Cartesia Makes Fish Audio Optional

This shifts Fish Audio from a default building block to an optional add on. Once Cartesia sells a full voice agent stack, the buyer no longer assembles STT, TTS, orchestration, and telephony from separate vendors. That matters because the product decision moves from picking the best voice model to picking the fastest path to a working phone agent, where the bundled provider usually captures the TTS seat and the usage revenue attached to it.1 2 3 4 5

• Cartesia now sells managed agents and telephony alongside its speech models, so TTS is packaged inside a larger workflow. In practice, a developer can connect a phone number, stream calls to an agent, and use Cartesia as the built in voice layer instead of wiring in Fish Audio separately.2 3 4
• This is the same integration move showing up across the market. Deepgram combines STT, TTS, and orchestration in one Voice Agent API at $4.50 per hour, while Retell stays model agnostic and lets customers swap providers. The competitive line is no longer just model quality, it is who controls the end to end agent runtime.5 6 7 8
• Fish Audio is still meaningful as a specialist supplier. It is a 2024 company with estimated revenue of $21M as of July 31, 2026, built around expressive TTS, voice cloning, and multilingual generation. But specialists get squeezed when platforms absorb their core feature into a broader product bundle.1 9

The next phase of voice infrastructure favors companies that own more of the live call stack. As Cartesia, Deepgram, and others keep bundling speech, orchestration, and telephony, standalone TTS vendors will win by becoming clearly better at a narrow job, or risk becoming interchangeable capacity underneath someone else’s agent product.2 5 6 8 9