According to theCUBE Research, only 5% of global enterprises are “future-built,” that is, they have fully operationalized AI across their business. The performance gap is striking. These future-built enterprises see 1.7× higher revenue growth, 3.6× stronger shareholder returns, and 2.7× greater ROI on AI investments compared to their peers.
In this episode of AppDevANGLE, I spoke with Kevin Kamel, Vice President of Product Management for AI at Mirantis, about why so many organizations struggle to move from AI experimentation to enterprise-scale value and what it really takes to close the AI value gap.
From AI Hype to Enterprise Reality
Kevin framed today’s AI moment as familiar territory.
“Everything that’s old is new again,” he told me. “This feels a lot like the early days of the internet.”
Just as in the dot-com era, enterprises are grappling with what the technology is truly good for, how to innovate on top of it, and how to translate experimentation into durable business outcomes. The difference this time is speed. AI is advancing far faster than earlier platform shifts, compressing learning cycles and amplifying the consequences of early decisions.
A major challenge, Kevin noted, is that enterprises can’t simply “hire their way out” of the problem.
“Organizations have to step back and ask what they need to become responsible for themselves, and what they can rely on vendors to do.”
That question of where ownership belongs sits at the heart of AI value realization.
Why AI Infrastructure Is Fundamentally Different
One of the clearest insights from the conversation was how dramatically AI infrastructure departs from traditional IT models.
AI workloads demand purpose-built environments:
- Specialized hardware such as GPUs and emerging accelerators (TPUs, LPUs, custom ASICs)
- High-performance networking like InfiniBand and RoCE
- Specialized storage capable of feeding massively parallel compute at scale
“These environments are extraordinarily expensive,” Kevin explained, pointing to modern AI racks consuming 150 kilowatts of power per rack. “It forces organizations to think very differently about utilization and efficiency.”
This is where the idea of AI infrastructure as a revenue engine, not just a cost center, becomes critical. Enterprises that treat AI factories as strategic assets rather than experimental labs are the ones beginning to pull ahead.
The Skills Gap Is the Real Bottleneck
While hardware procurement has become increasingly formulaic, Kevin was clear that talent and skills are now the limiting factor. Over the past 15–20 years, many enterprises deliberately moved “up the stack,” outsourcing infrastructure expertise to hyperscalers in order to focus on applications. AI is reversing that trade-off.
“Now we have a new foundational technology where this expertise is needed again inside the enterprise,” Kevin said.
The result is a stratified market:
- Proactive organizations are rapidly rebuilding platform and infrastructure skills in-house to make informed decisions and gain an edge.
- Reactive organizations are following hyperscaler defaults, often without fully understanding the long-term implications.
Kevin’s view was unambiguous: the organizations investing early in platform and AI infrastructure expertise will outperform those that wait.
Security, Data Movement, and Agentic Risk
Security challenges are compounding the skills gap. theCUBE Research shows that 70% of firms protect less than half of their AI-generated data, raising serious readiness concerns.
Kevin compared today’s agentic protocols, such as MCP, to the early days of REST APIs.
“MCP is great at interoperability, but we quickly found it’s also very good at exfiltrating data if you’re not careful.”
As AI agents automate data movement across systems, enterprises need new governance, policy enforcement, and visibility layers. The traditional assumption that data stays where it was created no longer holds. In Kevin’s words, “the network is becoming the AI supply chain.”
Operating AI at Scale With Humans in the Loop
Despite rising automation, Kevin emphasized that human-in-the-loop operations remain essential, especially as AI systems grow more complex.
New categories of operational tooling, often described as AIOps or event intelligence, are emerging to help teams manage this complexity. These systems correlate telemetry across infrastructure, services, and user experience, then surface actionable insights to human operators.
“There’s no way a single operator can manage all of this anymore,” Kevin said. “These systems highlight what matters so humans can focus on outcomes.”
This shift isn’t about removing people from the process; it’s about enabling them to operate at the scale AI demands.
Path to Maturity
For organizations early in their AI journey, Kevin pointed to the importance of honest self-assessment. Mirantis has developed a maturity model and readiness assessment to help enterprises understand where they stand relative to peers and what capabilities they need to build next.
The broader takeaway is clear: AI success is no longer about isolated proofs of concept. It’s about aligning platform strategy, skills development, infrastructure investment, and governance into a cohesive operating model.
Analyst Take
The enterprise AI value gap is not a technology problem; it’s an operating model problem. The 5% of future-built organizations succeeding today are not simply “using more AI.” They have rebuilt internal capabilities around platform ownership, infrastructure efficiency, data governance, and human-in-the-loop operations.
AI infrastructure is evolving into a strategic production system, closer to a factory than a lab. That shift demands new skills, new security models, and new ways of thinking about cost, performance, and risk. Enterprises that continue to treat AI as an experimental overlay on legacy IT will struggle to prove value. Those that reinvest in platform expertise and operational discipline will be the ones that convert AI ambition into measurable business outcomes.
As Kevin put it, this is a land race, and the winners will be determined less by who experiments first, and more by who operationalizes best.
