Most have done so without building the flexibility to adapt when conditions change
Across sixteen countries, a thousand senior executives have quietly revealed a paradox at the heart of the enterprise AI era: organizations are racing to adopt artificial intelligence while simultaneously losing the ability to govern, redirect, or even fully see it. IBM's Institute for Business Value, in collaboration with Oxford Economics, released findings in June 2026 showing that most enterprises are structurally bound to AI vendors they cannot easily leave, operating with dependencies they cannot fully map. The deeper question this raises is not technological but philosophical — whether the pursuit of capability, unaccompanied by the cultivation of control, is a form of progress at all.
- Seventy-one percent of enterprises cannot switch their primary AI vendor if needed — not merely by contract, but because AI has been woven so deeply into operations that extraction would cause collapse.
- Ninety-one percent of executives admit they do not fully understand their own AI dependencies, even as eighty-one percent acknowledge that a single week-long vendor outage would cause severe or critical disruption to their business.
- Data sovereignty requirements are tightening the trap further, with sixty-eight percent of organizations struggling to meet residency obligations across geographies — adding regulatory friction to an already rigid landscape.
- A mere seven percent of organizations have built advanced AI control capabilities, and they are pulling ahead: protecting fifty-five percent more operating profit from disruptions than their less-prepared peers.
- The competitive gap is widening in real time, and the window for building resilient, flexible AI systems is narrowing with every deployment decision made without governance to match it.
A new study from IBM's Institute for Business Value, conducted with Oxford Economics between February and April 2026, surveyed a thousand senior executives across sixteen countries and seventeen industries. Its central finding is unsettling: most enterprises have deployed AI at scale without building the flexibility to adapt when circumstances change.
Seventy-one percent of respondents said they could not switch their primary AI vendor if required — locked in not merely by contracts but by the deep operational embedding of AI systems. More troubling still, ninety-one percent admitted they do not fully understand their own dependencies across vendors, models, and infrastructure. This visibility gap is not abstract: over the past two years, surveyed leaders reported an average of six AI-related disruptions, and eighty-one percent said a seven-day vendor outage would cause severe or critical operational harm.
Data residency and sovereignty requirements deepen the problem. Sixty-eight percent of executives find it difficult to meet geographic compliance obligations, adding regulatory complexity to an already constrained landscape. Interestingly, seventy-two percent said they would accept a twenty percent cost increase to preserve strategic flexibility with their current vendors — suggesting the true cost of lock-in is the loss of optionality, not the price tag.
Most enterprises describe their AI environments as intentionally multi-vendor, but the reality is less deliberate: sixty-nine percent arrived there through independent business unit decisions, not coordinated strategy. The accumulation of decentralized choices has produced fragility dressed as diversity.
Only seven percent of organizations have developed what the study calls advanced AI control capabilities — the ability to adapt data, models, and infrastructure as conditions evolve. These organizations protect fifty-five percent more operating profit from AI-driven disruptions than their peers. The study's message is clear: the enterprises that build resilient, flexible systems now will not merely survive disruption — they will define the next competitive era.
A thousand senior executives across sixteen countries were asked a simple question: if you needed to switch your primary artificial intelligence vendor tomorrow, could you do it? Seventy-one percent said no. They are locked in—not by contract alone, but by the deep embedding of AI systems into the machinery of their operations. This is the finding of a new study by IBM's Institute for Business Value, released in mid-June 2026, and it arrives at a moment when enterprises are racing to adopt AI faster than they can govern it.
The study, conducted between February and April of this year in collaboration with Oxford Economics, surveyed a thousand respondents responsible for AI, data, technology, or related functions across seventeen industries. What emerges is a portrait of organizational vulnerability masquerading as progress. Yes, companies are deploying AI at scale. But most have done so without building the flexibility to adapt when conditions change—when vendors raise prices, deprecate models, impose usage restrictions, or simply fail.
The visibility problem cuts deepest. Ninety-one percent of executives surveyed admit they do not fully understand their organization's dependencies across AI vendors, models, and infrastructure. This is not a minor knowledge gap. It is the difference between being able to assess risk and being blindsided by it. Over the past two years, surveyed leaders reported an average of six AI-related disruptions. Yet eighty-one percent say that a single seven-day outage from a primary vendor would cause severe or critical disruption—the kind that effectively halts operations. The math is stark: most organizations have experienced multiple AI failures already, and they know a longer one could be catastrophic, yet they cannot see the dependencies that would allow them to prevent it.
