Your whole AI stack now runs on one vendor. What happens when you want to leave?

You standardised on one AI provider, and it was the right call at the time. One vendor, one contract, one way of doing things, and you moved fast while everyone else was still deciding. Then, a year on, you notice something. The provider's prices firmed up and you took it. A better model launched elsewhere and you could not really move. Quietly, without a decision being made, your options narrowed. The speed you bought early has turned into a dependence you did not choose.

This is worth understanding before it hardens further, because it is still fixable. Lock-in is rarely a single decision you can point to. It accumulates, layer by layer, until leaving feels impossible and the vendor knows it. The good news is that portability is a choice you can design in, and the bargaining power it buys you is real and lasting. It is really a question of how you adopt AI, which is exactly where most organisations go wrong.

Why is AI lock-in worse than the software lock-in you know?

Because it stacks in more places at once. With traditional software, the switching cost was mostly the data and the retraining. With AI it builds across several layers at once. There is the model itself, the orchestration that connects AI into your workflows, the data you have fed it, the records that prove your governance, and the working knowledge your team has built around one vendor. Each layer alone is a cost. Together they multiply, because unpicking one means touching all the others.

The layer that catches leaders out is orchestration. As you move from a few isolated AI uses to connected workflows where AI hands work between steps, that connective layer becomes the core of how your business runs. It is far harder to leave than a simple model subscription, because it is wired into your operations. This is the same lesson behind treating your own knowledge and architecture, not a single vendor's tool, as the real advantage.

What does keeping your options open actually look like?

It looks like the moves the most capable buyers are already making. Large enterprises are deliberately working with more than one AI provider rather than betting everything on one. ServiceNow signed deals with both OpenAI and Anthropic in early 2026, and other major firms have taken the same multi-provider path, treating optionality itself as a feature worth paying for. They are not being indecisive. They are keeping their pricing power and their freedom to adopt the next better thing.

The lower-altitude question is which AI vendor is best, as though you must pick one and commit. The better question is how to build so that no single vendor owns your future. That means keeping a clean separation between the layer that runs your workflows and the specific model underneath, so you can change the model without rebuilding the business around it. It is the same discipline as playing your own game with AI rather than the game a vendor sets for you.

So how do you build for portability?

You design the freedom to move in from the start, rather than trying to buy it back later at a premium. The aim is to capture the speed of a single vendor without surrendering your bargaining power. Work it in this order:

1. Separate your workflows from the model. Keep the layer that runs your business loosely coupled to any one provider's model, so you can swap the model without rebuilding everything around it.
2. Keep your data yours. Hold your data and its structure in a form you own and can export, not locked inside a vendor's format.
3. Use more than one provider where it counts. For your most important workflows, keep a credible second option live, so the ability to move is real rather than theoretical.
4. Price the exit before you sign. Ask what leaving would actually take at every renewal. If nobody can answer, the lock-in is already deeper than you think.
5. Own the know-how. Make sure your team understands the workflow, not just one vendor's buttons, so the capability stays with you.

What does this make possible?

An AI programme that serves you rather than the other way round. When portability is built in, you keep the upside of moving fast without the trap that usually follows. Your costs stay under pressure because the vendor knows you can leave. A better model arriving becomes an opportunity, not a source of regret. Your data and your direction stay in your hands. That is a far stronger footing than a dependence you did not choose and cannot easily undo.

Picture the next renewal. Instead of accepting whatever terms arrive because leaving is unthinkable, you negotiate from a real alternative, and the conversation changes entirely. When a stronger model launches, you adopt it in weeks rather than wishing you could. You bought the speed of AI early and kept your freedom too, because you designed for it. That combination, momentum without dependence, is what separates the businesses that command their AI future from the ones quietly handing it away.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.