Salesforce Partners With AWS, Google Cloud and Nvidia to Expand Agentforce Enterprise AI Ecosystem

TradingKey - On September 15, Salesforce (CRM) announced a series of AI partnerships, expanding its technical integrations with Amazon Web Services (AWS), Google Cloud, and Nvidia (NVDA). Centered on Agentforce, the company aims to connect enterprise data, AI models, AI agents, and cloud infrastructure to accelerate the deployment of artificial intelligence across scenarios such as sales, customer service, and government.

Following the announcements, Salesforce's pre-market decline narrowed to 0.3%.

Salesforce's new partnership with AWS focuses primarily on data interoperability, model selection, and workflows. Through the Model Context Protocol (MCP), enterprises can allow Amazon Quick to directly query real-time CRM data in Salesforce and execute sales or customer service tasks. AWS DevOps Agent will also connect to Slack, enabling technical teams to address infrastructure issues within the collaboration platform without switching applications frequently.

Meanwhile, Agentforce will connect to a variety of foundation models provided by Amazon Bedrock, including models from Anthropic, OpenAI, and Nvidia. The two sides will also expand Data 360's zero-copy data capabilities, enabling AI agents to access Apache Iceberg tables managed by AWS Glue and Amazon Aurora databases, reducing data migration while preserving existing enterprise permissions and governance systems.

Salesforce and Google Cloud plan to build a more unified AI tech stack. Salesforce has begun hosting customer traffic on Google Cloud infrastructure via Hyperforce and expects to drive more customer migrations in the fourth quarter. The integrations will allow enterprises to directly call Salesforce CRM data and features within Gemini Enterprise to inspect sales pipelines, generate account summaries, and process customer service tickets.

The partnership is also bidirectional: Salesforce's data and Tableau analytics capabilities will be brought into Gemini Enterprise, while Google Gemini will connect to Agentforce to help customers build Gemini-powered AI agents. Additionally, Salesforce Commerce Cloud will integrate with Google's Universal Commerce Protocol, enabling consumers to discover and purchase products directly within Google Search or the Gemini interface.

At the model level, Salesforce and Nvidia jointly launched Koa, Salesforce's first reasoning model specifically designed for customer relationship management scenarios. Koa was post-trained on NVIDIA Nemotron 3 Super using synthetic data simulating enterprise sales, customer service, and business process scenarios, without using customer data.

Salesforce stated that Koa can understand deal structures, service case lifecycles, and cross-industry workflows, helping Agentforce complete complex tasks that require multiple steps and tool calls. The company owns the model weights and conducts training and inference within its own trust boundary to meet enterprise requirements for data security and model control.

Koa is currently available to select Agentforce pilot customers and is expected to launch generally in the U.S. in the winter of 2026.

Salesforce and Nvidia will also bring Nemotron models and accelerated computing capabilities to Missionforce, offering AI systems deployable in private clouds, air-gapped networks, and other secure environments for government and highly regulated industries. The post-trained Nvidia models are expected to be available to select Missionforce customers in October 2026.