This article was first published in Les Échos.
Choosing Sovereign AI
By relying on the proprietary technologies of US giants, businesses and governments are jeopardizing their economic independence. This issue must take center stage at the AI summit in February, argues Gautier Uchiyama, founder of Daijobu AI.
We stand on the threshold of a fundamental societal choice regarding AI. In 2025, economic and political forces will have to take a position on sovereign AI. How should AI be integrated into the real economy? Broadly speaking, the answer reveals two approaches. On the one hand, companies are adopting proprietary, standardized solutions—most often from the US AI giants (OpenAI, Google, Anthropic) or their European or Asian competitors. On the other, a growing number of economic actors are equipping themselves with so-called “sovereign” AI.
These sovereign solutions are developed ad hoc for an established use case and installed on private, secure, localized infrastructure. They regularly rely on open-source solutions adapted to one or more specific needs. This is, for example, the choice made by the French government through the Interministerial Directorate for Digital Affairs (DINUM) and its Albert solution, which gives French civil servants secure access to leading open-source generative AI systems.
At the other end of the spectrum, standardized models from the AI giants make it possible to build solutions quickly, with performance often sufficient for a proof of concept or an initial approach. Thanks in part to the immense capital inflows they receive, they offer competitive prices—sometimes close to dumping—which suggests that prices could be rebalanced rapidly.
For decision-makers, choosing between these two approaches is crucial: the sovereign approach is often slower, but it enables the creation of a genuine competitive advantage. Although it sometimes requires substantial investment, it offers an equally substantial return: technological mastery, end-to-end control over data, and greater independence. And as these technologies become more widely accessible, the cost of entry is falling to increasingly affordable levels, including for SMEs!
Building operations on technology you do not own amounts to creating the conditions for your own economic dependence. In the era of “mass agentification” (that is, the growing use of generative AI to perform actions rather than merely converse with users), economic actors’ ability to control the entire technology chain will be a major differentiator. What is the value of a solution whose greatest barrier to entry is available free of charge, or almost, simply by connecting to Anthropic’s Claude? How much trust can be placed in a solution whose strategic data all passes through OpenAI’s servers?
Alongside these strategic considerations are other important issues: specializing AI models rather than using general-purpose models can reduce their size and therefore their energy consumption. LinkedIn, for example, developed its own generative AI (EON) on the basis of an open-source model, achieving a cost-effectiveness ratio 75 times higher than ChatGPT-4 on the tasks evaluated! At scale, these energy savings also represent an initial, albeit insufficient, shift in the environmental impact of AI development.
The issue is not merely economic. It is highly political: is “another AI” possible? How can the conditions be created for the development of AI that is sovereign, but also frugal and open? This question must take its rightful place in the crucial discussions to be held in Paris on February 10 and 11, 2025, at the AI Action Summit. France will host businesses, civil society, and no fewer than 100 countries to discuss the future of AI in our societies.
In terms of public policy, supporting sovereign AI requires developing training for technicians, but also—and perhaps above all—for decision-makers. Technology investment, particularly for smaller businesses, must be facilitated and financed. To build trust and encourage the emergence of localized solutions, protective digital spaces—such as the European digital space—for personal data and intellectual property must also be promoted. International solidarity policies will need to enable developing countries to participate fully in this fourth industrial revolution.
Finally—and this may be the keystone of the work ahead—we must enable open-source and academic innovation to flourish and play its catalytic role. This is the sine qua non for ensuring that the future of AI is built in public, rather than in the secrecy of the laboratories and data centers of a handful of giants.
Gautier Uchiyama is an AI entrepreneur and a specialist in innovation for development