Provider Pricing Power Is the New Lock-in: What AI Model Monetisation Means for Your Budget

In many AI budgets there is now a cost item that can change without your involvement: access to the model.

I recently described in public how an AI model I wanted to use became noticeably more expensive than its predecessor and dropped out of my flat-rate plan. The model was Claude Fable 5. The post resonated unusually widely. Clearly it touches something many people are feeling right now.

Back then my issue was the price. Now it is something more fundamental: the dependency behind it. Shortly after the price jump the model was unavailable for a while before it came back. That was the real moment for me. Had I built a productive process squarely on it, I would have had little room to switch at short notice.

Why this is not a procurement question

When a single provider decides on price, roadmap, access to the model and terms of use, that affects more than your IT architecture. It affects your budget, your predictability and potentially your ability to deliver.

A purchase can be renegotiated or replaced. Lock-in cannot, at least not in the short term and not without rebuild costs. That is exactly the difference. In my view, provider pricing power is the new lock-in.

Not against providers, but for the ability to switch

This is explicitly not an argument against a particular provider or a particular origin. Many models, including those from the large international providers, are excellent and legitimate options. The problem does not arise from choosing a provider. It arises when there is no real alternative to switch to.

Sovereignty here does not mean self-sufficiency. It means staying able to act when the conditions change.

Three operating models, one decisive question

For running AI in a company there are essentially three models:

• Self-hosted models. You run an open or licensed model on your own or rented infrastructure. Highest control, highest operational effort.
• External models via APIs. You use a provider's models through their interface, ideally with your own contract and your own keys. Plenty of flexibility, a clear dependency on the provider.
• Fully managed platforms. A service provider runs the model and the environment for you. Highest speed, lowest control.

None of these models is right in itself. The decisive question is not which one you choose today. It is: can you switch later without rebuilding your application?

That is the real capability. In technical terms it is model portability: building applications so that the model behind them stays replaceable rather than hard-wired.

What sovereignty actually means

Digital sovereignty is therefore more than the question of where your servers are located. The server location is part of it, but it is not the decisive part. What matters is whether you can switch provider, model or operating model without your operations faltering.

We summarise this in four controllable levers: control over model and cost, control over the keys, control over the data flows and a demonstrable way out. Provider pricing power hits the first lever directly. How these four levers work together, we have set out on our digital sovereignty page.

What you can do now

• Model the switching cost. What would it take to move your most important AI use case to another provider or operating model? If you do not know the answer, you have not yet assessed the risk.
• Make dependencies visible. Which AI services are business-critical today, and what happens if price, availability or terms change?
• Build in replaceability. An abstraction layer between application and model turns a full rebuild into a manageable switch.
• Design for governance from the start. Anyone bringing AI into core processes should provide for clear responsibilities, documented data flows and a rehearsed way out, not just a statement of intent.

Conclusion

The price of a single model is an annoyance. The dependency behind it is a business risk. The difference between the two is the ability to switch.

Sovereignty does not come from finding the perfect provider. It comes from not depending on a single one. So the first question for management, CFO, CIO and CISO is simply this: would you know what switching providers on your AI would cost, and where you would start?