Platforms Bundle Voice AI
Sesame AI
The key dynamic is that voice is becoming a feature inside much larger businesses, not a standalone market with durable pricing power. OpenAI can spread voice model costs across ChatGPT subscriptions, APIs, ads, commerce, and enterprise agents, while Amazon, Apple, and Google can use voice to strengthen devices, search, and cloud relationships. That makes it hard for independent voice startups to win on price alone, because platform players do not need voice itself to carry the full margin.
-
Independent vendors usually need each voice minute to pay for transcription, synthesis, model inference, and margin. Vapi, for example, charges a platform fee on top of telephony, STT, TTS, and LLM costs. Platform companies can hide those same costs inside broader bundles that already monetize elsewhere.
-
The specialists that hold up best are moving up or down the stack. Cartesia bundles STT, TTS, and deployment into one control plane, and Deepgram sells an integrated speech stack with enterprise deployment options. Both are trying to avoid becoming a cheap interchangeable voice endpoint.
-
Even the strongest pure play, ElevenLabs, is expanding beyond raw voice generation into agents, dubbing, marketplaces, and enterprise software. That is the same pattern, voice quality attracts users, but the larger business has to come from owning workflow, distribution, or adjacent products.
The market is heading toward a split. Horizontal voice will be bundled by platforms and priced close to infrastructure cost, while independents that survive will either own a full enterprise workflow, own a regulated deployment niche, or pair voice with hardware and distribution the way Sesame is attempting with companions and smart glasses.