The Base URL Lever: Why Gateway Choice Sets Agency Inference Margin

Agencies that standardize on one OpenAI-compatible gateway turn model swaps into a config change, so per-token cost becomes a negotiable line item instead of a rewrite. The same decision concentrates dependency risk: when a single host stumbles, every client integration behind it stalls. The strategic play is one primary gateway plus a tested fallback endpoint, priced and documented before a client retainer depends on it.