NVIDIA CEO and Salesforce Unveil CRM Reasoning Model at Dreamforce
At Dreamforce, NVIDIA's Jensen Huang joined Salesforce's Marc Benioff to announce Koa, a CRM reasoning model built on NVIDIA's Nemotron 3 Super. The model is post-trained on Salesforce's proprietary data and runs entirely within Salesforce's infrastructure. Huang emphasized AI safety as an engineering priority and dismissed the idea that AI threatens jobs.
The Koa model represents a notable shift in enterprise AI deployment, as it was post-trained on a proprietary synthetic dataset derived from 27 years of Salesforce CRM deployments. Because NVIDIA's Nemotron 3 Super is an open model, Salesforce retains full control over the weights and runs inference entirely on its own infrastructure, with no customer data used during training or inference phases.
Huang framed AI as a transformative infrastructure layer comparable to electricity and the internet, while emphasizing that safety must be treated as an engineering priority rather than an afterthought. He also pushed back against fears that AI would eliminate jobs, instead arguing that productivity gains would expand possibilities for enterprises and workers alike.
This partnership could reshape how enterprises approach AI adoption, as the fully on-premises model addresses privacy concerns that have slowed CRM AI deployment. Companies may now weigh the benefits of proprietary reasoning models against the control and security of open, self-hosted alternatives. Workers in sales and customer service could see their roles evolve toward AI-assisted workflows, though the long-term employment impact remains uncertain. The emphasis on safety as an engineering problem may also influence industry standards for AI deployment.