Deepgram lands US$130M to expand real-time Voice AI infrastructure

Deepgram, a US-based API platform that powers real-time voice interactions, has secured US$130 million in Series C funding at a US$1.3 billion valuation. The round was led by AVP and backed by a mix of institutional, strategic, and academic investors including Twilio, BlackRock, SAP, ServiceNow Ventures, and several major universities.

This marks a significant milestone for Deepgram, which positions itself as the backbone of the Voice AI economy. While the voice tech space includes consumer-facing players like Alexa and Siri, Deepgram focuses on B2B infrastructure—providing speech-to-text, text-to-speech, and speech-to-speech APIs used by developers and enterprises building voice-first products.

For marketers and customer engagement teams, this translates into tools that can fuel faster, more human-like conversational experiences—without building models from scratch.

Short on time?

Here’s a table of contents for quick access:

Why the raise matters for marketers

Unlike traditional speech recognition tools, Deepgram’s APIs are built to handle fully duplex (simultaneous two-way) conversations with low latency and high customization. That makes it a strong fit for live customer service, conversational marketing, and automated voice agents that need to sound less like robots and more like humans.

With this new round, Deepgram plans to scale its infrastructure globally, double down on developer tools, and support domain-specific voice models for industries like finance, healthcare, and QSR (quick-service restaurants).

Marketers operating in high-volume, voice-centric environments—like call centers, drive-thrus, or outbound sales—could gain access to voice tech that feels less like a novelty and more like a reliable backend service.

Deepgram lands US$130M to expand real-time Voice AI infrastructure

What’s Deepgram building?

Deepgram’s core pitch is speed, accuracy, and real-time processing at scale. Some of the flagship models and tools in its portfolio include:

Nova-3: A speech-to-text model optimized for accuracy in live conversations.
Aura-2: An enterprise-grade text-to-speech engine for generating human-like voices.
Flux: Built specifically to handle interruptions during calls—something most voice agents struggle with.
Voice Agent API: Real-time voice AI API that plugs into customer service and voicebot workflows.
Saga: Deepgram’s own “Voice OS” to orchestrate voice interactions across platforms.

These tools are designed for flexibility: they can be deployed in the cloud, self-hosted, or run on-premises depending on compliance needs. SDKs are available to speed up implementation.

Strategic move into restaurant automation

Deepgram is also moving beyond tech infrastructure with the acquisition of OfOne, a voice automation platform built for restaurant drive-thrus. OfOne’s AI handles live orders with over 95% containment, freeing up staff and improving accuracy.

Now rebranded as Deepgram for Restaurants, the product targets national QSR chains looking to automate high-volume ordering environments. Will Edwards, OfOne’s co-founder, now serves as GM of the new division .

This vertical expansion hints at how Deepgram could apply its infrastructure in niche but high-value domains—and gives marketers in food service a new tool to rethink customer experience.

What marketers should know

Here’s how B2B marketers and customer teams can make use of Deepgram’s expanding Voice AI stack:

1. Reimagine voice channels as strategic touchpoints

Voice calls are often overlooked compared to chat or email. Deepgram’s stack makes it easier to deploy AI-driven agents that can engage, upsell, and assist in real time—without sacrificing quality.

2. Reduce friction in service experiences

Flux and the Voice Agent API are engineered to handle messy, real-world conversations—interruptions, slang, and all. This could improve customer satisfaction and deflection rates in automated systems.

3. Unlock new data from spoken conversations

Deepgram’s APIs can generate transcripts, detect sentiment, and extract keywords, giving marketers access to voice data that was previously locked inside call recordings.

4. Customize for industry-specific lingo

From financial terms to medical jargon, Deepgram models can be tuned to understand domain-specific language. That means fewer misinterpretations and more relevant experiences across regulated industries.

Voice AI infrastructure is getting serious

Deepgram’s latest raise shows that voice isn’t just a UI novelty—it’s a foundational channel for how businesses communicate. As voice-first experiences spread across customer service, sales, and commerce, companies will need infrastructure that can scale with them.

For marketers, that means not just watching the Voice AI trend from the sidelines, but actively testing how these tools can improve engagement and automation without compromising the human touch.

This article is created by AI with human assistance, powered by ContentGrow. Ready to explore content marketing automation solutions? Book a discovery call today.