By AI infrastructure teams are being forced to make long-term architectural bets while GPU hardware, networking fabrics like InfiniBand, RoCE, and NVLink, and the model layer all change faster than any planning cycle can accommodate. Organizations that build on monolithic, tightly coupled stacks are discovering that each infrastructure shift requires a full rebuild, creating compounding disruption, technical debt, and cost. The enterprises most exposed are those that treated early AI infrastructure decisions as permanent.
In this interview on TFiR, Richard Borenstein, SVP of Growth and Business Development at Mirantis, walks through how Mirantis built k0rdent AI as a composable, declarative OS for AI infrastructure, what lessons from the OpenStack era apply directly to AI platform engineering today, and how pre-validated reference architectures with partners like Nvidia, Dell, and SuperMicro reduce time to production for global enterprises.
Guest: Richard Borenstein, SVP of Growth and Business Development at Mirantis
Show: TFiR
Here is what every platform engineer and AI infrastructure architect needs to know.
Technical Deep Dive
Q: Why is vendor lock-in a specific and serious risk in AI infrastructure today?
Richard Borenstein, SVP of Growth and Business Development at Mirantis, explains that enterprises recognize the same pattern from previous infrastructure generations: a single monolithic architecture that does not evolve at the organization’s pace and does not respond to operational needs. The AI infrastructure landscape is in permanent beta. No one knows what dominant AI hardware looks like in five years, the model layer is changing faster than infrastructure teams can track, and networking architectures including InfiniBand, RoCE, and NVLink fabrics are all evolving simultaneously. Any infrastructure strategy that requires a full rip-and-replace when one of these layers shifts puts the organization in a constant state of disruption.
“There was a time for monolithic architecture. That time has passed. Now what companies prioritize is composability and long-term flexibility.” — Richard Borenstein, SVP of Growth and Business Development, Mirantis
Q: What is composable declarative architecture and how does it solve AI infrastructure lock-in?
Borenstein describes composable declarative architecture as the ability to swap individual components, such as accelerators, storage backends, or emerging tooling, without rebuilding the entire stack. Mirantis built this principle into Mirantis k0rdent AI from its Kubernetes-native foundation. The design allows organizations to adopt a new accelerator, plug in a different storage backend, or integrate new tooling without starting over, which is a direct requirement for enterprises with compliance, security, and operational continuity obligations that cannot treat infrastructure as disposable.
“Composability is how you give organizations the freedom to evolve without the cost of chaos.” — Richard Borenstein, SVP of Growth and Business Development, Mirantis
Q: Why does Mirantis call k0rdent AI the OS for AI infrastructure?
Borenstein uses the OS analogy deliberately. An operating system does not lock users into specific applications; it enables them. Mirantis k0rdent AI is positioned to function in the same way for AI infrastructure, providing the foundational layer that allows organizations to run, swap, and extend their AI workloads and tooling without being bound to a single vendor’s application or hardware stack. The intent is to act as the enabling layer across the full AI infrastructure estate.
“An OS doesn’t lock you into applications, it enables them. That’s why we call k0rdent AI the OS for AI infrastructure.” — Richard Borenstein, SVP of Growth and Business Development, Mirantis
Q: What lessons from the OpenStack era is Mirantis applying to AI infrastructure today?
Borenstein points to two key lessons. First, the value of human capital alongside the platform itself: enterprises need professional services, managed services, trained and certified personnel, and a partner that supports them at the human level, not just at the technology level. Second, the necessity of designing for continuous evolution. Infrastructure must be capable of changing on a daily basis if required. Organizations that lack the tooling to support that pace end up in long development cycles beholden to multiple vendors.
“We become partners with them not just at the technology level, but we support them at the human level.” — Richard Borenstein, SVP of Growth and Business Development, Mirantis
Q: What is the risk of assembling open source AI infrastructure components without a platform partner?
Swapnil Bhartiya frames the pattern accurately: day zero and day one installation of open source components is straightforward. The compounding operational burden arrives with patching, security management, functionality updates, and scalability requirements. Without the tooling and human support to manage that ongoing complexity, projects fail after the initial deployment phase. This is a known failure pattern from the OpenStack era that Mirantis built its services model around addressing.
