According to external sources, Canadian AI Cohere CEO Aidan Gomez recently re-emphasized that government and key institutions could risk being cut off at a critical time if they relied on foreign firms for long periods of time to provide front-line AI models. This concern rose rapidly after the United States restricted overseas access to the Mythos model of Anthropic this month.
Gomez indicated to Wealth that over-reliance on a single external supplier essentially places critical capabilities under the control of others. Once access is withdrawn, the service may be interrupted directly. As AI progresses towards access to hospitals, financial markets, military logistics and public services, such risks are not only business issues, but are also beginning to be discussed at the national strategic level.
Sovereignty AI points to chain control
In Gomez, it appears that the so-called “sovereign AI”, core is not just left locally, but, more importantly, has control over the underlying infrastructure and the model itself. This includes data centres, chips and operational models, rather than renting capacity from overseas companies only through API.
He indicated that the true manageable model should be for clients to deploy the model directly within their own infrastructure, rather than sending data to the vendor server. According to him, Cohere had provided such deployment to Governments and businesses, and clients could run models in their own data centres, even in systems physically isolated from the Internet, and Cohere could not close down services remotely.
Calculus is still a reality.
However, the article also points out that the advancement of sovereignty AI faces real constraints. Front-line model training and deployment requires a great deal of computing, capital and energy, and it is difficult for most countries to have a complete technology warehouse on their own.
According to AI Research Institute Epoch AI, the United States and China collectively hold about 90 per cent of the global computing resources. Except for these two countries, there are not many companies with large front-line model development and deployment capabilities, and Cohere and the French Mistral are a minority.
Cohere promotes transnational alliances
This means that, even if many countries wish to reduce external dependence, it will be difficult to complete all the links independently in a short period of time. Gomez therefore argues that the realistic path is not one in which each country duplicates a whole system, but rather a division of labour among credible partners.
Cohere has been driving this direction in recent months. In April this year, the company announced the acquisition of a German AI company, Aleph Alpha, with the goal of establishing a transatlantic AI company across Canada and Germany. The British Financial Times had previously reported that the consolidated company valued approximately $20 billion and would continue to follow the Cohere brand, with Gomez continuing to head it.
Two months before the deal, Canada signed a joint statement of intent of artificial intelligence with Germany and launched the Sovereign Technology Alliance. According to the article, Cohere ' s deal with Aleph Alpha was seen as an attempt to further bring this political cooperation to the commercial level.
Gomez also mentioned that during last week ' s G7 discussions, there was general agreement between multinational leaders and AI corporate executives that there was a need for greater coordination of standards and regulation in democracies. He further argued that cooperation should not remain rule-based, but should also extend to joint inputs of computing, data centres and modelling capabilities to reduce reliance on a few United States or Chinese enterprises.