Data residency and sovereignty requirements compound the problem. Sixty-eight percent of executives say meeting these obligations across different geographies is challenging, creating additional friction when trying to move AI systems or data between environments. This is not abstract compliance theater. It is a real constraint on flexibility, one that grows more complex as regulations tighten and as enterprises operate across more jurisdictions.
Yet there is a counterintuitive finding buried in the data: seventy-two percent of surveyed executives say they would accept a twenty percent cost increase to maintain their current AI vendors if it improved strategic flexibility. This suggests the real cost of lock-in is not primarily financial—it is the loss of optionality, the inability to adapt. Ana Paula Assis, IBM's Senior Vice President and Chair for EMEA and APAC, frames it this way in the study foreword: the stakes are no longer technical but economic. Any loss of control translates directly into margin pressure, compliance exposure, or business disruption.
The study identifies a small group of organizations—only seven percent—that have built what it calls advanced AI control capabilities. These companies design their systems to adapt data, models, and infrastructure as conditions change. The performance gap is substantial: organizations with these capabilities see less downtime and protect fifty-five percent more operating profit from AI-driven disruptions than their peers. This is not marginal improvement. This is the difference between resilience and fragility.
Why, then, do most organizations remain locked in? The answer lies partly in how they arrived at their current state. Seventy-three percent of surveyed companies describe their AI environments as intentionally multi-vendor. But the drivers tell a different story. Sixty-nine percent cite independent business unit decisions—different teams choosing different tools without coordination. Another sixty-nine percent point to geographic necessity. Only fifty-seven percent mention legacy complexity from mergers and acquisitions. In other words, most enterprises have multiple vendors not by strategic design but by accident, by the accumulated weight of decentralized decisions and operational realities.
The study offers no easy solutions, only a roadmap: organizations must strengthen control and oversight as AI adoption expands. The window for doing so is narrowing. The enterprises that build flexible, resilient systems now will outperform those that remain locked in. The rest will face a widening competitive gap, constrained not by the limits of AI itself but by their inability to control it.
Notable Quotes
AI has introduced new forms of dependency that evolve faster than traditional governance, procurement, or technology cycles were designed to handle. The stakes are no longer technical; they are economic.— Ana Paula Assis, IBM Senior Vice President and Chair, EMEA and APAC
The Hearth Conversation Another angle on the story
Why does vendor lock-in matter so much right now? Companies have always had dependencies on their technology providers.
Because the pace of change in AI is unlike anything that came before. A vendor can deprecate a model, change pricing, or impose new restrictions in weeks. Traditional governance cycles—procurement, contracts, technology reviews—operate on quarterly or annual timelines. AI moves faster than the structures designed to manage it.
The study says 91% of executives don't understand their own AI dependencies. How is that possible at this scale?
Because AI has been embedded so quickly and across so many parts of the organization. A data science team in one division uses one model. Engineering uses another. Finance uses a third. No one has a complete map. And the dependencies are not just about which vendor you use—they're about data formats, model architectures, infrastructure requirements. The complexity compounds.
If 72% would pay 20% more for flexibility, why haven't they already switched?
Because switching is not just expensive—it's operationally disruptive. You can't simply move a system from one vendor to another. You have to retrain models, migrate data, rewrite applications. And if you get it wrong, you halt operations. The cost of the switch itself often exceeds the cost of staying locked in.
The study mentions that only 7% of organizations have advanced control capabilities. What do those companies do differently?
They design for adaptability from the start. They use abstraction layers that let them swap vendors without rewriting core systems. They maintain multiple models in parallel. They architect their data pipelines to be vendor-agnostic. It requires discipline and foresight, but it pays off in resilience and negotiating power.
What happens to the 93% that don't build this way?
They become increasingly dependent on their vendors. As AI becomes more central to operations, that dependency becomes a competitive liability. They lose flexibility, they lose negotiating power, and they become vulnerable to disruption. The gap widens.