Urgency defined the early stages of the enterprise AI race. Boards applied pressure, the markets priced in AI productivity gains and the main fear was being left behind.

Leaders responded by going deep with whichever AI provider looked dominant, building fast while deferring the harder questions about what they were actually building toward.

The pressure hasn't disappeared, but what’s driving it has changed. Research by AI platform Dataiku and The Harris Poll* shows 65% of CEOs say they worry more about over-investing than under-investing — a reversal from the posture that defined the early stages of the race. Revenue growth now edges out productivity as the leading measure of AI success, a signal that experimentation is no longer the metric boards are watching.

The personal stakes are sharpening alongside it, and 77% of CEOs believe a peer will be ousted this year due to a failed AI strategy or an AI-driven crisis.

"CEO confidence in deploying AI fell even as the investment rose," says Florian Douetteau, Dataiku’s CEO and co-founder. "The more these companies put in, the less certain their leaders became, because each new system showed them how much they did not yet control."

The Structural Risk Nobody Budgeted For

Many organizations locked themselves into vendor relationships that are now difficult to unwind. Pricing is opaque, consumption is unpredictable and capabilities keep shifting.

Companies that went all-in with one provider built their workflows around those capabilities, assuming the relationship would hold. Then a contract renewal arrived, or a model got deprecated, or a competitor shipped something better. What had looked like a vendor decision turned out to be a structural one — "a bit like pouring cement around the furniture," says Douetteau, "then learning the furniture is going to move four times before you finish the house."

Nor is the risk limited to commercial terms. As AI infrastructure becomes more geopolitically sensitive, access can be shaped by regulation, export controls or government action that reaches beyond the vendor relationship itself. A contract can govern price, service levels and usage rights; it cannot fully protect an enterprise from a policy decision that changes who can access a model, where it can be used or under what conditions.

More than three-quarters of CEOs (76%) say their organizations are overly exposed to operational or strategic risk from relying on too few AI vendors, and 67% have questioned or challenged AI vendor decisions made by their CIO or other team members in the past year. Underneath those tensions, the integration burden is compounding: 74% of IT decision-makers say fragmented AI tools are a significant obstacle to scaling, according to a Dataiku/Morning Consult survey.

As AI scaled across the enterprise, the decisions shaping it were scattered across teams, vendors and systems. The accountability, meanwhile, stayed with the CEO. Douetteau points to a telling gap in his own survey data: 70% of CEOs claim ownership of AI strategy, but only 6% are involved in the day-to-day decisions. "That gap is where the dependency accumulates," he says, "because the person who holds the full picture is never the person watching it form."

The Solution: Make AI Flexible By Design And Owned By Default

Rather than engaging in a futile hunt for the perfect vendor, CEOs are increasingly realizing that it's better to preserve the ability to adapt as vendors, models and economics continue to shift — and that keeps what the business has learned, not just what it has licensed.

A vendor relationship gives you access to a capability for as long as the vendor allows it. An orchestration layer above any single provider gives you something different: the freedom to swap a model without rebuilding the work underneath it and the ability to keep the logic, governance and institutional knowledge embedded in those systems under enterprise control.

“The layer above the models and the systems is what lets a company add a vendor, swap a model, connect a new data source and keep its governance and the work already done,” says Douetteau. “We built Dataiku to be that layer.”

Dataiku is a governed AI environment that gives teams one place to build, deploy and adapt AI across the vendors, models and systems they already use, while preserving control and traceability.

The stakes of getting this right extend well beyond operational efficiency, as 81% of CEOs say their AI decisions are already shaping or securing their long-term legacy. The companies that emerge strongest from this period won't necessarily be the ones that moved fastest — they'll be the ones who built systems flexible enough to evolve as the market shifts around them, and legible enough that someone still knows what they're doing a year later.

“The question I would put to any CEO is a narrow one,” says Douetteau. “Which pieces of their company's judgment have they moved into working systems they control? And could they explain, to a regulator or to themselves, what those systems are doing?”

The companies that win this period won’t just be the ones that preserved flexibility. They’ll be the ones who made sure the judgment, governance and workflows their AI depends on remained under their control.

For a deeper look at the survey findings behind this piece, explore the Global AI Confessions Report: CEO Edition .

*Research was conducted online by The Harris Poll on behalf of Dataiku (February–March 2026), surveying 900 CEOs at companies with annual revenue of $500M or more across the US, UK, France, Germany, UAE, Japan, South Korea and Singapore.