Jensen Huang made a notable appearance at Salesforce's Dreamforce conference, joining CEO Marc Benioff onstage to mark the launch of Koa. The new system is described as Salesforce's first reasoning model built specifically for CRM tasks, and it runs on NVIDIA Nemotron 3 Super, the latest generation of NVIDIA's open models aimed at enterprise reasoning workloads.
The announcement reflects NVIDIA's broader strategy of positioning its Nemotron models as building blocks for software vendors that want to embed advanced reasoning into vertical products, rather than competing head-on with general-purpose frontier models. For Salesforce, Koa is meant to automate more complex CRM operations — pipeline analysis, next-best-action recommendations, customer data synthesis — with a level of reasoning beyond the conversational assistants already available through Agentforce.
Huang's framing during the appearance, about AI systems now being able to 'know everything and do anything,' echoes NVIDIA's familiar marketing language around agentic AI in enterprise settings. Beyond the rhetoric, the event mainly confirms a deepening partnership between the two companies, with NVIDIA supplying compute infrastructure and base models while Salesforce handles the application layer and distribution to its enterprise customer base.
The move fits a broader pattern of enterprise software vendors differentiating their AI offerings through open-weight models fine-tuned for specific business tasks, rather than relying solely on generic proprietary APIs. It remains to be seen how meaningfully Koa will stand apart from other CRM agents built on large language models, particularly in terms of inference costs and accuracy on real-world workflows.