Only a small fraction of global enterprises have crossed the chasm from AI experimentation to real, enterprise-scale value. According to TheCUBE Research, just 5% of organizations qualify as “future-built” – meaning that they’ve fully operationalized AI across their business.
That number alone is eye-opening. But, TheCUBE Research goes on to reveal a performance gap that should keep every CIO and CDO up at night.
Future-built enterprises are seeing 1.7X higher revenue growth, 3.6X stronger shareholder returns, and 2.7X greater ROI on AI investments than their peers. That isn’t a rounding error. It’s an entirely different class of business outcome.
I recently had the opportunity to talk through this reality with the AppDevANGLE team, and the conversation focused on a question I hear constantly: Why do so many organizations struggle to move from AI pilots to enterprise-scale value?
The gap is not primarily an AI model problem. It’s an operating model problem.
It feels like deja vu.
In my view, today’s AI moment resembles the early internet era, when businesses knew something profound was happening but none of us could understand the full impact of what was happening. In those days, companies experimented with websites and early e-commerce not because they had a perfect blueprint, but because they couldn’t afford to miss what was obviously a once-in-a-generation platform shift.
But there is one major difference now, comparing AI to cloud: speed.
AI is moving faster than the internet movement did, faster than cloud did, and arguably faster than mobile did. The technology curve is steep and we’re seeing compressed timelines. You don’t get five years to “wait and see.” Decisions made today about infrastructure, data governance, and platform design can lock in cost structures and operational limitations for years.
That’s why I think one of the most important questions enterprises can ask right now is: What do we need to become responsible for ourselves and what can we reasonably rely on vendors to do?
Organizations can’t hire their way out of this. AI isn’t simply a staffing gap. It’s a responsibility gap. Enterprises need to decide where ownership lives, and do it quickly. The winners will be the ones who understand their AI stack as a production environment, not an innovation side project.
Why AI Infrastructure is Fundamentally Different
One of the reasons so many companies are stuck in “pilot purgatory” is that they’re trying to run AI as if it were traditional IT.
We know from our work with customers that this approach fails.
AI workloads demand infrastructure that departs dramatically from the environments most enterprises built over the last two decades. This isn’t just “compute plus storage.” AI requires purpose-built systems and the expertise to build and operate them. They often include the following.
- Specialized hardware: GPUs, and increasingly other accelerators such as TPUs, LPUs, or custom ASICs;
- High-performance networking: InfiniBand and RoCE are becoming table stakes at scale.
- Specialized storage that can feed massively parallel compute without bottlenecks.[Text Wrapping Break]
This is not theoretical. Modern AI racks can consume 150 kilowatts per rack, and the next-generation Rubin Ultra NVL576 racks are expected to draw nearly 600 kW. At those levels, you’re not just managing IT. You’re managing industrial-grade infrastructure with massive power requirements.
Which leads to the most important mindset shift enterprises must make:
AI Infrastructure is Becoming a Revenue Engine, not a Cost Center
Organizations that treat AI like “labs” tend to optimize for experimentation. That’s fine early on, but it doesn’t scale. The organizations pulling ahead treat AI infrastructure more like a factory: a production system where utilization, efficiency, throughput, governance, and security are all part of the same discipline.
The Skills Gap is now the True Bottleneck
If you ask many enterprises where they’re struggling, they’ll say hardware constraints: “We can’t get GPUs.” That can be true. But it’s also increasingly solvable.
The deeper bottleneck is skills and specifically, platform and infrastructure skills that many enterprises intentionally deprioritized over the last 15 to 20 years.
Cloud changed the enterprise. For a long time, the smart move was to move “up the stack.” Outsource infrastructure complexity to hyperscalers. Focus on applications. That operating model was rational, at the time.
But AI reverses that trade-off.
AI is a new foundational technology and it requires expertise inside the enterprise again. Not necessarily because everyone needs to build infrastructure from scratch, but because enterprises need enough internal capability to make informed decisions, validate vendor approaches, design platforms correctly, and avoid getting trapped in expensive defaults.
This is creating a stratified market:
- Proactive organizations are rebuilding platform competency now by investing in AI infrastructure expertise, platform engineering, data governance, and operational tooling.
- Reactive organizations are following hyperscaler defaults, hoping those defaults will automatically translate into business outcomes.[Text Wrapping Break]
I don’t believe they will. Defaults rarely produce differentiation. And in AI, differentiation is the entire point for competitive advantage.
If enterprises want to close the value gap, they must close the skills gap and that means treating AI platform expertise as a core strategic capability.
Security, Data Movement, and the Rise of Agentic Risk
We’re seeing that at the same time, security complexity is compounding the skills shortage.
TheCUBE Research reports that 70% of organizations protect less than half of their AI-generated data. That should be alarming. AI does not simply “process” data – it creates it, transforms it, amplifies it, and moves it.
And with the growth of AI agents, the risk profile changes again. Agentic protocols such as MCP have some parallels to the early REST API era. They’re great for interoperability and they can become great for data theft, known as data exfiltration, if governance and controls don’t keep pace.
This is why enterprises must revisit a long-standing assumption: that data stays where it was created.
That assumption is broken.
In the age of AI agents, the network becomes the AI supply chain. Enterprises need new governance models, policy enforcement layers, and observability that treats data movement as a first-class operational risk.
Operating AI at Scale Requires Humans in the Loop
Another myth worth dispelling is that AI operations will eventually be “hands off.” In reality, the opposite is happening.
As systems become more complex, the operational burden expands not because people are inefficient, but because humans can’t manually track the full behavior of modern AI stacks.
That is why new categories of operational tooling are emerging, often framed as AIOps, event intelligence, or operational AI. These systems correlate telemetry across infrastructure, services, and user experience and elevate actionable insights to human operators.
This isn’t about removing people from operations. It’s about allowing people to operate at the scale AI requires.
No single operator can manage the entire surface area anymore. The winning operating model will be human-in-the-loop, AI-assisted operations, where automation expands human reach rather than pretending humans can be eliminated from the process.
The Path to Maturity: Assessment, Ownership, Discipline
For enterprises early in their AI journey, one of the most important steps is brutally simple. Be honest about your readiness.
Organizations need clarity about where they are on the maturity curve in terms of what foundational capabilities exist, what is missing, and what the next phase of investment should look like.
At Mirantis, we’ve built a maturity model and readiness assessment framework specifically for this reason. To help organizations measure themselves against peers and develop a roadmap grounded in platform ownership, infrastructure efficiency, operational discipline, and governance maturity.
Because the truth is that AI success is no longer defined by isolated proofs of concept. It’s defined by whether enterprises can build a cohesive operating model that makes AI repeatable, reliable, and accountable.
Closing the AI Value Gap
The “future-built” 5% aren’t winning because they have access to secret models or magical GPUs.
They’re winning because they built the right systems around AI:
- Platform ownership that enables informed decision-making.
- Infrastructure efficiency that turns AI spend into sustainable throughput.
- Governance frameworks that keep data safe as it moves across systems.
- Human-in-the-loop operations that scale reliability and control.
We are all learning quickly that AI is not an overlay on legacy IT. It’s a new production paradigm. This is a land race that will ultimately be won by those who operationalize best. This is an existential opportunity for most organizations to accelerate their timeline towards “future-built” infrastructure so that they can best leverage AI.
