Zoom & Nvidia: Inside the AI Companion 3.0 Enterprise Launch

As enterprise software companies race to embed AI capabilities into their platforms, computational demands of running sophisticated language models at scale have forced most to seek partnerships with infrastructure specialists. 

It’s a change of tactic that reflects just how resource-intensive modern AI has become.

Zoom Video Communications is the latest to take this route, announcing a tie-up with semiconductor manufacturer Nvidia to power the next version of its AI assistant. 

The video conferencing platform is leveraging Nvidia’s Nemotron open technologies to build AI Companion 3.0, which promises to adapt to individual work patterns through model customisation and Retrieval-Augmented Generation (RAG). 

RAG, as it’s known in the industry, allows AI systems to pull from external knowledge sources when generating responses – particularly useful for company-specific queries that generic models struggle with.

What makes the system interesting is its routing mechanism.

Rather than throwing everything at one model, Zoom has built a hybrid architecture that intelligently sends queries to either a small language model (SLM) or a large language model (LLM) depending on what’s being asked. 

The SLM handles straightforward tasks requiring quick responses, while the LLM tackles heavier lifting. 

It’s an approach that acknowledges a basic truth: not every question needs the full firepower of a massive language model.

“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,” says Xuedong D. Huang, Chief Technology Officer at Zoom.

How Zoom AI Companion works across Microsoft, Google and Salesforce platforms

The real test comes with integration. 

Zoom’s CTO says that “with the help of Nvidia Nemotron open technologies, we’re accelerating the development of our enterprise retrieval-augmented generation capabilities, allowing AI Companion to work seamlessly with Microsoft 365, Microsoft Teams, Google Workspace, Slack, Salesforce and ServiceNow.”

That cross-platform functionality matters in workplaces where employees routinely juggle half a dozen tools throughout the day. 

Nobody wants yet another assistant that only works within a single ecosystem. Xuedong D frames it as delivering “powerful, security-focused and scalable AI experiences to our customers at rapid speed.”

The system already handles real-time transcription, translation and summarisation within Zoom’s own environment. 

Now the company wants to extend those capabilities to multi-step operations that span meetings, chat sessions and documents across different platforms. 

It’s an ambitious scope that puts Zoom in direct competition with Microsoft Teams and Google Meet, both of which have the advantage of drawing on their parent companies’ substantial AI research operations.

How Nvidia brings reasoning models to Zoom’s federated architecture

Zoom plans to deploy Nvidia’s reasoning models for more complex tasks. 

These AI systems break down problems into sequential steps rather than attempting to answer everything in one go. 

The companies are working with the Llama Nemotron Super-based reasoning model, though specific deployment timelines remain vague.

“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,” says Kari Briski, Vice President of Gen AI software at Nvidia.

Nvidia is using the system internally, which serves as both validation and a marketing tool. For Nvidia, partnerships like this extend its reach in enterprise AI beyond selling chips and accelerators.

Zoom has its sights set on finance, healthcare and government sectors. 

These industries face stringent requirements around data privacy and compliance, which explains the emphasis on federated architecture. 

Keeping sensitive data on an organisation’s own infrastructure while accessing AI capabilities addresses concerns that have made regulated industries cautious about adoption.

The hybrid approach also tackles an increasingly pressing issue: cost. 

Running every query through large language models burns through computational resources and energy. 

By routing simpler tasks to smaller models, Zoom can keep operational costs down while maintaining performance where it counts.

“Our collaboration with Zoom to bring smart and efficient AI reasoning to customers is just beginning,” Kari says.