An Old Problem Requires A New Solution 

Traditional file environments were built on a set of assumptions that no longer hold: that data lives in discrete locations, that access is largely localized, that governance can be applied after the fact and that infrastructure can be managed through predictable refresh cycles.

In practice, that led to fragmented, difficult-to-govern environments optimized for capacity rather than usability. Data exists, but it lacks consistency, accessibility and often the context needed to make it useful beyond its original purpose. Data that can’t be reliably accessed or trusted is functionally useless to AI.

For a while, that trade-off was tolerable. The model worked because it was predictable. Organizations could plan around known costs and performance characteristics, so even if the system was inefficient, it remained manageable.

Now, that predictability is breaking down. According to research conducted by my company, 94% of organizations report challenges managing unstructured data, yet only 16% rank it among their top investment priorities. That gap reflects a familiar pattern in enterprise IT: underinvestment in the foundational layer that supports everything else.

AI disrupts that pattern by increasing demand on that layer in ways it was never designed to handle. It requires access that preserves context across systems and locations. Put differently, what was once a manageable inefficiency is now an active constraint, exposing a long-standing blind spot and turning it into a limiting factor for growth.

How AI Pours Fuel On A Fragmented System 

AI drives three simultaneous pressures: a surge in data volume, a sharp increase in access frequency and a much higher requirement for consistency and governance. Each of those is manageable in isolation, but together, they expose the limits of fragmented infrastructure. 

Fragmentation remains the norm, with 79% of organizations in our research reporting inconsistent data access across locations. The average enterprise relies on four separate systems for storage, actively breaking the continuity and context that AI systems depend on. It’s no surprise, then, that 90% of organizations report challenges when trying to scale AI.  

These findings align with broader industry trends. Cisco reported that more than two-thirds of organizations struggle to efficiently access high-quality data, warning that AI is exposing gaps in data visibility and governance that many organizations have historically tolerated.

What’s striking is how often this is framed as a problem of models and talent when, in reality, the culprit is scaling on top of misaligned environments with inconsistent data access and fragmented governance. That creates a ceiling that no model upgrade can break through. When the underlying data lacks consistency and context, even the most sophisticated AI produces unreliable results.

The Confidence Gap 

One of the more revealing dynamics in enterprise AI adoption is psychological. Leaders broadly believe they’re ready, with 70% of respondents saying their infrastructure can support AI at scale. Yet only 26% say they could easily recover from a cyberattack, and 70% took more than a week to fully recover from one. 

This gap between confidence and capability is evident across the industry. IBM’s 2026 Tech Leader Study found that of the 2,000 CIOs and CTOs surveyed, only 11% feel fully prepared for large-scale AI deployment, despite significant investment and growing responsibility for AI systems. Together, these patterns suggest that many organizations are measuring readiness by aspiration rather than operational resilience.

These findings point to a form of operational overconfidence. Infrastructure is often evaluated based on whether it functions under normal conditions, while AI stresses edge cases such as latency and governance gaps, which are exactly where most environments break down. ​

If there’s a single litmus test for AI readiness, it’s this: how well does your data environment perform under pressure? You don’t have an AI strategy if your data can’t survive a bad week. 

From Capacity Thinking To Data Value Thinking 

It doesn’t matter as much where your data lives or how much an organization has as whether it can be used and trusted in real time. This shifts the focus from storage as a function of volume to data as an operational asset.

Although AI is now the top IT investment priority for 59% of organizations, infrastructure strategies remain tied to older models built around expansion and refresh cycles. Those models assume data can be organized and governed in discrete intervals, but AI operates continuously, exposing the limits of that approach almost immediately. Clearly, there’s a mismatch between how data is managed and how it is expected to behave.

Organizations with fewer systems and more centralized environments are often able to recover faster from disruptions and deliver more consistent performance across locations. This means shifting from managing storage capacity to managing data utility, designing environments where data remains usable wherever it is created or accessed. ​

Until that shift happens, AI progress will remain constrained by the same issue, and companies will continue investing in increasingly sophisticated tools, only to find them constrained by the same underlying limitations. Luckily, this represents a clear opportunity for those willing to rethink the foundation on which it depends.​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?