Salesforce and Nvidia Unveil Domain-Specific Reasoning Model for Enterprise Work

Salesforce and Nvidia have introduced Koa, a new reasoning model purpose-built for the functions at the heart of Salesforce’s business: sales, marketing, and customer support. The model is constructed on Nvidia’s open-weight Nemotron family and signals a deepening collaboration between the enterprise software giant and the AI chipmaker.

Koa’s distinguishing feature is its specialization. Rather than competing as a general-purpose assistant, the model is trained specifically to execute the day-to-day tasks that enterprise revenue and service teams perform — qualifying leads, drafting outreach, resolving support tickets, and similar workflows. That vertical focus places it directly alongside Salesforce’s Agentforce platform, which the company positions as a way for customers to build, deploy, and manage enterprise-grade AI agents.

The choice of Nvidia’s open-weight Nemotron as a foundation is notable. Open-weight models give enterprise vendors like Salesforce the ability to fine-tune, audit, and deploy AI systems inside their own infrastructure — an appealing proposition for corporate buyers with strict data-governance requirements. The partnership also reinforces Nvidia’s strategy of supplying not just silicon but model software to the enterprise ecosystem.

For Salesforce, the launch extends a broader push to weave reasoning-capable AI throughout its customer relationship management suite, moving beyond copilot-style assistance toward autonomous agents that can complete multi-step work. The move comes as enterprise software vendors race to embed agentic AI before standalone AI labs and startups capture the same budget dollars.

Salesforce (CRM) shares traded at $259.43 on Tuesday, up 0.26% from a prior close of $258.75, giving the company a market capitalization of roughly $200.9 billion. Nvidia (NVDA) slipped 0.73% to $223.67 from a previous close of $225.31, with a market cap of approximately $5.56 trillion.

The pairing is strategically complementary: Salesforce contributes decades of customer-interaction data patterns and workflow expertise, while Nvidia supplies the model architecture and the accelerated computing infrastructure on which such models are trained and served. Whether domain-specialized reasoning models can outperform general frontier models on enterprise tasks — at lower cost — is the question the release’s framing implicitly raises.

Source: original release

What to watch

• Details on Koa’s pricing, availability, and integration timeline with Agentforce
• Benchmark or performance comparisons against general-purpose models on sales and support tasks
• Salesforce’s next earnings report for commentary on AI product adoption and monetization
• Further joint announcements from the Salesforce–Nvidia partnership