Cohere: Enterprise AI Sovereignty Requires Full Control Over Agent Stack

At VB Transform 2026, a conference focused on generative AI agent applications in business, Rashad Alao, Vice President of Products at Cohere, shared his vision for enterprise AI sovereignty. According to him, true independence from vendors and regulatory compliance are only possible with full control over the technology stack—from hardware infrastructure to agent frameworks and data management systems.

Alao, who previously worked on responsible AI teams at Google and Meta*, noted that sovereignty extends beyond running models behind corporate firewalls. Critical industries like finance and healthcare must independently determine where data is stored and how queries are processed. This applies not only to GPUs and private clouds but also to model routing systems, search tools, and agent frameworks handling enterprise data.

The expert also refuted the claim that reduced inference costs make local models obsolete. He argued that the shift to complex AI agents capable of multi-step tasks significantly increases token consumption. Cohere proposes an alternative pricing model not tied to token volume, enabling enterprises to optimize costs and select models tailored to specific tasks.

To illustrate this approach, Alao cited a major Canadian bank using Cohere’s local models for regulated processes while redirecting less critical tasks to more powerful cloud models via the North platform. The company also introduced new solutions, including the Command A+ model with 218 billion parameters and the open-source North Mini Code model optimized for single-GPU deployment on Nvidia H100.