The early stages of the enterprise AI race were marked by urgency. Boards of directors pushed for quick adoption, markets anticipated AI-driven productivity gains, and the primary concern was keeping pace with competitors. Leaders quickly aligned with dominant AI providers, prioritizing speed over clarity on future goals. While pressure remains, the underlying factors have shifted.

A study by AI platform Dataiku and The Harris Poll reveals that 65% of CEOs now worry more about over-investing than under-investing, a stark change from the initial stages of AI adoption. Revenue growth has now surpassed productivity as the main metric for AI success, highlighting a shift in boardroom expectations toward tangible outcomes rather than mere experimentation. The personal stakes for leaders have intensified, with 77% predicting a peer’s departure due to unsuccessful AI strategies or crises within the year.

CEO confidence in deploying AI fell even as investment rose, says Florian Douetteau, Dataiku’s CEO and co-founder. The more these companies put in, the less certain their leaders became.

Unforeseen Structural Risks

Many companies have become entrenched in vendor relationships that are difficult to alter. Pricing remains unclear, consumption varies, and capabilities frequently change. Organizations that committed to a single provider based their workflows on those capabilities, anticipating stability. However, when contracts came up for renewal, models became obsolete, or superior offerings emerged, the reality of undesired dependencies surfaced. Douetteau likens this to pouring cement around the furniture, then learning the furniture is going to move.

The risk extends beyond commercial terms. As AI infrastructure gains geopolitical significance, access is subject to regulatory influences, export controls, and governmental actions beyond vendor agreements. Contracts dictate pricing and service but cannot shield enterprises from policy changes affecting model access and usage conditions.

Over three-quarters of CEOs, about 76%, feel their organizations are too reliant on a limited number of AI vendors, and 67% have questioned vendor decisions made by their team. Fragmented AI tools further complicate matters; 74% of IT leaders report this as a significant barrier to scaling efforts.

As AI has spread across enterprises, decision-making is divided among teams, vendors, and systems. CEO accountability, however, remains centralized. Douetteau highlights a discrepancy: 70% of CEOs claim ownership of AI strategy, but only 6% engage with day-to-day decisions. That gap is where dependency accumulates, he notes.

Creating Flexibility through AI Design

CEOs are shifting away from seeking the perfect vendor, focusing instead on maintaining adaptability as vendors, models, and economics evolve. A vendor relationship offers access to capabilities while permitted, but an orchestration layer across providers offers the advantage of switching models without losing underlying work, retaining enterprise control over logic and governance.

The layer above models and systems lets companies add vendors, swap models, and connect new data sources while keeping governance intact, Douetteau explains. Dataiku is built to be that layer. This governed AI environment allows teams to build, deploy, and adapt AI across existing vendors and systems, ensuring control and traceability.

Ensuring flexibility and retaining control over AI judgment and workflows are critical. 81% of CEOs recognize that their AI decisions will shape their long-term legacy. Success in this period will come from building adaptable and understandable systems.

The essential question for any CEO, Douetteau poses, is which aspects of their company’s judgment have they integrated into systems they control, and can they explain these systems’ operations?

Companies that excel during this period will not only retain flexibility but will also ensure their AI systems remain under their governance and judgment.

For an in-depth view of the survey results, see the Global AI Confessions Report: CEO Edition.

*Research was conducted online by The Harris Poll for Dataiku, surveying 900 CEOs from companies with annual revenues of $500M or more across the US, UK, France, Germany, UAE, Japan, South Korea, and Singapore between February and March 2026.