Deepgram

Voice & speech AI · https://deepgram.com

Verdict: Hold · Confidence: medium · Reviewed 2026-10-15

Deepgram sells speech-to-text, text-to-speech, and voice-agent APIs. On January 13, 2026 it announced a $130 million Series C at a $1.3 billion valuation led by AVP. It says more than 1,300 organizations and 200,000 developers build on the APIs. Revenue was not in the release.

Rubric

DimensionScore (1–5)Notes & sources
Real customers & revenue quality
Are there named paying customers, recurring revenue, and retention evidence?
3/5
Deepgram says more than 1,300 organizations and 200,000 developers use its APIs, and that it has processed over 50,000 years of audio and transcribed over 1 trillion words. Twilio is quoted as a partner. No revenue figure was in the Series C release.
Unit economics
Do gross margins survive inference/compute costs? Is pricing durable?
UnknownUnknown — no public figure found.
Moat
Proprietary data, distribution, switching costs, or tech that is hard to copy?
3/5
The company cites 2025 U.S. patents including US 12,380,880, US 12,334,075, and US 12,499,875, plus self-hosted deployment. Speech APIs are contested by hyperscalers and other voice vendors, so patents and usage scale are the disclosed edge.
Platform risk
Could a model provider or incumbent ship this as a feature? (higher score = lower risk)
3/5
Deepgram owns its speech models and offers cloud or on-prem APIs, which lowers dependence on a single LLM vendor. Customers can still switch STT/TTS providers, and the voice-agent layer competes with orchestration startups.
Team & execution
Shipping cadence, leadership track record, stability.
4/5
CEO Scott Stephenson closed a $130 million round with strategic investors including Twilio, ServiceNow Ventures, SAP, and Citi Ventures, and acquired OfOne for restaurant voice. The release says OfOne delivered more than 95% containment.
Funding vs. traction
Is capital raised justified by verifiable traction?
2/5
A $1.3 billion valuation and $130 million round are disclosed. Revenue is not, so the valuation cannot be tied to a public ARR figure despite large usage claims.
Hype signals
Demo-driven marketing, vague metrics, valuation outrunning evidence? (higher score = fewer red flags)
3/5
Usage and patent numbers are specific. The release also calls several models 'the world's most' accurate or cost-effective without publishing revenue or margins.

Listing source: https://deepgram.com. Revenue, users, and valuation figures appear only when publicly sourced; otherwise they are marked unknown.