The AI Stack
Sign in

Deepgram, Inc.

Provides APIs for speech-to-text, text-to-speech, and real-time voice agents.

Updated April 2026

Overview

Founded
2015
Headquarters
San Francisco, CA, US
Segment
Audio & Speech

Product overview

Deepgram builds real-time speech-to-text models like Nova-3, text-to-speech models like Aura-2, unified Voice Agent APIs, and audio intelligence tools for summarization and sentiment analysis. Enterprises and developers in healthcare, customer support, media, and conversational AI use them for transcription, analytics, and voice agents. They stand out with ultra-low latency under 300ms, high accuracy in noisy conditions, custom model training, and cost-effective per-minute pricing.

Revenue model

Usage-based API pricing: Pay As You Go (free $200 credit, then Nova-3 STT $0.0077/min, Aura-2 TTS $0.030/1k chars, Voice Agent $0.075/min); Growth tier ($4K+/year prepaid credits, ~20% discount); Enterprise custom pricing with self-hosting and priority support.

Moat

  • Proprietary Technology
  • Patents/IP
  • Brand
  • Cost Advantages

Deepgram's competitive moat stems from its superior accuracy, speed, and customization in voice AI technologies like speech-to-text and real-time transcription, bolstered by 10 patents in AI, neural networks, and machine learning, plus flexible deployment options and a strong reputation among enterprises like NASA and Spotify.

Headwinds

Large tech companies like Google, Amazon, and Microsoft could commoditize speech AI capabilities through their cloud platforms.

Active layers