in order to do that. But the idea here is that as the market sort of matures and there's more expertise within, uh, the labor market, what they'll do is actually hire their own people to go operate that. And maybe they would want to change the pricing model and things to that effect.

The key bit here is that our heritage basically puts us in a position where we can sort of evolve with the customers. Do you think Your back experience really applies to the AI insanity that exists today? I mean, Yes.

Uh, we see a lot of parallels that are happening. Very Much so. Yeah.

Um, Like what, Uh, well what Are the parallels that you are, are resonate most with you? Yeah, a hundred percent. So, in the early days of OpenStack, and I wasn't actually with motis in the early days, but in the early days, one of the challenges again was this sort of skill gap that was there.

People didn't know how to build a private cloud. They didn't know the different types of services to do that. They didn't know how to operate a cloud internally within an enterprise and things like that.

And they basically had this challenge that they needed to develop the skills within their organization, but they needed to deliver a solution. Today for the enterprise, same kind of a thing is happening here on the AI side. People go ahead.

They don't necessarily know when exactly their chips will get in, right? The chips are gonna be here in three weeks. Well, what are we supposed to do?

Like how exactly are we supposed to support this? The same sort of on demand requirement exists today, and Mirantis basically can service this. There is also the, uh, compute aspect that's part of this ecosystem, which in some cases people have already adopted OpenStack and they're looking for something a little bit more mature.

And again, they'll actually come to us because, you know, they associate Morans with that OpenStack, uh, heritage. Anything you want to add to That? Yeah, I think there's two other things which I'd add and, and Kevin touched on it.

The first one is the operational complexity. When you start talking about environments that are so vertically integrated as many AI is, you really have to have a skill base that understands the full stack because the complexity of virtualization linked to networking. Uh, there was a question earlier about RDMA networks, it's all interlinked.

Um, and these workloads have changed enormously from the abstracted workloads in the cloud native space and the years of running large scale infrastructure, operating it, deploying it, solving the technical challenges is really what we bring to the bring to the party today. Um, more than that, a lot of the technologies that are underneath the hood for OpenStack are the same tech that's underneath the hood within large scale AI infrastructure. And we'll touch more on multi-tenancy later on.

But more and more we're seeing multi-tenancy requirements even inside the enterprise. Yeah. We work with a lot of banking customers and they're coming to us and saying, how do we enable multi-tenancy inside our businesses the same way their cloud provider does so we can better share those resources, better create, um, strict isolation and separation of duties.

So it's all of that experience of running essentially running clouds that we're bringing back into the AI space. Yeah, Thank you. Also somewhat common for prospects to come in and actually already be running OpenStack, they've attempted to do this themselves and now they're like, ah, this isn't really where my main value to the market is gonna be in running this.

This is sort of like a commoditized thing at this point. Why don't I go find the experts in this so that I can focus on like unique value that I can bring to the market? So it's a, it's not like an implicit thing.

It's explicit in those cases where they're basically saying they're in that ecosystem now. Um, so one of the kind of, you know, and this ties into the question there. One of the kind of unique things about mirantis is that we have a broad set of capabilities because we've been servicing the infrastructure market for some time.

Um, we not only have the product, but we actually have an in team security or in-house security team. Um, and that's not something that's just like, oh hey, we have SOC two or something like this. We've actually worked with enterprise customers to develop FedRAMP environments to basically check off those particular requirements for them.

Um, in some cases we've helped to actually run those particular environments. So this is something that like a lot of people in this space just aren't set up to do. Um, from a managed services perspective, again, people make these large investments, they don't necessarily know when the investment's gonna arrive, the chips and things like this.

They may need help actually running those particular environments until they can actually train and onboard staff and get that expertise in house. We can help bridge that. And then for smaller projects and things like that, if somebody wants us to sort of solve a particular scaling problem or, uh, you know, run something for a certain period of time or develop a custom piece of software, we actually have a whole professional services team in-house, the global team that we service the world with.

And that again, is sort of like a unique capability that only Mirantis has. Question Josh, with American Sound, um, does that mean you have some larger channel partners that are augmenting your professional services and managed services? Or are you doing all of that in-house?

Uh, right now we're doing that in-house. Um, but it is something that we're open to working with partners around. Um, there's definitely an opportunity for the partner to basically, you know, add value into the ecosystem.

We would, we would welcome that. Okay. Question before you move off this slide.

Um, Michelin Murphy Worldwide technology, uh, in terms of your FedRAMP, uh, and DOD certifications, um, kind of my ears up, how a lot of the folks that I deal with in the DOD space, um, or in the Fed space, um, require air gap environments and that necessarily means that, uh, cloud access or things that are are cloud native or cloud based, um, often just get yanked right off the table. Can you talk a little bit how you work around that? We do, um, well we actually directly support air gap.