“You can install everything, you can download everything, but then you have to deal with all the patches, the security, the functionality, the scalability, and that’s where you get stuck.” — Swapnil Bhartiya, CEO and Co-Founder, TFiR
Q: How are hardware vendors and software platforms converging in AI infrastructure and where does Mirantis fit?
Borenstein describes an accelerating bundled approach in which hardware, GPU, and operations infrastructure cannot be evaluated in isolation. Hardware vendors including Dell and SuperMicro are partnering with Mirantis to ensure customers receive pre-validated, tested, and accredited reference architectures that are ready to deploy. The strategic rationale for hardware vendors is that maximizing customer utilization and satisfaction requires the software and services layers that sit above the hardware. Mirantis positions itself as the AIOS layer that enables enterprises to set up, optimize, and build for the long term regardless of the hardware combination underneath.
“The more simplified we can create that reality for them, the greater benefit we create and the better use case that we expose.” — Richard Borenstein, SVP of Growth and Business Development, Mirantis
Q: How does Mirantis handle global enterprises with sovereignty, security, and compliance requirements across geographies?
Borenstein notes that the enterprises Mirantis serves are global in nature, operating outposts subject to different sovereignty, security, and financial requirements across geographies. Mirantis has built those requirements into the platform’s capability set as a design goal rather than an afterthought. The managed services model extends this further, allowing Mirantis to run and manage the platform day-to-day for enterprises that require operational continuity across distributed, regulation-sensitive infrastructure estates.
“The companies we deal with are global in nature and they have outposts and they’re beholden to certain sovereignty and security and financial requirements that are different across geographies.” — Richard Borenstein, SVP of Growth and Business Development, Mirantis
Q: How does Mirantis approach initial deployment for enterprises that are not yet operating at full scale?
Borenstein is direct that Mirantis does not require enterprises to commit to full-scale infrastructure on day one. The approach is to start with smaller infrastructure footprints to get organizations running, then scale as requirements grow. The underlying customer base consists of some of the largest enterprises in the world with significant infrastructure estates, and the goal is to make complexity manageable at every stage of that growth rather than requiring the full investment upfront.
“We don’t need to boil the ocean on day one. We’re happy to start with small amounts of infrastructure to get you up and running.” — Richard Borenstein, SVP of Growth and Business Development, Mirantis
Resources and Documentation
- Mirantis k0rdent AI, composable declarative OS for AI infrastructure from Mirantis
- Mirantis Managed Services, managed operations and day-to-day platform management for enterprise AI infrastructure
- Nvidia DGX Reference Architectures, pre-validated GPU infrastructure specifications referenced in the interview
- Dell AI Infrastructure Solutions, hardware partner for pre-validated Mirantis reference architectures
- SuperMicro AI Infrastructure, hardware partner for pre-validated Mirantis reference architectures
***
👇 Click to Read Full Raw Transcript
Swapnil Bhartiya: There’s one more concern which is vendor lock in. That anxiety is there, especially in the AI infrastructure space. How does a composable architecture actually protect enterprises from making decisions that, I mean that won’t harm them years from now? It will totally eliminate the whole tribal knowledge and the whole technical debt they will incur otherwise.
Richard Borenstein: Yeah, I mean, I think that is palpable fear that companies have is that they will get locked into that scenario again where they’re beholden to a single monolithic architecture that doesn’t evolve at their pace, that doesn’t respond to their needs. There was a time for that. That time has passed. Now what companies prioritize is composability and long term flexibility. You have to build in iterations, change ads and deletes into your capability set. So composability is the architectural answer to a landscape. In permanent beta, nobody knows what the dominant AI hardware looks like. In five years. The model layer is changing faster than any infrastructure team can track. Networking architectures for AI like InfiniBand or Rocky or NVLink fabrics are all evolving simultaneously. If your infrastructure strategy requires ripping and replacing every time one of these layers shifts, you’re going to be in a constant state of disruption. So that’s what we solved for. We made k0rdent AI the ability to build on a composable declarative architecture, which means organizations can swap components without rebuilding the entire stack. You can adopt a new accelerator, plug in a different storage backend, integrate emerging tooling without starting over. That’s not an accident. It’s a design principle we’ve held since our Kubernetes native foundation. And it matters enormously for enterprises that have real compliance, security and operational continuity requirements, they can’t afford to treat infrastructure as disposable. Composability is how you give organizations the freedom to evolve without the cost of chaos. It why we call k0rdent AI the OS for AI infrastructure. An OS doesn’t lock you into applications, it enables them. No.
