AI Is Making Enterprise IT More Complicated, Not Less

AI Was Supposed to Clean Up the Mess

Kyndryl surveyed 2,000 business and technology leaders across five continents and 12 industries. The finding: companies adopting AI are spreading their workloads across more infrastructure types, not fewer. Mission-critical applications now sit almost evenly distributed among public cloud, private cloud, mainframes, and on-premises systems.

AI is the top reason companies are increasing their use of mainframes and edge computing. It's the second-biggest driver of private-cloud expansion. And it's the leading reason organizations are upgrading networks and application portfolios.

That overturns a decade of enterprise-tech wisdom. "Modernization" used to mean getting off the mainframe, shrinking the data center, and moving toward cleaner public-cloud architecture. The promise was consolidation. The reality, according to Kyndryl's report published October 1, is fragmentation.

Why Old Systems Are Suddenly Valuable Again

Mainframes provide reliable access to core business data—the customer records, transaction histories, and operational ledgers that AI models need to be useful. Edge infrastructure supports low-latency workloads, which matter when you're running inference at the point of sale or on a factory floor. Private environments offer control over sensitive models and proprietary information.

So instead of replacing these systems, companies are keeping them and adding AI on top. The result is an enterprise technology estate that looks less like a single streamlined stack and more like a loose federation of incompatible systems that somehow have to talk to each other.

AI adoption now outranks cost reduction, legacy replacement, and the shrinking supply of legacy expertise as a modernization priority, according to the report. Companies are apparently willing to tolerate additional complexity if it produces measurable business outcomes.

The Pivot Nobody Saw Coming

This creates an unexpected opening for Kyndryl itself. The company spun out of IBM in 2021, carrying the reputation of a traditional infrastructure outsourcer—the kind of firm you hired to manage the mainframes and data centers everyone was supposedly leaving behind.

Now it's positioning itself as the orchestrator of a deliberately messy, AI-era technology landscape. The pitch is no longer "move workloads off legacy systems." It's "make every layer—mainframe, cloud, edge, and on-premises—work together."

That's a business pivot disguised as a research report. Kyndryl is betting that the next phase of enterprise technology isn't a race to eliminate legacy systems. It's a market for companies that can make incompatible systems behave like one operating environment.

Constellation Research describes the same shift as the beginning of the "enterprise AI rewrite." Companies are moving AI from prototypes into production and evaluating it by growth and business outcomes rather than isolated productivity experiments. Capital One, for example, is using CoreWeave to expand AI across the organization and evaluate AI agents at scale.

What This Means for the Next Decade

The oddest part is that AI—the technology that was supposed to simplify enterprise architecture—is helping make it more complicated. The winners may not be the vendors promising a clean replacement. They may be the ones willing to manage the mess.

For years, the enterprise-tech narrative has been about reduction: fewer vendors, fewer platforms, fewer moving parts. Kyndryl's data suggests the opposite is happening. AI doesn't care whether your data lives on a mainframe or in a Kubernetes cluster. It just needs access.

So instead of choosing between old and new, companies are choosing both. And someone has to make that work. That's the market Kyndryl is claiming—not as a legacy provider clinging to old systems, but as the company that understands why those systems suddenly matter again.

The irony is hard to miss. The technology everyone thought would force a great simplification is creating the conditions for a great complication. And the companies best positioned to manage that complication may be the ones nobody expected.