Okay. So we have to, in order to actually work with those particular requirements, we basically have an air gap bundle that we maintain as part of this. Um, that is actually super useful even when there aren't explicit security requirements like listed here.

Um, most people today are running in these sort of closed environments and things like that and they just expect an air gap bundle to be present in order to do a deployment. Mm-hmm. And then of course, that's how we maintain up and upgrade it as well.

Interesting. Okay. Yeah.

Thank you. Okay. My turn now and apparently we're all introducing ourselves today on Guy Courier with futurum.

Um, two questions actually. Uh, first is, um, to what degree is accordant, uh, the mirantis overall stack paradigm, let's say, for everything versus very specifically for ai? 'cause I saw the AI not part of the name and it's just this little bug for it, so I'm wondering if it's a flavor of some new overall structure.

That's question one. So Sean's gonna go deeper on this as part of his section. Um, so I won't belabor it here, but Accordant is basically a, a general product for handling multi cluster, multi-cloud requirements.

So it allows us basically to come in and basically manage what are these like sprawling enterprise estates that you start to find now where there's like thousands of Kubernetes clusters and the platform architects who, you know, are supposed to basically be governing and managing that don't have tools through which to do that today. That's the sense that I've gotten and that's pretty good positioning to my sense, which is there's a lot of things that you're doing and now AI and AI infrastructure is just a huge part of it. So we're subsuming that into our overall.

Yeah, I think, yeah, from our point of view, it's a use case of the bigger pool. The question that we had earlier, which asked about how does OpenStack fit to this picture and how does our experience with private cloud fit to this picture? The reality is it's still an infrastructure play, it's a more complex infrastructure play, but it's still managing of large scale infrastructure and this is what Accordant was built to do.

And then we are able to wrap specific use case capabilities on top of that, that that bring that customer experience that's unique to ai. Maybe it's unique to enterprise, unique to the federal space or the air gap space. I mean, all those banking customers I spoke about are all air gap.

So the, the follow up question is to what degree you called it a product, but to what degree is it, um, defining of what, uh, you are aiming your customer's towards or, and to what degree is it a foundation for, um, the kind of, customization may be the wrong word, but tailoring of the overall solution We should think of packaging here? Um, so accordant basically has a couple different packages that are here. One of them is really designed for the requirements of an enterprise today.

Again, these sort of like sprawling estates where like these clusters are just all over the place. They're up in the cloud, they're on premises and things like that. That's a different set of concerns than what say a neo cloud would have, where actually they're trying to be very deliberate with a greenfield type environment and just making sure that they have everything in place upfront, but there's a common set of requirements that are here.

People want to make sure that they are able to consistently run these environments, that they have unified observability across all of it. That they basically have a, a good understanding as to what the configuration is supposed to be and where things are running out of configuration. Not necessarily that that's a bad thing, but just that they have line of sight against that.

And that those types of capabilities are actually super useful both for an enterprise context and then also if you're building one of these AI clouds, but you would apply the technology slightly differently. Okay. Thanks.

Cool. Um, talk build versus buy. Um, you know, this is a choice that a lot of our customers have to think through today.

Um, one of the really interesting things here is that a lot of this stuff actually commodity things that people have lost track of over time, they don't necessarily know how to do bare metal provisioning. Mm-hmm. You know, there's some teams that are still experts in doing this, but you know, not everybody has those capabilities in house.

Not a lot of unique value in sort of like building that capability from scratch. Um, and Mortis basically has a lot of expertise in things that's like bare metal provisioning and sort of OpenStack and like operating these estates at scale. Um, we bring this economy of scale here to the table and one of the things that our customers really need to think about is that does this really make sense for me to actually go build a solution that does these things?

Or would it make a lot more sense to actually go and find a vendor, you know, tis who basically has already built a a capability that's like this and gives me the degree of customization necessary, the ability to white label the ability to brand the interface so that I can just very quickly go to market. And our thesis is, is that it's actually not a good idea for these folks to actually go build platforms that are like this. That this is something where we can basically take our domain expertise and we can help them move faster and get to market and monetize things.

Um, above and beyond that, this concept of composability that we're talking about, um, really provides a lot of flexibility and optionality for our customers. This is a very fast moving space. Things are changing every day.

It's crazy if you read these articles, there's like a new technology, a new thing that's happening all the time. But composability really allows coordinate AI to do is to sort of keep up with changes that are happening out in the ecosystem. If there's a new training platform, if there's a new inference platform, if your team releases a proprietary inference platform or something like this, it comes very, very easy for us to integrate that directly into the ecosystem and then you can go extend that out to the market.

One of the other things that this really helps you with is around lock-in and things like this. So if you've gone ahead and you're actually working with a particular vendor now you don't have to keep working with them just because your particular AI stack integrated with that. The composability really allows us to come in and basically say, Hey, you know, you no longer wanna work with this particular storage provider or the CSI anymore.

