Technical debt, one of those phrases that has escaped from the CIO’s deck and entered the boardroom, distills to sacrificing long-term stability for short-term velocity. Now, with sovereign AI and computing on the front burner, enterprises are rapidly accumulating another kind of debt, one that stems not from age and incompatibility, but from geography and control — and may prove just as consequential.
Sovereignty debt grows whenever a technology decision quietly reduces a company’s freedom of action. It may increase, for example, when data can only live in one place, when a workload depends on one provider or is hard to replicate across geographies or port to another IT environment, or when a particular model is so deeply embedded in a process that replacing it would mean rebuilding the process.
Companies usually make these choices for good reasons, seeking lower costs, faster deployment, greater consistency or simpler management. The trouble comes when an arrangement that once made life easier becomes difficult or prohibitively expensive to change.
Kyndryl’s 2026 State of Modernization Report reveals that companies are increasingly running into this problem. Nearly two-thirds of organizations surveyed expect their use of sovereign cloud to increase, yet only 6% say sovereignty principles are built into their digital transformation efforts from the outset. Meanwhile, 37% report having moved workloads back to mainframes, partly because of sovereignty and governance requirements.
The sovereignty ambition–execution gap
of organizations moved workloads back to the mainframe
of organizations expect their use of sovereign cloud to increase
consider sovereignty principles from the outset of their broader transformation strategy
Navigating instability
Broadly speaking, the push for digital sovereignty is reversing the past two decades of enterprise technology strategy. Companies consolidated data centers, standardized applications, reduced the number of suppliers and moved more workloads onto common cloud platforms.
But concentration can foster dependency and, therefore, vulnerabilities. Important data may sit in one jurisdiction while AI models, cloud platforms and applications depend on others. While enterprises built efficient systems, their resilience became tied to stability.
Stable the world is not. Around the globe, governments are passing new laws requiring customer and business data to remain local. We have already seen governments elect to block access to AI models based on geopolitical concerns, and how, even after access is restored, “nerfed” models can provide very different product capabilities to different users in different parts of the world.
To be clear, none of these developments necessarily requires an enterprise to abandon a supplier or immediately repatriate a workload. The important question is whether the enterprise could do so if circumstances demanded it.
AI makes this problem more urgent because dependency is moving up the stack. A cloud service hosts an application. An AI system increasingly interprets information, recommends actions and, in the case of agents, may act on the company’s behalf. Models may be trained in certain ways due to the location of their creators.
That means sovereignty can no longer be reduced to where data sits. Audrey Tang, in an article published by The Kyndryl Institute, explores the next wave of data sovereignty, which must ask where decision-making lives. When models may subtly influence recommended actions or outputs, it follows that in addition to where decisions are made, how and where a model is trained matters.
Sovereignty debt grows whenever a technology decision quietly reduces a company’s freedom of action.
Keeping options open
Modernization should also be understood as a path to achieving sovereignty objectives, not only as a mechanism to improve efficiency and scale innovation faster.
Data can be made more portable or reside in sharded databases with global coverage. Business logic can be separated from a particular model via post-training or additional steps. CIOs can deploy model “hot-swapping,” allowing for rapid switches of AI tasks to match geography or model provenance.
As AI reduces the cost of writing and fixing software, it can also reduce the cost of planning for sovereignty. AI-powered tools can help organizations map dependencies more quickly and accurately, identifying where they are exposed to particular vendors or where potential challenges exist. AI is also making it faster and cheaper to understand legacy systems, replicate functionality and port applications between platforms or clouds. This makes it easier to replace proprietary components and reduce dependence on any single technology stack.
But businesses must take the time to build a muscle for this new skill. CEOs and CTOs who navigate this transition effectively are pursuing sovereignty as strategy, a continuous process that turns sovereignty principles into the driver of strategic optionality, operational agility and sustainable transformation.
To be clear, few multinationals will make their own chips, build every cloud they use or train every model from scratch. That would be costly and, frankly, unrealistic. Sovereignty is not self-sufficiency or autarky at the corporate level. Rather, it is the preservation of meaningful choice and designing for a future where maintaining this choice is a first-order priority.
The modern-day sovereignty debt crisis emerged from the twin tempests of geopolitical uncertainty and AI capability. And we are still in the very early stages. The silver lining in all of this is that choice is actually code for improved resilience, and more durable businesses are better for customers, shareholders, employees and societies.
