Skip to main content

Deepgram

Category: Voice / Speech AI

A foundational AI platform providing high-performance, real-time speech-to-text and text-to-speech APIs for developers and enterprises building voice-first applications. Deepgram was founded in 2015. The company is led by Scott Stephenson. Based in San Francisco, USA. Team size: 200+. Total funding raised: $215M total. Latest round: Series C ($130M, Jan 2026) at $1.3B valuation. Key investors include ["AVP","Madrona","Wing VC","NVIDIA","Tiger Global","Y Combinator"].

Founded
2015
Headquarters
San Francisco, USA
Team size
200+
Total funding
$215M total

Value proposition

Delivers the industry's highest accuracy and lowest latency for voice AI at the lowest total cost of ownership, enabling human-like real-time conversational interactions.

Products and solutions

["Nova-2 (Speech-to-Text): High-accuracy, GPU-accelerated transcription","Aura / Aura-2 (Text-to-Speech): Low-latency, human-like voice synthesis","Voice Agent API: End-to-end framework for real-time conversational AI","Flux: Conversational speech recognition model with advanced turn detection","Language AI: Unified API for summarization, sentiment analysis, and NLU"]

Unique value

Utilizes a radical end-to-end deep learning architecture that bypasses legacy phoneme-based models, allowing for direct audio-to-text processing with higher contextual awareness.

Target customer

Software developers, enterprise engineering teams, contact center operators, and AI agent builders.

Industries served

["Contact Centers & Customer Experience","Media & Entertainment","Healthcare & Telemedicine","Finance & Banking","Technology & Software Development"]

Technology advantage

GPU-native infrastructure and proprietary foundational models provide sub-300ms latency and superior accuracy in noisy or domain-specific environments, outperforming legacy 'Big Tech' providers in both speed and cost-efficiency.

How they differentiate

Deepgram utilizes a proprietary end-to-end deep learning architecture that bypasses legacy phoneme-based models, delivering sub-300ms latency and higher accuracy for real-time conversational AI compared to 'Big Tech' providers.

Main competitors

["AssemblyAI","OpenAI (Whisper)","Google Cloud Speech-to-Text"]

Key partnerships

["NVIDIA (Strategic technology and investment partner for GPU optimization)","Amazon Web Services (AWS Global Startup Programme & SageMaker integration)","Twilio (Preferred voice AI partner for telecommunications infrastructure)","Voximplant (Collaboration on production-ready voice agents for developers)","Y Combinator (Alumnus and strategic ecosystem partner)"]

Notable customers

["Spotify","NASA","Citibank","Twilio","Viber"]

Major milestones

["Achieved Unicorn status with a $1.3B valuation in January 2026","Acquired YC-backed AI startup OfOne in 2026","Acquired communication AI startup Poised in 2024","Launched Nova-2 and Aura-2 foundational models for speech-to-text and text-to-speech"]

Growth metrics

Reached $21.8M ARR in Oct 2024; supports over 200,000 developers and 400+ enterprise customers.

Market positioning

High-performance Voice AI platform leader targeting developers and enterprises building real-time AI agents.

Geographic focus

Global, with primary market concentration in North America, Europe, and Asia-Pacific.

Patents and IP

Holds 10+ active patents covering hardware efficiency for AI, transformer-based speech architectures, and domain-specific language modeling innovations.

About Scott Stephenson

Scott Stephenson is a physicist-turned-entrepreneur who co-founded Deepgram after researching dark matter at the University of Michigan, where he built a lab two miles underground. He is a Y Combinator alumnus (W16) and has led Deepgram to become a $1.3 billion unicorn in the voice AI space, pioneering foundational models for speech-to-text and real-time voice agents.

Official website: