Armonk, New York
Nine out of ten executives at large companies do not actually know which AI systems they rely on. This is not just speculation; it is the main finding from IBM’s Institute for Business Value, which surveyed 1,000 senior executives across 16 countries and 17 industries between February and April 2026. The study, called The Calculus of AI Sovereignty, reveals a governance crisis that is easy to overlook. For companies using IBM Watsonx and other enterprise AI platforms, these findings possess real consequences for operating profit, not only theoretical risk.
The IBM Global Study Reveals Executives Do Not Understand AI Dependencies — And the Numbers Are Damning
The most striking number in the report is 91%. Almost all respondents say they do not fully understand their AI dependencies across vendors, models, and infrastructure. In real terms, this means a chief technology officer at a Fortune 500 company might know which supplier invoices are paid each month, but may not know what would happen to the company’s operations if that vendor changed its pricing, stopped supplying a key service, or had a long outage.
Only 9% of executives surveyed said they had a strong understanding of their dependencies on AI vendors, models, and infrastructure. The other 91% are essentially operating devoid of clear visibility, and problems are already starting to appear.
Leaders surveyed reported an average of six AI-related disruptions over the past two years, mostly caused by vendor services. Still, 81% say a seven-day vendor outage would cause severe or critical disruption, stopping operations. Six disruptions in two years are not a minor issue; it shows a pattern. The fact that executives admit a week-long outage would cripple them, even after experiencing smaller disruptions, suggests that simply being aware of the problem is not leading to action.
Cloud Security Vendor Lock Is Now an Enterprise-Level Balance Sheet Problem
The cloud security vendor lock problem has graduated from an IT procurement headache to a boardroom crisis. 71% of respondents say switching their primary AI vendor or model would be difficult, underscoring substantial operational barriers. More than half of the executives surveyed think the situation is even worse: 57% believe replacing a core AI model would require major changes or a full system rebuild, and 56% say it would take at least six months to move core AI systems and applications to another vendor.
Think about what six months of migration would mean for a bank using AI for fraud detection, or a hospital network using AI to manage patient scheduling. If a vendor raises prices, limits usage, or stops supplying a product, these organizations face a tough choice: pay whatever is asked or deal with months of disruption. Conor Mlacak, CIO of Staples Canada, put it simply: “Vendor lock-in creates imbalance. Once you’re locked in, you lose leverage.”
The financial risk is real. Organizations pay 2.8 times more for token processing when their data is not placed correctly for model execution. This extra cost adds up across all AI workloads, quietly increasing expenses that are rarely noticed as a single line item.
Data Silos and the Illusion of Multi-Vendor Strategy
This part of the study is especially concerning. Many executives think they have solved the dependency problem by using several AI providers, but they have not. Most organizations surveyed—73%—say their AI environments are intentionally multi-vendor, but in reality, this variety is often driven by internal and operational factors rather than careful planning.
Data silos are what create this false sense of security. When different business units choose their own AI tools—which happens in 69% of surveyed organizations—it may look like diversification on paper, but it actually leads to fragmentation. Each unit creates its own dependencies and residency issues. The company ends up with several hidden lock-ins instead of none.
The findings show a growing gap between the widespread use of AI in business operations and the governance needed to manage it. In most companies, governance was designed for procurement cycles that last years. But AI vendor relationships, with their model changes, pricing updates, and access limits, change in just weeks.
IBM Watsonx and the Architecture of Control
This is exactly the kind of environment IBM Watsonx was designed to address. Instead of treating sovereignty as just another compliance requirement, IBM’s approach, explained further through its IBM Sovereign Core platform, makes control a core part of the AI system. The idea is that organizations should be able to change data sources, swap models, and shift infrastructure as needed, without having to rebuild everything.
The IBM study puts forward the idea of “selective AI sovereignty.” This means organizations focus their control efforts on the most important systems, such as fraud detection, risk management, and core decision-making, while allowing greater flexibility in lower-risk areas, such as translation or routine automation. This stratified approach is practical. Full control over every part of the AI stack is not realistic or cost-effective for most companies, but selective sovereignty is.
The performance difference between companies that manage this well and those that do not is large. Organizations with the best AI control protect 55% more operating profit from AI disruptions. Yet only 7% of organizations surveyed have reached this level. That 7% did not get there by accident; it is the result of careful planning made years before the disruptions happened.
AI Sovereignty Study Findings: What Executives Are Actually Willing to Pay
One of the clearest signals in the report is what executives say they would pay for the flexibility they lack now. Seventy-two percent of surveyed executives say they would accept a 20% cost increase to keep their AI vendors if it gave them a more strategic leeway. In other words, most senior leaders would willingly pay a 20% premium to get out of the difficult situation they are in.
This is not simply a prediction about the future. Executives are describing a current problem serious enough to put a price on it. The AI sovereignty study shows that there is already strong demand among executives for flexible, auditable, and portable AI architecture, but there are still not enough reliable solutions available.
Sixty-eight percent of surveyed executives say it is hard to meet data residency and sovereignty requirements across countries, making it complicated to move AI systems or data between environments. For multinational companies operating under the EU AI Act, India’s data localization rules, and US federal AI governance requirements simultaneously, this is not just one problem. There are many overlapping legal rules, and any mistake could lead to regulatory trouble if data crosses the wrong border. Imperative Is Not Awareness — It Is Architecture
The IBM report is helpful, but its biggest value may be in changing how we think about the issue. AI dependency is not simply a technology risk for IT departments to handle. It is an economic problem that should be discussed alongside capital allocation and supply chain resilience.
Companies that see AI architecture as a strategic factor, rather than just a set of separate vendor choices across different business units, will protect more of their earnings when the next disruption occurs. And with an average of six disruptions already reported, another disruption is not a question of if, but when.
The executives who close that 91% visibility gap first will not just be better prepared. They will also have a much stronger competitive position than those still operating without clear insight.
Source: IBM Newsroom
