Fish Audio amid bundled voice stacks

Voice is shifting from a component market into a workflow market. The winning products are no longer simple endpoints that turn text into audio, they are stacks that handle listening, reasoning, speaking, tool use, and deployment in one product. That matters for Fish Audio because a buyer building a phone agent or creator workflow increasingly wants one vendor that ships the whole loop, not a separate TTS API plus extra plumbing.1 2 3 4 5 6

• Deepgram has moved from speech recognition into a bundled agent stack. Its Voice Agent API combines STT, LLM orchestration, and TTS in one WebSocket flow, and even supports Cartesia or other TTS providers underneath. That makes TTS a feature inside a larger agent product, not the product itself.2 3 7
• ElevenLabs is expanding the same way from creator voice tools into a broader platform. Its product set now spans text to speech, speech to text, voice cloning, conversational agents, and a ready to deploy phone receptionist. By June 2026 it had reached an estimated $600M ARR, showing how much larger the bundled opportunity is than standalone voice generation.4 8
• Fish Audio is still small relative to the leaders, with estimated revenue of $21M as of July 31, 2026, versus Deepgram at $100M and ElevenLabs at $600M. That gap suggests the market is rewarding companies that sit closer to the full customer workflow, whether that workflow is creator production, enterprise phone automation, or real time agent infrastructure.1 8 9

The next phase is a land grab for the control layer that owns the live conversation and the surrounding workflow. Voice quality will still matter, but it will increasingly be sold as one module inside broader agent and media platforms. The companies that package voice with orchestration, monitoring, telephony, and business actions will capture most of the value.2 4 5 6