OpenAI, Microsoft, Google. Three names that appear in every AI roadmap. And three dependencies that most organisations did not choose — they endured them.
AI vendor lock-in is the situation where an organisation can no longer change its AI vendor without prohibitive operational, financial or strategic cost. It differs from classic lock-in by its speed of formation and its often invisible character until it is too late.
In certain rapid integration contexts, AI vendor lock-in can form in 12 to 18 months. Once integrations are made and teams trained on a specific model, exit cost becomes structurally prohibitive.
AI vendor lock-in is not a future risk. It is a reality already installed in most organisations. It is not an API problem. It is not a cloud problem. It is a structural dependency on a capability you do not control. AI vendor lock-in is faster, deeper and sometimes less visible than classic technology lock-in. It concerns data, models, prompts, workflows, evaluations, team habits, contracts, processes and decisions.
Your organisation can depend on a model whose performance, conditions, limits or behaviours change without internal decision.
Changing vendor does not only mean changing API. It may require recalibrating prompts, evaluations, guardrails, quality processes, training and integrations.
Data, embeddings, fine-tunings, usage histories or configurations can reinforce a capability the vendor controls.
Real cost is not only technical. It is organisational, contractual, operational, human and strategic.
For each structuring AI vendor:
We assess your reversibility and structure your dependency doctrine. Exclusively executive committees.
let's talk