Cohere Earns 85% From Private Deployments

Cohere’s revenue mix shows it is being bought more like enterprise infrastructure than like a metered AI API. In practice, private deployments mean the customer runs the model inside its own cloud or data center, signs a multi year software contract, and pays for control, security, and compliance. That shifts Cohere away from volatile token spend and toward steadier licensing revenue with better visibility and stronger gross margins.

• This model fits regulated buyers that cannot send sensitive data into a shared third party service. Cohere’s cited customers include Oracle, RBC, Fujitsu, and LG, and these deals are described as multi year private deployment contracts rather than pure API consumption.
  1 sacra 2 sacra
• The unit economics are different from consumer and developer heavy labs. When customers run models on their own infrastructure, Cohere avoids paying to host every inference request itself, which supports SaaS like gross margins estimated at 70% to 80%.
  1 sacra
• This has become a real lane, not a side niche. Mistral is also selling private on premise deployments and implementation support, while OpenAI and Anthropic have scaled faster through cloud APIs and apps. Cohere’s mix shows it is leaning hardest into the sovereign and enterprise controlled end of the market.
  3 sacra 4 sacra 5 sacra

The next step is turning private deployments into a broader enterprise stack. As Cohere layers agents and workflow products like North on top of its models, it can use the deployment relationship as the wedge, then expand from model licensing into higher value software embedded inside large customer environments.