By Enterprises across every industry are racing to stand up AI infrastructure — and the vast majority are making the same miscalculation. They believe that because the components are open source, the stack is free to assemble. It is not. The real cost of the DIY approach is not licensing fees. It is the compounding operational burden: every integration failure, every version conflict, every 3am incident becomes the sole responsibility of the team that built it. And by the time that team finishes building, the underlying technology has already moved on.
The challenge is structural. A production-grade AI infrastructure stack requires dozens of interdependent decisions — GPU operators, network fabric, storage tiers, multi-tenancy architecture, how data science teams access the environment, and how day-two operations are handled at scale. Each of those decisions is effectively a six-month project in its own right. Enterprises that treat this as a DIY exercise are not just taking on technical risk. They are diverting their most skilled engineers away from product differentiation and into permanent infrastructure maintenance.
The speed of change compounds the problem. Open weight models, new GPU generations, evolving inference frameworks, and tightening regulatory requirements around AI sovereignty and GDPR compliance mean that a stack assembled today may be architecturally obsolete within 18 months. Organizations that build from scratch face a brutal choice: rebuild entirely or maintain a deprecated architecture while competitors on managed platforms stay current automatically.
The enterprises that are navigating this most effectively are those that have moved away from the build-versus-buy framing entirely. The more useful question is: where does your organization create differentiated value? The answer is almost never in the plumbing. It is in the models, the data, the applications, and the outcomes. Pre-integrated platforms that abstract away the infrastructure complexity — while preserving the flexibility to choose technologies and enforce sovereignty at the inference routing layer — are increasingly the architecture of record for serious enterprise AI deployments.
Mirantis, with its deep operational history running OpenStack at hyperscale for Fortune-class enterprises and its subsequent evolution into Kubernetes and container orchestration, has positioned its k0rdent AI platform as the answer to this exact problem: open but opinionated infrastructure that lets enterprises stay current without rebuilding from scratch every 18 months.
The Guest: Richard Borenstein, SVP of Growth & Business Development at Mirantis
Key Takeaways
- The DIY AI infrastructure tax is enormous: assembling a production AI stack requires dozens of interdependent decisions, each a potential six-month project, diverting top engineering talent from differentiation to plumbing.
- The real risk of building your own stack is not the upfront cost — it is owning every integration failure, every version conflict, and every 3am incident, with no shared accountability.
- By the time most enterprises finish a DIY AI infrastructure build, the technology has moved on and they are maintaining a deprecated architecture.
- Mirantis’s k0rdent AI platform delivers pre-integrated, certified, and battle-hardened AI infrastructure with built-in inference routing for AI sovereignty and GDPR compliance.
- Mirantis’s OpenStack and Kubernetes heritage — managing the majority of the Fortune 1000 at scale — gives it the operational credibility and playbooks to guide enterprises from day-zero architecture through day-two optimization.
***
Read Full Transcript & Technical Deep Dive
Speaking with TFiR, Richard Borenstein of Mirantis defined the current state of enterprise AI infrastructure adoption — and the dangerous gap between what enterprises want to build themselves and what they are actually capable of sustaining over time.
What Is the DIY AI Infrastructure Tax?
The concept of a “DIY AI infrastructure tax” sits at the center of Mirantis’s market position. Richard Borenstein defines it not as a financial line item but as a compounding operational liability that accumulates across the full AI infrastructure lifecycle — from day-zero architecture through day-two operations and ongoing optimization. This framing directly addresses the misconception that open source components make AI infrastructure free to assemble and maintain.
Q: Why can’t enterprises just leverage open source tooling and wire it all together themselves?
Richard Borenstein: “The AI infrastructure do-it-yourself tax is enormous and most enterprises are still underestimating it. Assembling a production-grade AI stack means making dozens of interdependent decisions. Which GPU operator, which network fabric, which storage tier, how you handle multi-tenancy, how you expose the environment to data science teams, how you handle day-two operations. Each of these is like a six-month project in and of itself. And companies that try to build it from scratch are committing their best engineers to the guts and the plumbing rather than differentiation.”
Borenstein draws a sharp distinction between the perceived freedom of open source assembly and the actual operational cost that accrues when an enterprise owns every layer of a custom-built stack. The framing is explicit: engineering talent redirected to infrastructure maintenance is engineering talent not building competitive advantage.
Pre-Integration as De-Risking, Not Convenience
A recurring theme in Borenstein’s analysis is the reframing of pre-integrated platforms. The conventional enterprise objection to managed or pre-integrated stacks is a perceived loss of flexibility or vendor lock-in. Borenstein inverts this: pre-integration is primarily a risk management instrument, not a convenience feature. k0rdent AI, Mirantis’s AI control plane, is positioned as the embodiment of this philosophy.
Q: What does Mirantis’s pre-integrated AI platform actually provide that DIY cannot?
Richard Borenstein: “What we see with k0rdent AI is that pre-integration isn’t just about convenience, it’s about de-risking the whole process. When a customer deploys our stack, the Nvidia components are certified, the integrations are tested and our professional service teams have battle-hardened playbooks from real deployments. So the alternative, assembling it yourself, means you own every integration failure, every version conflict, every 3am incident. That’s nobody’s problem but yours. The real risk isn’t the licensing cost or even the time to build. It’s that by the time you finish building, the technology has moved on and you’re maintaining a deprecated architecture. Pre-integrated platforms let organizations stay current without rebuilding from scratch every 18 months. That’s what you’re investing in by working with us.”
