Zoom expands AI companion with NVIDIA Nemotron

Zoom and NVIDIA are partnering to expand Zoom’s Artificial Intelligence Companion by integrating NVIDIA Nemotron open technologies into a federated architecture. The collaboration introduces a next-generation hybrid language model approach that intelligently routes queries between Zoom’s proprietary small language models, optimized for low latency and specific skills, and a fine-tuned large language model for deeper reasoning. Zoom says the framework balances speed, cost, and accuracy and will power AI Companion 3.0 across industries such as finance, healthcare, and government.

The technical stack includes a new 49-billion-parameter large language model built on NVIDIA Nemotron and developed with NVIDIA NeMo tools, alongside other Nemotron-based reasoning models such as the Llama Nemotron Super. Zoom’s federated architecture is patent pending and already used for real-time transcription, translation, and summarization. The partnership leverages NVIDIA GPUs and software to accelerate development, improve lower-cost model decision making, and enhance retrieval-augmented generation capabilities. The company highlights integrations with Microsoft 365, Microsoft Teams, Google Workspace, Slack, Salesforce, and ServiceNow to streamline enterprise workflows.

Zoom frames the work as a responsible Artificial Intelligence foundation for enterprise deployments, emphasizing security and privacy controls. The company states it does not use customer audio, video, chat, screen sharing, attachments, or other communications to train its or third-party models. Zoom and NVIDIA position the collaboration as a way to deliver customizable, private, and scalable AI experiences that improve collaboration and automation while aiming for optimized cost efficiency, quality, and latency for customers and government organizations.