Rachad Alao, vice president of product engineering at Cohere, took the stage at VB Transform 2026 in Menlo Park to lay out a compelling argument for AI sovereignty. Speaking to a packed room of enterprise leaders and technical experts, Alao emphasized the need for organizations to maintain control over their AI systems to protect sensitive data and ensure adaptability. His stance is clear: true sovereignty in AI requires mastery over the entire agent stack, from infrastructure to governance.
### What Cohere Offers: Full Control Over AI Systems
Cohere, a Canadian enterprise AI startup, aims to empower organizations by offering solutions that ensure tight control over AI operations. According to Alao, this control extends beyond merely downloading open models or running AI applications behind corporate firewalls. For banks, hospitals, and governments operating mission-critical systems, having control over where data resides and how AI is governed is crucial. Cohere’s approach involves leveraging private-cloud infrastructure and governance systems to manage AI models and their interactions with enterprise data.
This vision aligns with the idea that AI operations should be conducted in jurisdictions that organizations can understand or directly control. Control over the entire stack—from GPUs to agent frameworks—is vital for maintaining sovereignty, Alao argued. This approach allows enterprises to manage their AI resources efficiently while safeguarding against potential data security risks.
### Competitive Context: The Debate Over Model Utilization
During the conversation, VentureBeat CEO Matt Marshall questioned the economic viability of locally deployed models, given that inference prices are dropping. Alao responded by highlighting that the complexity and consumption of AI tasks are increasing even faster. As enterprises transition from simple chatbots to more sophisticated agents capable of reasoning, calling tools, and navigating complex workflows, token utilization skyrockets.
Cohere stands out by not billing customers based on token consumption, unlike some other providers. Instead, Alao emphasized helping enterprises tackle their most challenging problems securely and privately. By routing tasks according to the required intelligence and regulatory demands, Cohere aims to optimize model usage, not maximize it. This approach can potentially lead to cost savings and more efficient AI operations for businesses.
### Real Implications: Smaller Models for Most Enterprise Needs
Cohere’s strategy includes the use of its open-source North Mini Code, which competes with larger proprietary models. While acknowledging that larger models might perform better on complex tasks, Alao argued that for 80% of use cases, smaller models are more effective and cost-efficient. This pragmatic approach allows developers to achieve their objectives without incurring unnecessary expenses associated with larger models.
For founders and engineers, this means that focusing on the right model for the task can lead to significant efficiency gains. It also underscores the importance of understanding the specific needs of their business operations. By avoiding the indiscriminate use of larger models, enterprises can streamline their AI processes and reduce costs, making AI solutions more accessible and practical.
### What Happens Next: Decisions for Founders and Engineers
As AI continues to evolve, the debate over model utilization and sovereignty will likely intensify. For founders and engineers, the challenge lies in balancing the need for advanced AI capabilities with the imperative of maintaining control and optimizing resources. Cohere’s approach offers a roadmap for those looking to implement AI solutions that are both effective and sustainable.
For investors, the takeaway is clear: startups that prioritize AI sovereignty and efficiency are worth watching. As enterprises seek more tailored and secure AI solutions, companies like Cohere that offer control over the full agent stack are well-positioned to meet these demands.