Zoom and Nvidia team up to push the next era of custom enterprise AI
Zoom has partnered with Nvidia to deliver a faster, more secure and more customisable AI experience for enterprises. The move expands Zoom's federated AI architecture with Nvidia's Nemotron open technologies to power AI Companion 3.0 across finance, healthcare and government.
The goal is straightforward: better reasoning, lower latency and improved cost efficiency by routing each task to the right model in real time. For teams, that means smarter assistance, deeper integration with existing tools and less overhead managing multiple AI systems.
Hybrid model routing: SLMs for speed, LLMs for depth
Zoom is rolling out a hybrid language model approach that routes queries between its proprietary small language models (SLMs) and a fine-tuned large language model (LLM). SLMs handle focused skills and low-latency tasks; the LLM takes over when complex reasoning is required.
This sits inside a federated architecture that selects the best model for each job. It can tap both open and closed models, aiming to balance accuracy, speed and cost while meeting enterprise privacy expectations.
Works across your existing stack
Using Nvidia Nemotron, Zoom's AI Companion extends across Microsoft 365, Microsoft Teams, Google Workspace, Slack, Salesforce and ServiceNow. It's built to reason through multistep tasks, automate workflows and offer intelligent assistance inside the tools your teams already use.
Under the hood: Nemotron + Zoom's new 49B LLM
Zoom is introducing a 49-billion-parameter LLM based on Nvidia Nemotron and developed using Nvidia NeMo tools, targeting a practical balance of speed, cost and accuracy. Nemotron reasoning models are used to enhance decision-making and task orchestration across the enterprise.
Looking ahead, Zoom plans to incorporate optimal models dynamically, including a Llama Nemotron Super-based reasoning model, to sustain a balance of performance and cost as use cases evolve.
Privacy, cost and latency get first-class treatment
Federated model selection plus Nvidia's AI infrastructure aims to deliver private, efficient and scalable AI operations. By mixing open and closed models, Zoom expects improved cost efficiency, deeper reasoning capabilities and faster AI workflows without sacrificing quality.
What this means for your team
- IT leadership: Centralised control over which models run where, with options to meet data residency, privacy and budget constraints.
- Developers: Clear path to RAG-based features and workflow automation inside core productivity suites; lower latency on SLM-driven tasks.
- Product and ops: More reliable summarisation, decision support and multistep task handling embedded in day-to-day tools.
- Security and compliance: Federated architecture supports private deployments and selective use of open/closed models to align with policy.
What Zoom and Nvidia are saying
"We've increased our speed and enhanced lower-cost model decision making using Nvidia GPUs and AI software stack, helping to optimise AI Companion's core capabilities and enable faster go-to-market timelines," said Zoom chief technology officer X.D. Huangt. "With the help of Nvidia Nemotron open technologies, we're accelerating the development of our enterprise retrieval-augmented generation [RAG] capabilities, allowing AI Companion to work seamlessly with Microsoft 365, Microsoft Teams, Google Workspace, Slack, Salesforce and ServiceNow. This partnership allows us to deliver powerful, security-focused and scalable AI experiences to our customers at rapid speed."
Kari Briski, vice-president of GenAI software at Nvidia, added: "The integration of Nvidia Nemotron into Zoom AI Companion allows enterprises like Nvidia to experience a private, powerful and personalised work environment for enhanced productivity. Our collaboration with Zoom to bring smart and efficient AI reasoning to customers is just beginning."
How to prepare and evaluate
- Map use cases: Identify where SLM speed vs. LLM reasoning matters (e.g., meeting notes vs. cross-system incident analysis).
- Line up data for RAG: Define retrieval sources (docs, wikis, tickets, CRM) and access rules to avoid data leakage.
- Test integrations: Validate performance inside Microsoft 365, Teams, Google Workspace, Slack, Salesforce and ServiceNow.
- Set guardrails: Establish prompt policies, red-teaming, audit logs and fallback behaviours for sensitive workflows.
- Measure ROI: Track latency, task completion rates, human review time, quality scores and compute spend.
For more on the stack behind this announcement, see Nvidia's overview of Nemotron models here and Zoom's AI Companion product page here. If your teams are upskilling for enterprise AI and RAG workflows, explore curated programs by role at Complete AI Training.
Your membership also unlocks:
