Cohere VP: Enterprise AI sovereignty requires control of the full agent stack
Cohere VP Rachad Alao emphasized the need for control over the entire "stack" for enterprise AI solutions, aiming to secure data, infrastructure, and supply chain independence.
Enterprise AI sovereignty requires organizations to have complete control over the entire agent stack, including GPUs, cloud infrastructure, governance systems, and model routing, said Rachad Alao, vice president of product engineering at Cohere, at the recent VB Transform 2026 conference.
Alao stressed that true AI sovereignty extends beyond simply downloading an open model or running applications within a corporate firewall. He highlighted that control over data residency and AI operations is paramount in critical sectors such as finance, healthcare, and government.
Alao also challenged the notion that falling token prices diminish the need for AI usage optimization. He noted that overall consumption is increasing significantly as organizations move to more complex, multi-step agents that reason through problems and utilize tools. Such workflows substantially increase processing power and token usage.
Instead of maximizing token consumption, Cohere aims to help enterprises solve their hardest problems privately and securely. According to Alao, organizations should use the most appropriate model for each task. For instance, a Canadian bank uses Cohere models locally for regulated tasks and cloud-based models for less sensitive but more demanding jobs.
Alao also mentioned the company's North Mini Code open-source model, which is efficient and cost-effective for most use cases, even if larger models may be more capable for edge cases. The reporting is based on interviews and conference remarks focusing on technical control and ensuring data sovereignty.