What it really costs over its life, not what the licence costs. The line item everyone remembers is the per-token price. The lines everyone forgets are integration, evaluation, monitoring, the engineer who maintains it, the human review time, and the migration when the underlying model is retired.
Technical view
Components include inference and infrastructure, embedding and index storage and refresh, engineering for integration and ongoing maintenance, evaluation dataset construction and running, observability tooling, human review labour, security and compliance work, and periodic migration cost driven by model deprecation. Costs scale differently: inference with usage, maintenance with the number of integrations, and migration with the number of tuned components.
Related
Part of AI Terminology Index, section: Product and Adoption.