Salesforce and Nvidia's Koa model targets frontier labs' enterprise turf

Salesforce introduced a new AI model called Koa as one of its biggest announcements at this week’s Dreamforce conference. Koa is the company’s first reasoning model, built on Nvidia’s open-weight Nemotron model; Salesforce and Nvidia worked together to post-train it specifically for sales, marketing, and customer-support tasks, using synthetic data that mimicked customer patterns rather than any actual Salesforce customer data.

Koa illustrates how enterprise AI needs are diverging from what frontier labs offer. Where proprietary labs want enterprises uploading files, code, prompts, and feedback directly into their models, Salesforce is instead offering customers an open-weight alternative that’s cheaper to run, doesn’t ingest customer data, and can be routed automatically through an AI “gateway” depending on the task. Koa will sit alongside the other models Salesforce offers inside its Agentforce platform, where customers build agents for tasks like answering customer-service questions or scheduling appointments.

“We’ve built many small task-specific language models, which are part of Agentforce’s portfolio,” Jayesh Govindarajan, EVP of Salesforce AI, told TechCrunch. “But reasoning has always been something that we’ve relied on the frontier model providers for. Until now.” Before Koa, agents that needed to reason through long or multi-step tasks routed those prompts to a frontier model like Claude or ChatGPT through Agentforce’s gateway. Govindarajan said the missing piece had always been a pre-trained base model with clear data provenance — something Nemotron provided, unlike Alibaba’s Qwen, about which he said Salesforce has no visibility into training data. Nvidia VP Kari Ann Briski said Nemotron’s architecture makes it token-efficient for inference, which she called central to the “tokenomics” of running reasoning at scale.

Salesforce isn’t abandoning Anthropic or OpenAI, however: the company also just announced a partnership called Claudeforce that lets companies use Claude as their AI interface while keeping their data inside Salesforce’s own infrastructure.

Why it matters

Koa signals that large enterprise-software vendors increasingly want their own reasoning models rather than routing everything to frontier labs — a shift that could shrink the volume (and pricing power) of enterprise workloads flowing to Claude and ChatGPT.