You wanna switch to this other one for whatever reason. That becomes a very, very easy thing within this ecosystem to go swap out. Um, and then of course there's our domain expertise, which we think is, is sort of very, very unique in this place.

We don't think that there's a lot of these folks that are coming to market with if use, again, this is like a very new technology who have this sort of pedigree and domain expertise that we're bringing to the table here. Are you gonna talk more about the open source part of it? Yes.

Yeah. Slide. Okay.

That's later. Okay, thank you. And talk a bit about competitive differentiation.

So, you know, one of the unique things about mirantis again is that we've been here for some time. We actually have a lot of great relationships in both the open source and the vendor ecosystem. So if you need some sort of consulting, if you need help, what's the best approach to this?

We can basically help connect you with partners and things like this. Um, but the other thing is that if you've made choices, if you've really committed to a particular vendor and you're gonna be working with them for five or 10 years, um, one of the things that we are able to do is basically just work with whatever that choice is. So while we may not support it today, we very easily can support it tomorrow through this composable solution.

Now this composability, and I know we've talked a lot about this already and we're gonna keep talking about it, is pretty unique to our solution. We're unaware of any other market offerings that have this degree of configurability that allows you as the operator to just come in and add commercial offerings and to remove them and to tailor that. Um, and you don't necessarily need us to do any of this thing.

There's a user experience that basically allows you to do this stuff yourselves. Um, and then again, sort of going back to this, you know, scale and pedigree, um, we have been in the market for a lot longer than the other folks. We basically have a proven track record of delivering these large scale clouds.

We have the references to prove that. Um, and we think that that basically gives us an edge over other folks today. Talk a bit about one of our customers.

Neville, who's Neville? So Nebel is a sovereign AI cloud that's located in Netherlands. Uh, we met these guys after they ran into a bunch of problems.

So Arnold, who is the CEO of Nebel, basically tried to, tried to do this himself. Um, and when he tried to do this himself, he basically ran into a whole bunch of trouble just with the core tech stack. Just a lot of complexity getting this thing working.

You know, he has all of the compute, he's got the chips, he is, you know, a great relationship with the different providers that are out there, the hardware providers. And, uh, his challenge ended up being that like working with the different drivers, working with the operators and things like that just doesn't work the way that it's supposed to. Requires a bunch of tuning, really, you have to know how this stuff works.

And if you're starting from scratch, you might not have that expertise. Um, tenant isolation ended up being a really big problem for him. Again, these ecosystems are not multi-tenant by design.

They were originally intended for an enterprise market where the enterprise basically was the customer and you would share it with a whole bunch of different teams. But if you're building a cloud on top of this ecosystem, it's a different set of concerns. Like, we really need tenant isolation, we need like true multi-tenancy and that needs to be enforced.

Um, so there were challenges related to that. And then also, um, in order to sort of work around that, one of the things that we sort of see in the market is what people will do is basically stand up lots and lots of different Kubernetes clusters. And that's sort of the work around, right?

Like each customer will have one or more Kubernetes clusters. And on the surface that sounds great, right? Like I've at least isolated at that particular level, um, which is a partial solution.

The challenge ends up being that like, yeah, thousands and thousands and thousands of clusters. There's a lot of operational sprawl that falls from this and like it's just not a scalable solution. Um, people pretty quickly come to the realization that like, that that model of like manually operating all of these clusters, like isn't gonna actually get them to a good place.

I have to imagine in that scenario too, there's a lot of stranded GPU resource, which is a very expensive resource to be inefficiently allocated. Correct? Yep.

Especially when you're trying to create strict multi-tenancy because you know, today you are kind of scaling unit is an eight GPU box. That's, you know, 400, $450,000 worth of hardware, right? Forget about all the networking costs that goes around that.

It's almost a rounding error these days. It's just kind of scary. All right.

Um, so Sean's gonna go through this in his section. He's gonna be doing a deep dive in a minute here. But the idea here is that Horton makes it easy to manage all of those thousands of clusters that we were talking about.

In fact, it can actually create the clusters and do the full lifecycle management of everything that's involved there, including the tear down. Um, we have implicit capabilities as part of the ecosystem for doing that tenant isolation. That's basically like a solved problem.

Um, and then above and beyond that is that Cordem basically enables a degree of automation here, such that you have this operational efficiency. It's a, it's a little bit startling because the technical teams for these types of folks tend to be pretty small. Um, you know, it's not uncommon to basically find that like, Hey, I've got a whole entire data center and I've got like a dozen guys.

That's it, working at my whole entire tech team. Um, so like automation is sort of at a premium here. Um, and the idea here for somebody like Nebel, like the net result is basically this improved operational efficiency.

These challenges are all solved and that they can actually go out and focus on getting their offering into the market. You know, getting monetization going, figuring out how to partner all of these types of things. And that these technical challenges aren't things that they need to go, like solve each one of the individual problems.

We basically take care of all of this for them.