Salesforce and Nvidia Launch New Business Intelligence Model

Salesforce and Nvidia Launch New Business Intelligence Model

2026-09-16 companies

San Francisco, Wednesday, 16 September 2026.
Salesforce and Nvidia introduced Koa, an advanced reasoning model trained on 27 years of enterprise data without using actual customer information, achieving three times fewer errors than existing software.

Technical Architecture and Data Security

The model, built on NVIDIA Nemotron 3 Super architecture, leverages 27 years of Salesforce CRM intelligence derived from synthetic datasets rather than customer data [1][4]. This distinction ensures that no customer data crosses the trust boundary during training or inference, addressing critical enterprise security concerns [2][4]. Salesforce retains control over model weights, post-training, and inference within its own trust boundary, differentiating it from reliance on external frontier models [1][3].

Performance Metrics and Efficiency

Performance benchmarks indicate significant improvements over previous iterations. Koa demonstrates 11% higher precision in action-calling and 2.1 times greater reliability in recalling customer context compared to prior models [2]. In internal CRM benchmarks measuring tasks like updating opportunities and routing cases, the system exhibits 3x fewer errors [1][2]. The model is optimized for token efficiency, potentially lowering inference costs compared to general-purpose frontier models like Claude or ChatGPT [3][5].

The shift towards open models is evident in the broader market context. Open model usage increased from 30% at the beginning of 2025 to 70% as of September 2026 133.333 [4]. This infrastructure supports Missionforce, extending AI capabilities to government and regulated organizations via private clouds and air-gapped networks [1]. Salesforce (NYSE: CRM) and Nvidia (NASDAQ: NVDA) stock tickers reflect the companies involved in this partnership [1][5].

Deployment Timeline and Pilot Programs

Availability is staged for late 2026. While announced on September 15, 2026, general availability for the Koa model is scheduled for winter 2026 in U.S. regions [1][2]. Currently, the model is accessible to select pilot customers including 1-800Accountant, Formula 1, UChicago Medicine, and Xero [1][2]. Siemens AG is currently utilizing Salesforce Agentforce at scale, providing a live enterprise deployment of the company’s agentic AI tools [6].

Strategic Implications and Future Outlook

The initiative positions Salesforce as an enterprise alternative to closed frontier models, aiming to reduce token usage costs and adhere to embedded security requirements [3][5]. Salesforce is expanding its AI-powered contact center globally through partnerships with Bandwidth Inc. and Maestro, covering more than 25 countries [6]. Investors are advised to monitor customer adoption and deployment metrics over simple product launch announcements to gauge the success of Salesforce’s AI initiatives [6].

Sources


Enterprise AI Autonomous Agents