Deepgram Explores Local Voice AI and Air-Gapped Laptops
Deepgram's Nick Mann discusses voice AI infrastructure, hardware demands, and the goal of running models natively on disconnected laptops.
Voice technology is increasingly becoming a primary way people interact with their devices, replacing traditional screens and clicks. This transition places heavy demands on the underlying compute, storage, and partner ecosystems that power listening and speaking systems.
Deepgram Inc., which broadened its Flux model to support 10 languages earlier in the year, requires substantial infrastructure to sustain both its internal research staff and a massive base of external users. Nick Mann, staff technical program manager at Deepgram, discussed these operational realities during an interview at the Dell “AI Leadership Symposium” broadcasted by SiliconANGLE Media’s livestreaming studio, theCUBE.
According to Mann, the company employs a research team that demands significant computational resources and data. Operating in the market for 13 years has allowed Deepgram to optimize its models for challenging acoustic environments, including emergency medical services and air traffic control. Despite these capabilities, the company operates under the assumption that its current software will eventually be outperformed, prompting continuous retraining cycles using improved data.
Infrastructure Partnerships and Local Deployments
Many of Deepgram's users choose to self-host the software rather than relying solely on the cloud. A recent collaboration with Fortanix Inc. facilitates private, on-premises voice AI deployments tailored for regulated sectors. Meanwhile, Dell Technologies Inc. provides the necessary hardware infrastructure while simultaneously utilizing Deepgram's models within its own Customer Solution Center Innovation Lab.
Mann highlighted that Dell's broad customer base spans enterprises and smaller organizations alike. These deployments range from compact Nvidia Corp. GB10 systems managing interactive voice response for smaller businesses to much larger installations.
Expanding to Personal Hardware
Beyond traditional data centers, Mann noted that the company is aiming to bring its voice technology directly to the consumer hardware level, focusing on the AI-enabled laptops utilized by students. The goal is to enable the models to execute natively and in air-gapped environments, bringing intelligence directly to where users operate.
What it means for developers
As voice architecture shifts toward local execution and localized hardware, developers face increasing demands for efficient compute and versatile model integration. For those looking to experiment with various architectures, developers can try top AI models cheaply through one API at https://apixoai.online, simplifying access to leading artificial intelligence tools without managing multiple separate provider keys.
Source: Deepgram targets voice AI on air-gapped laptops — SiliconANGLE AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
