Overview
The cloud-exit metaphor is being applied to artificial intelligence too casually. Frontier-model application programming interfaces create visible, metered cost. Open-weight models appear to offer lower spend, greater control, and a route away from vendor lock-in.
That comparison is incomplete.
For a regulated financial institution, replacing a token bill with graphics processing units does not automatically improve return on investment. It can move costs and accountability into capacity planning, model serving, evaluation, cyber controls, resilience testing, specialist staffing, audit evidence, and incident response.
The Bank of England and Financial Conduct Authority report both rising AI use and rising concern about third-party dependencies, model complexity, hidden models, data risks, resilience, and cybersecurity in UK financial services.1
The Decision
Treat model choice as a risk-adjusted portfolio decision: retain frontier APIs where capability and elasticity justify the premium; use managed or self-hosted open-weight models only where the workload clears business-quality, economics, …