A policy brief unveiled in early July outlines for the first time the path to French strategic independence in artificial intelligence. Its assessment is brutal: staying in the global race would require a budget effort of historic proportions, with no guarantee of success.
A Brief Breaking with the Comfort of Followership
Published on July 8 by Le Grand Continent, the brief dubbed “Operation Prometheus” proposes a three-year plan to equip France and its partners with a frontier AI laboratory capable of competing with the best American labs. The initial diagnosis leaves no room for ambiguity:
« Importing our artificial intelligence flows amounts to depending on others for a growing share of our productivity, our industrial power, and our security. » - Le Grand Continent
The verdict is clear: France currently has no frontier model, meaning no artificial intelligence at the highest level of global power. The authors of the brief point to a concrete warning sign: recent US export controls on Anthropic’s best models illustrate the vulnerability of a country that does not control its own technological chain.
An Economic Engine Already at Work Across the Atlantic
The urgency is not just a matter of future forecasting. According to the brief, AI is already, and by far, the main driver of American growth. In the most advanced companies, AI spending relative to payroll already exceeds 10%, a proportion set to grow to the point where it could one day exceed the share allocated to salaries themselves, according to Le Grand Continent.
This dynamic is reshaping a geopolitical balance of power where two powers hold the keyboard:
- the United States, which controls access to the most advanced models and does not hesitate to restrict their export;
- China, the second pole of the technological frontier;
- France, the only country identified by the brief as currently meeting the conditions to become the third frontier nation - provided it is made an absolute priority.
The Bill: Up to $700 Billion Over Three Years
The proposed costing gives a sense of the scale of the effort demanded of the nation. The targeted computing power trajectory is as follows:
- 2 gigawatts (GW) in 2027;
- 7 GW in 2028;
- 12 GW in 2029, equivalent to the level of today’s major American laboratories.
The associated cost is skyrocketing: $170 billion as early as 2027, then over $300 billion in 2029, for a cumulative total of around $700 billion over three years, according to the brief. Compute - meaning the computing power infrastructure - alone accounts for 95% of the expenditure.
Relative to national wealth, the effort would represent between 4.5% and 8% of French GDP, including 1.5% of public investment per year. The brief does not hide the significance of such a choice:
« This would be a central budgetary decision in the mandate of the future President of the Republic. » - Le Grand Continent
Funding would not rely solely on French public finances. The authors envision contributions from private capital, but also from middle powers sharing an interest in escaping dependence on the Sino-American duopoly, at a time when these two powers are gradually shutting off the AI taps to their clients.
A Two-Tiered Architecture: Science and Infrastructure
To achieve this ambition without the state substituting itself for private initiative, the brief proposes an architecture in two distinct blocks:
- An autonomous scientific laboratory, in which the public sector would hold only a minority stake of 25%, while retaining strategic control mechanisms;
- An infrastructure program - compute, energy, real estate - managed directly by the state, backed by a special, fast-track “Prometheus Act” designed to speed up administrative procedures, complemented by a financial planning component.
This ambition also requires building a tight-knit but world-class team: the brief aims to attract the world’s top researchers to a workforce of 1,700 people, according to Le Grand Continent.
The Sinews of War: Energy, Land, and Public Acceptance
Beyond financing, the brief identifies structural bottlenecks that public debate is only beginning to grasp. The construction of data centers mobilizes considerable energy and land resources, a reality that the magazine Challenges describes as the “sinews of the AI war,” pointing out that this government priority is often accompanied by a lack of consultation with local populations and unaddressed local environmental impacts.
The points of tension identified by the brief include:
- the sheer scale of the budgetary cost in a context of already strained public finances;
- the political acceptability of an unprecedented public investment effort;
- the availability of the energy required to power the computing centers;
- the mobilization of land, often in conflict with environmental issues and local acceptance.
A Choice of Mandate, Not a Technical Project
The brief thus reframes the debate on artificial intelligence where it belongs: not merely in the realm of technological innovation, but in that of civilizational choice and sovereignty. Deciding to invest several percentage points of GDP over three years in computing infrastructure means choosing, for the next five-year presidential term, between two trajectories: organized dependence on foreign powers, or the agreed-upon effort of strategic independence.
The bet is risky, the cost is certain, but the alternative - the absence of choice - could prove even more costly: that of a France, and a Europe along with it, permanently placed under the technological tutelage of digital empires it failed or refused to confront in time.
