Routing Layer Captures TTS Value

The strategic shift is that the control point in AI voice is moving up from the speech model to the routing layer. Platforms like Retell and Vapi decide which TTS engine handles each call, so Fish Audio is often one slot in a larger stack, not the product a customer is locked into. When buyers can swap vendors for a few cents less cost or lower lag, voice quality alone stops protecting price.1 2 3 4 5 6

• In practice, orchestration means a developer builds the phone agent once, then plugs in providers underneath it. Vapi supports custom TTS, provider keys, and ordered voice fallback, so the platform can keep the call running even if one voice vendor fails. That makes the platform sticky, while the underlying TTS vendor becomes easier to replace.2 3 4
• This routing layer is already where value is accruing. Retell reached an estimated $60M annualized revenue by April 2026, versus Fish Audio at an estimated $21M by July 2026. The larger revenue pool sits with the company that owns call logic, testing, deployment, and monitoring, not just the speech generation step.1 5
• Comparable vendors show how pricing pressure spreads. Cartesia depends on Vapi and Retell for downstream distribution, and those platforms support multiple TTS providers with explicit fallback across vendors. ElevenLabs research also shows customers mixing providers based on latency, cost, and developer experience, with one cited gap showing ElevenLabs at roughly 5x Cartesia per minute.6 7 8

Going forward, standalone TTS vendors will keep getting pushed toward wholesale pricing unless they own a workflow the routing layer cannot easily copy. The winning defenses are deeper integration into enterprise call flows, proprietary voice assets, or a move up the stack into full agent infrastructure, where the customer lives day to day and where switching gets much harder.1 2 5 6