CEOs told IBM that AI already makes a quarter of their operational decisions, a figure they expect to hit 48% by 2030. Ask those same executives participating in the IBM Institute of Business Value (IBV) how well they understand the systems behind those decisions, and the confidence dissipates: just 9% say they have an excellent understanding of their AI dependencies, and only 25% report high control over their models. “Vendor lock-in creates imbalance,” says Conor Mlacak, chief information officer of Staples Canada. “Once you’re locked in, you lose leverage.” This blind spot has a price: enterprises reported an average of six AI-related disruptions in the past two years, and 81% say a weeklong outage at their primary AI vendor would be severe enough to halt the business.

The costs compound quietly, too: misaligned data placement when running AI far from where it actually lives, for example, drives token-processing costs up 2.8 times, or roughly USD50 million a year for a USD20 billion enterprise, according to IBM’s modeling. The IBM IBV report frames the real test as a board-level question: can you govern, and move, the data your AI depends on? For a CDO, that’s not somebody else’s homework — the report treats data as the layer every model and application depends on, and only 25% of executives report high control over their AI models, compared with 38% over their data. IBM calls the fix selective sovereignty: invest in control in proportion to business risk, not chase full ownership across every layer.

Download the IBM IBV report to learn how to:

• Own the lever with the most headroom. Only 25% of executives have high control over their AI models, but 38% do over their data. Budget and attention should follow that gap. It’s the one part of the stack a CDO can move without waiting on a vendor.
• Tier control instead of chasing it everywhere. Full sovereignty on commodity AI wastes budget; reserve it for systems that touch revenue, risk, or core operations.
• Treat data placement as a budget line. Misaligned data and model location costs 2.8 times more in token processing. That’s a margin problem, not an IT one.
• Fund flexibility like insurance, not a luxury. 72% of executives would pay 20% more to keep multiple AI vendors on hand. Budget optionality before disruption hits, not after.
• Write exit rights into contracts, then test them. 57% say swapping a core AI model means significant decoupling or a full rebuild, and moving AI training and operational data to a new environment takes 145 days on average. Negotiate portability now, and run the failover drill before a vendor runs it for you.