[Florian Douetot (플로리안 두에토 데이터이쿠 CEO), Dataiku CEO] In June, an overseas government's export-control measure led to a top-tier AI model being fully blocked within hours. That put the spotlight back on the issue of who “owns” the AI companies depend on.

In Europe and Asia, the question of AI ownership has been raised steadily for years among parliaments, corporate boards and experts. In the United States, there is not yet even an established term that clearly defines it, but the essence of the issue is no different. It could be called AI sovereignty or AI independence, but the core question is one. If someone else can stop the AI that supports a company's core operations at any time, can the company really say it owns that AI?

Having a contract with a supplier does not guarantee control over AI. Most companies already rely on outside suppliers for core AI functions and operations. They simply have not yet faced the risks that could arise if they lose that control.

A structure that depends on external AI for the core of corporate operations.

There is a fact most CEOs do not state publicly. Companies think they own AI, but in reality they depend on AI developed by external suppliers. That means suppliers have the authority to design and change the AI models, and as shown by Anthropic's suspension of its Fable 5 service, use can also be restricted at any time by decisions from external actors such as governments.

This risk does not stem only from government decisions and is a problem that has existed for some time. If a supplier raises prices, stops a function a company relies on or changes its product development direction, the company can be easily shaken. That is because the cost of changing existing systems or switching suppliers becomes greater than the cost of maintaining them.

Dataiku's 'Global AI Confessions Report: CEO Edition, 2026', conducted jointly with the Harris Poll on 900 CEOs, shows this reality in numbers. The survey found 70 percent of CEOs worldwide said they are leading AI strategy in their organisations. But only 6 percent said they are involved in most of the key decisions that determine whether that strategy succeeds. That is closer to the delusion of believing they have a strategy than to a real strategy.

That situation is intertwined with an uncertain business environment. In the 2026 KPMG US CEO Outlook Pulse Survey, 65 percent of U.S. CEOs said they lack confidence in market conditions to make major investment decisions. Even so, CEOs see AI investment as an essential task that cannot be delayed. That means core corporate operations depend on an AI foundation they do not control themselves.

CEOs under pressure to deliver AI results.

In the Dataiku survey, 80 percent of CEOs said their role could be at risk if their company fails to produce measurable AI results by the end of this year. Another 77 percent said they expect other CEOs to step down in 2026 due to AI strategy failures or AI-related crises.

CEOs' responsibility for AI results is growing heavier. In Boston Consulting Group's 'CEOs and Boards Survey', about 60 percent of CEOs were found to think boards are pushing too hard to speed up AI transformation. That means they are demanding speed on a foundation that is not sufficiently controlled.

What true AI ownership means.

When most CEOs say they own AI, what they actually mean is infrastructure. Examples include a sovereign cloud, a cloud environment that guarantees control over data and AI operations, self-built local models and whether regulatory requirements are met. That is the easiest area in which to describe ownership, but it is only part of true ownership. Having direct possession of the environment in which AI runs does not show whether people in the organisation properly understand the models inside it, can change them directly when needed or whether the AI performing core tasks still depends on a supplier. Suppliers, in particular, can adjust prices or decide to halt services at any time. Infrastructure is the foundation of AI ownership, but it is not all of it.

True ownership is completed in three dimensions, and infrastructure is only the first. The second is capability. The key is who builds the system, who understands it and who can modify and manage it. If an organisation cannot explain the process by which an AI model reached a particular conclusion, then regardless of what the contract says, the company is merely a user with operating rights. The third is an economic perspective. It means who secures the value created by AI and whether the company can flexibly change direction as markets shift.

Companies' perceptions are already changing. Two years ago, the biggest concern was falling behind in competition, but now 65 percent of CEOs said they are more worried about investing too much in the wrong supplier than about delays in adopting AI. That shows market concerns have shifted completely. Respondents who cited revenue growth as the most important criterion for evaluating AI success rose to 28 percent, a near doubling from a year earlier. That figure is based on markets surveyed in both 2025 and 2026, while the 2026 figure for the full sample was 25 percent.

But if pricing structures are opaque, usage is hard to predict and the cost of switching suppliers is deliberately designed to be high, companies cannot fully secure the value AI creates. Ultimately, the key is choice. The moment a company loses the freedom to change direction, important AI-related decisions are driven by decisions made by external actors.

True ownership of core AI assets.

For companies today, a bigger problem than falling behind in the AI race is treating handing over core control to the outside as the right AI strategy. The article says CEOs worldwide also appear to recognise that risk. In Dataiku's survey, when asked what matters most for AI success, CEOs put governance ahead of people and technology. That means the most important thing in AI innovation today is not technical level or speed, but control. If a company cannot control AI on its own, it is hard to say it owns it. The true value of ownership becomes clearer the moment control is lost.

This ownership issue is not simply about choosing a particular AI model supplier. What matters is shifting the perspective on how to control and use AI. Companies should use the latest AI models where needed, but business logic, organisational knowledge and work processes that determine corporate decision-making and competitiveness should be built on a foundation the company can control directly. They must also be able to review, revise and continuously develop those elements when needed. AI models themselves will become increasingly general-purpose, but the experience and judgement accumulated by organisations are competitive strengths that cannot be easily replaced. What companies need to secure today is not AI technology itself, but their own knowledge and decision-making capabilities accumulated through AI.

Anthropic's suspension of its Fable 5 service was a warning that showed in advance the risks companies can face in the AI era. It should not be viewed only as a geopolitical issue. The core of AI sovereignty ultimately lies in who holds decision-making power over AI. As more models and suppliers emerge, this question will be repeated. The winners in the AI era will not be the companies with the most powerful AI, but the companies that can place AI under their own decision-making.

There is only one AI companies must build. It is AI they can control.