Aidan Gomez, the CEO of Canadian AI firm Cohere, has been preparing for potential scenarios in which foreign governments might suddenly restrict access to advanced AI models. His concerns became particularly relevant earlier this month when the U.S. government imposed limitations on foreign access to Anthropic's AI, Mythos, highlighting the very risks Gomez had predicted.
“This has reminded everyone of the reality that depending on a centralized entity poses a significant structural risk,” Gomez shared with Fortune. “Access can be revoked at any time, effectively shutting down services.”
The United States has predominantly led the global AI landscape, controlling a substantial share of the world’s computational resources and housing many of the top research labs. China stands as its closest competitor, and together, the two nations dominate 90% of global computing capabilities, according to Epoch AI research.
Beyond the U.S. and China, the pool of significant AI developers is limited. Toronto-based Cohere, established in 2019, and Paris-based Mistral are among the few notable firms working on scalable AI models outside this binary dominance. As AI integrates into essential sectors, including healthcare, finance, military operations, and public services, the stakes regarding who controls this technology and its operational status are rising, tying into broader national interests.
Gomez advocates for democracies to shift from relying on a few foreign entities for critical AI services. He believes they should establish independent systems that allow for comprehensive control over all components of AI infrastructure—this includes data centers, the underlying chips, and the development of models, rather than depending on external companies for access via APIs.
“Relying on rental AI from external sources poses a national security threat,” Gomez asserts. “Complete control is crucial.”
Cohere has already implemented a model where its AI can be deployed within clients' own infrastructures. Instead of transmitting data to Cohere's servers, customers obtain the model and necessary software to operate it in-house, including on systems that are entirely disconnected from the internet—thereby ensuring Cohere has no remote access or capability to terminate services.
The concept of "Sovereign AI" has emerged as a broad term, often representing various degrees of independence. Some interpretations focus on controlling the data used to train AI systems, while others emphasize the ability to create new advanced models independently. There are also definitions that prioritize keeping sensitive data within national boundaries. This ambiguity contributes to ongoing political discussions and enables governments and companies to claim progress in AI sovereignty without adhering to stringent requirements.
Gomez believes that the foundation of true sovereignty lies in domestic infrastructure.
“To secure sovereignty, it all starts with having infrastructure that a country controls,” he noted. “A home-grown infrastructure champion is essential.”
He contends that only once a nation has a domestic player capable of managing the critical elements—chips, power, and data centers—of its AI systems can it ensure that foreign entities cannot easily disrupt services.
Nevertheless, the aspiration for complete AI sovereignty encounters significant challenges related to computation, capital, and energy resources. The high costs associated with developing and maintaining cutting-edge systems mean that most nations may not be able to independently own every aspect of AI infrastructure. According to Gomez, a practical approach to sovereignty must involve forming alliances.
Instead of each country attempting to recreate the entire AI ecosystem independently, Gomez proposes that democracies collaborate around a select few "champions" and share capabilities through trusted partnerships.
“It isn’t feasible for every nation to possess all layers of the AI stack internally… they should seek various partners for each layer,” he explained. “If we disperse our resources too thinly, we won’t succeed. It’s essential to support our champions rather than dilute our efforts.”
Cohere is currently looking to establish partnerships with other "middle power" countries and firms. In April, the company announced its plan to acquire Germany's Aleph Alpha, with the goal of creating a transatlantic AI powerhouse based in both Canada and Germany.
This move aligns with a Joint Declaration of Intent on Artificial Intelligence that Canada and Germany established prior to the acquisition, aimed at fostering a sovereign technology alliance. The combined entity, maintaining the Cohere brand under Gomez's leadership, is projected to be valued at approximately $20 billion, as reported by the Financial Times.
During recent G7 meetings, Gomez noted a shared understanding among leaders and tech executives regarding the necessity for democratic nations to align on policy standards and regulations. He emphasized the importance of not only coordinating but also investing in shared resources to ensure that vital infrastructures are not dependent on a handful of American or Chinese firms.
In Gomez's view, the political landscape concerning AI mirrors the technology itself.
“Centralized reliance on a single entity is inherently risky; we must eliminate single points of failure,” he argued. “As a computer scientist, I know that lesson well: avoid constructing systems with vulnerable points.”