Swapnil Bhartiya: Once again, so what said, thank you. I’m also kind of curious looking at your and Mirantis background roots in OpenStack, because the scale at which OpenStack deals with nobody can match. Have you seen any patterns, lessons from that era that you’re bringing into how you folks are building infrastructure for AI today?
Richard Borenstein: Yeah, absolutely. And I wonder if this will surprise you a little bit in terms of the answer. But it’s not just the fact that we’ve built the most capable and robust AI platform for the modern era, it’s that we have the human equity that we apply against it as well. So our professional services and managed services is what we learned. These companies need help and they need handholding at times and they need to be able to have their people trained and certified and have the ability to support these engagements with that human capital. And so that’s a big takeaway for us. Something we’ve been doing really well for, as I mentioned, the Fortune thousand companies that we’ve been so proud to work with for decades plus, is that we become partners with them not just at the technology level, but we support them at the human level. And that allows them to do things like call us in when they need help or, or even use our managed services that actually run their platforms and manage the day to day for them. So that’s a really big one. It’s the combination of that kind of service and support with the platform itself. The other is just the nature of evolution as being something you have to factor in. As I was mentioning earlier, this needs to be able to change on a daily basis if required. And if you don’t have the tooling to do that, then you will be stuck in long development cycles, beholden to a multitude of vendors. Whereas we have aspired to and are proving that we can help companies achieve their corporate goals and their revenue goals by finding the right partner to be able to manage that complexity for them. And it’s not just a local equation. The companies we deal with are global in nature and they have outposts and they’re beholden to certain sovereignty and security and financial requirements that are different across geographies. And so to be able to build that into your capability set is also something we’ve worked hard to do so that we could be that answer on a global basis for the world’s leading companies.
Swapnil Bhartiya: And open source actually make it more tempting that you feel like. Yeah, we can just put it all together. Day zero, day one is the easy part. You can install everything, you can download everything, but then you have to deal with all the patches, you have to deal with security, you have to deal with functionality and the scalability and that’s where you get stuck. And then suddenly your project fails. This may be a bit off topic, but I do want to talk a bit about because you folks are in the infrastructure space, there are a lot of, you know, challenges these days. Of course, hardware is a big challenge. Availability of hardware, cost is big challenge. Now aws, everybody’s kind of working on building their own chips as well. Nvidia is entering the software space as well, where do you see hardware vendors and software platform converge or diverge as AI infrastructure evolve? Where do you see Mirantis in that world?
Richard Borenstein: Yeah, so that is really kind of the model that I envision being accelerated in the near term. And we’re starting to see this bundle kind of approach, which is you can’t look at these pieces in isolation. You have to understand how your hardware works with your gpu, works with your operations infrastructure. And you may have seen Jensen’s five layer cake. You have to actually keep that in mind and think about how you build using that. And what that does is it really gives you some kind of clarity of mission, if you will. To a certain extent, you want to package together the solutions that make the most sense for your organization. And the hardware vendors realize that they need these additional layers and capabilities to ensure that their customers maximize utilization and satisfaction with their products. So that’s why we work together with Dell, with Super Micro, to ensure that when we get in front of the customer, we have a very simple story, which is this is, as I mentioned, with our Nvidia relationship pre validated, we have the reference architectures, it’s tested, accredited and ready to go. And so the more that the hardware vendors can do that, the more they’ll make it easily digestible for their companies to not only use what they’ve bought, but to have the level of satisfaction and confidence to go out and continue purchasing at scale. So it’s an important part of their growth. And that’s where we come in, is we are that AIOS that ensures that you can set it up, optimize, maximize and build for the long term. And we don’t need to boil the ocean on day one swap like we go in there. We’re happy to start with small amounts of infrastructure to get you up and running, but the truth is the companies that we work for are literally some of the biggest in the world that require enormous estates of technology. And the more simplified we can create that reality for them, the greater benefit we create and the better use case that we expose.
