ElevenLabs Vertically Integrating Voice Stack

This is a move to turn a premium voice model into a full voice operating system. If developers buy ElevenLabs for both the voice that speaks and the model that listens, then add its agent layer on top, they stop shopping each piece separately. That matters in a market where teams often swap STT and TTS vendors based on latency, price, and setup friction, and where standalone voice quality is getting easier to copy.1 2 3 4 5

• ElevenLabs has pushed down from TTS into STT with Scribe v2 Realtime, which it says delivers roughly 150ms transcription across 90 plus languages and 93.5% accuracy, and it plugs directly into ElevenAgents. That makes the listening layer a native part of the same workflow as the speaking layer.1 2 3
• The direct target is companies like Deepgram, which also pitch a unified stack of STT, TTS, and voice agent orchestration. The difference is where each started. Deepgram expanded up from transcription, while ElevenLabs expanded down from voice generation and cloning, where it already had developer mindshare and enterprise demand.4 5 6
• This also pressures horizontal agent builders like Retell, Vapi, and Decagon partners, whose value comes from mixing best of breed components. Once ElevenLabs bundles the core models with the agent runtime, some customers will accept slightly less flexibility in exchange for one bill, one API, and fewer moving parts in production.5 7 8

The next phase is a fight over who owns the default build path for voice apps. If ElevenLabs keeps improving latency, compliance, and deployment options while bundling more of the stack, more teams will start with its all in one product first and only unbundle later if they hit edge cases on price or control.2 5 6