Salesforce Builds on Nvidia for an AI Model Meant to Steal Enterprise Work

The Shift Away from Big AI Subscriptions

At the Dreamforce conference, Salesforce announced its first reasoning model, Koa, built on Nvidia's open-weight Nemotron. The model is fine-tuned specifically for sales, marketing, and customer support.

The key dynamic lies in what it offers large companies: a way to avoid closed models like GPT-4 or Claude. Enterprise customers can use a smaller model that keeps sensitive data in-house and reduces token consumption for specific corporate tasks.

Why Koa Is a Nightmare for Big Labs

For the AI labs, this is a direct attack on their business model. They profit from companies sending data into their APIs. With this move, Salesforce keeps corporate data within its infrastructure, handling the model locally without relying on expensive cloud services from OpenAI or Anthropic.

This is attractive to large customers because they don't need a universal model capable of writing poetry or solving quantum physics. They need a model that reliably resolves a support ticket.

The Hidden Vendor Lock-in Continues

The catch remains that Koa further locks clients into the Salesforce ecosystem. It's a cheaper alternative to frontier AI, but for the customer, it's just swapping one dependency for another. The Agentforce platform ensures that API savings are traded for deeper CRM integration.

Who Wins the Enterprise Segment

The adoption of similar industry-specific models will determine if big AI companies can keep enterprise clients. The proof will be whether companies start hosting smaller, fine-tuned models themselves, or eventually stick with the big ones just under a different wrapper.

Lilith's verdict

We are watching a battle over who gets the data and API money, rather than true innovation. Why pay OpenAI for math skills when you just need to apologize for a delayed package?