Deepgram closed a $130M Series C at a $1.3 billion valuation in mid-January, led by AVP. What's notable: the company was already cashflow positive in 2024, meaning this wasn't a survival round.
The San Francisco-based company provides speech-to-text, text-to-speech, and speech-to-speech APIs for enterprise contact centers, sales automation, and voice-enabled applications. Current deployment: 400+ enterprise customers including Twilio and Vapi, plus 200,000 developers.
The numbers suggest genuine traction. Deepgram has processed over 50,000 years of audio and transcribed more than 1 trillion words. Annual usage has grown 3.3x over the past four years. That's the kind of growth VCs pay for.
What they're positioning against: Unreliable voice AI. Taco Bell famously pulled its drive-through voice ordering system in 2025 after persistent accuracy issues. Deepgram's pitch centers on low latency, interruption handling, and production-grade accuracy for multi-speaker scenarios.
Their product stack includes Nova-3 for speech recognition, Aura-2 for synthesis, and Flux for streaming. Implementation typically uses WebSocket connections for real-time transcription, with speaker diarization available (though developers report configuration complexity in production multi-speaker scenarios).
The company recently acquired OfOne, which claims 93% order accuracy for restaurant voice systems. That acquisition suggests Deepgram is moving beyond infrastructure toward vertical-specific solutions.
For CTOs evaluating voice AI: Deepgram's free tier covers basic testing, but production deployments at scale require enterprise pricing. The WebRTC streaming implementation is well-documented, though teams report speaker identification accuracy varies based on audio quality and overlap.
The "Stripe for Voice AI" positioning is aspirational but directionally correct. Voice AI infrastructure is consolidating around a few providers who can deliver reliability at scale. Whether Deepgram becomes that default platform depends on execution over the next 12-18 months.
The funding gives them runway to prove it. We'll see.