The OpenStack and Kubernetes Heritage as an AI Infrastructure Differentiator
Mirantis’s credibility in enterprise AI infrastructure is grounded in a specific operational history. The company built its reputation running OpenStack at scale for some of the most performance-intensive infrastructure environments in the world — including large-scale scientific and telecommunications deployments. That heritage informed its subsequent expansion into Kubernetes, container orchestration, and developer platforms, and now directly shapes its approach to AI infrastructure with k0rdent AI.
Q: How does Mirantis’s OpenStack and Kubernetes background change the way it approaches AI infrastructure?
Richard Borenstein: “It really is the benefit of our experience that allows us to have an informed opinion and really migrate our capability set and our knowledge into this very robust platform and capability set. That came from earning a reputation by helping enterprises run OpenStack at scale, particularly in industries with complex and performance-intensive infrastructure demand. We were early adopters of cloud-native technologies, expanded from OpenStack into Kubernetes and container orchestration, developer platforms, full lifecycle management. Throughout that evolution, we’ve remained committed to openness, operational control, and customer choice of technology. And we’ve extended all of that now to meet the emerging requirements around AI infrastructure with a very key focus on digital sovereignty, AI, and modern virtualization.”
Open but Opinionated: The Mirantis Philosophy
Borenstein articulates Mirantis’s product and services philosophy using a specific phrase — “open but opinionated” — that captures the balance between customer flexibility and the firm prescriptive guidance that comes from deep operational experience. This is particularly relevant in the context of AI infrastructure, where the combinatorial explosion of technology choices creates paralysis for many enterprise teams.
Q: How does Mirantis balance giving customers open source flexibility without leaving them to figure it all out themselves?
Richard Borenstein: “The way we look at it, we are open but opinionated. We have great flexibility, but we have deep experience and some very strong opinions in how you do it. But we want the customer to first and foremost have the flexibility to achieve what they want to do. Now, in order to figure out how we can best map to them, we do pretty intensive assessments because we’re not really just selling software or a platform, we’re really building a long-term partnership and often it’s design partnership all the way through to production.”
Managing the Full AI Infrastructure Lifecycle: Day Zero Through Day Three
One of the structural gaps Borenstein identifies in enterprise AI infrastructure programs is the overemphasis on day-zero build activity and the underestimation of what follows. Day-one deployment, day-two operations, and ongoing day-three optimization — iterating on components, evaluating new technologies, managing existing Kubernetes and OpenStack clusters alongside new AI workloads — represent the bulk of the long-term cost and complexity. k0rdent AI is designed to manage all of these simultaneously.
Q: What does Mirantis mean by full lifecycle management for AI infrastructure?
Richard Borenstein: “We’re trying to discuss outcomes with our customers and obviate them from the deep technological education that’s required on a daily basis to maintain all of this — not only for the build, the day zero, but the day one and the day two and the day three: operationalizing, optimizing, iterating and evaluating key components on an ongoing basis. That’s where we built our control plane in k0rdent AI to be able to do all of these things simultaneously, as well as being able to extend and manage existing infrastructure like OpenStack and Kubernetes clusters. As you rise to the moment of AI, you need to be able to manage those both effectively.”
AI Sovereignty, Inference Routing, and GDPR Compliance
For enterprise AI deployments operating across jurisdictions — particularly in Europe and regulated industries — AI sovereignty is not an optional feature. It is a compliance and legal obligation. Mirantis has built inference routing directly into the k0rdent AI platform to address this requirement, allowing organizations to enforce model selection, data residency, and security policies at the infrastructure level rather than relying on application-layer controls.
Q: How does Mirantis address AI sovereignty and GDPR compliance requirements in its platform?
Richard Borenstein: “We’ve also built into the platform the ability to do the inference routing so you can make sure about sovereignty and GDPR and security. And we are kind of the pioneers in that private cloud evolution. So this has given us a very informed capability and this ability to help guide companies to achieve those outcomes through the means of our interactions.”
Enterprise Scale and Fortune 1000 Deployments
Mirantis’s claim to enterprise credibility is grounded in the scale of its customer base. Borenstein references management of the majority of the Fortune 1000 as the operational context within which k0rdent AI’s playbooks and professional services have been developed. This scale creates a self-reinforcing knowledge base: the more large enterprise deployments Mirantis operates, the more battle-hardened its implementation playbooks become.
Q: Who are Mirantis’s primary enterprise customers and what are they asking for?
Richard Borenstein: “We manage the majority of the Fortune 1000, some of the biggest companies in the world, and how do they figure out what to do? There’s the do-it-yourself approach and some people are able to do that, but we want customers to not have to worry about those pieces of the puzzle. We are going to help walk them through the full lifecycle as we’ve done previously, from design and architecture to deployment to operations. And with our track record we’ve got the credibility and the bona fides to do that. So we can talk about what models and where — and then we’ve also built into the platform the ability to do the inference routing.”
Watch the full TFiR interview with Richard Borenstein here
