Deepgram Acquires Daily to Own the Voice AI Stack and Expands Enterprise Teams

Deepgram acquired Daily, the creator of the open-source real-time orchestration framework Pipecat, to expand its infrastructure for speech-to-text, text-to-speech, and voice agents, according to TechCrunch. The deal follows a $130 million Series C funding round led by AVP at a $1.3 billion valuation, reported by Slator, bringing total funding past $215 million.

Founded in 2015 out of Y Combinator's W16 batch, the San Francisco-based voice AI platform serves more than 200,000 developers and 1,400 enterprise clients. Deepgram processed over 50,000 years of audio and transcribed over one trillion words by 2024. The company is now opening an Asia-Pacific headquarters in Singapore under newly appointed VP and GM Sriram Ved following a strategic investment from EDBI, reported by Yahoo Finance.

Building Voice Infrastructure from Hardware to the Edge

Deepgram designs its speech models and inference software in-house. Its model portfolio includes Nova-3 for speech recognition, Aura-2 for speech synthesis, and the Flux model family, which handles turn detection and conversation timing directly inside recognition and synthesis engines, according to audioXpress.

To manage high-volume workloads, Deepgram optimized Nova-3 on AWS Trainium2 chips by replacing standard PyTorch layers with custom kernels and building a lightweight data-parallel runtime in Rust, according to AWS. The setup increased audio throughput per dollar by 11% to 70% compared to previous compute options.

Outside cloud environments, Deepgram delivers virtual private cloud, on-premises, and hardware-level speech setups. The company partnered with Dell Technologies to move speech models onto air-gapped, AI-enabled laptops, reported by SiliconANGLE.

To build its deep learning models and custom serving runtime, the company is hiring a Research Scientist, Voice ($150,000–$250,000, US Remote), who will own the deep learning stack, optimize model architectures for inference at scale, and design experimental programs for ASR and TTS.

Leadership Rooted in Physics and Scaled Systems

Co-Founder and CEO Scott Stephenson studied dark matter particle physics at the University of Michigan, earning a master's degree in physics before starting Deepgram to solve the challenge of searching recorded laboratory audio. He co-founded the company alongside CTO Noah Shutty, a Stanford graduate and former Google software engineer, and CTO Adam Sypniewski, who holds a Ph.D. in physics from the University of Michigan and previously conducted research at Soar Technology.

That research background shaped Deepgram's focus on processing raw audio waveforms through end-to-end neural networks rather than traditional heuristic signal pipelines.

Hiring Focus: Vertical Solutions and Commercial Expansion

Deepgram is expanding into specialized vertical markets, starting with physical environment AI integrations in food service. In San Francisco, the team is building backend platform infrastructure and hardware integrations through the Software Engineer - Deepgram for Restaurants ($150,000–$230,000, San Francisco, CA), who will design backend systems connecting machine learning models with POS, loyalty, and payment hardware. Deployments in that unit are led onsite by the Enterprise AI Strategist, Deepgram for Restaurants (San Francisco, CA), who oversees end-to-end customer execution.

To support enterprise sales across technical buyer personas, Deepgram is scaling its commercial team with remote positions. The Corporate Account Executive ($240,000–$260,000, Remote US) will drive revenue through self-prospecting and technical platform sales, while the Sales Development Representative ($50,000–$60,000, Remote US) will prospect new opportunities with software leaders.

See all open roles at Deepgram