Jensen Huang Joins Benioff at Dreamforce to Unveil Salesforce's Koa Reasoning Model

The Dreamforce keynote paired two of the industry's most visible CEOs to launch Koa, Salesforce's first native CRM reasoning model, built on NVIDIA's Nemotron 3 Super. The partnership positions Salesforce's Agentforce and CRM stack on NVIDIA silicon and model foundations, deepening NVIDIA's push into vertical enterprise AI beyond raw GPU sales.

Koa is designed to reason over Salesforce's structured CRM data — pipeline, accounts, cases — to power autonomous sales and service agents. Building it on Nemotron 3 Super rather than a third-party frontier model signals Salesforce's intent to own more of its AI stack while leaning on NVIDIA for the model backbone, a middle path between building from scratch and renting OpenAI or Anthropic APIs.

Huang's 'know everything, do anything' framing captures the enterprise pitch: agents that combine broad world knowledge with deep access to a company's own system of record. It lands the same week AWS and Salesforce announced their own expanded integration (Salesforce context in Amazon Quick, Agentforce model choice via Bedrock), underscoring that Salesforce is playing every cloud and chip vendor simultaneously.

The competitive context is a scramble for the enterprise-agent layer, where CRM data is the moat. The open question is differentiation: Nemotron-based reasoning must prove it beats simply wiring GPT-6 or Claude into Salesforce data. Watch for Koa benchmarks and whether customers prefer a Salesforce-native model over model choice through Bedrock or Azure.