A policy analysis from June 2026
When Emmanuel Macron stood at the Grand Palais in Paris in March 2018 and launched France’s national AI strategy under the banner “AI for Humanity,” he was drawing on a report by mathematician and Fields Medal laureate Cédric Villani that argued France had the research talent, the institutional architecture, and the industrial ambition to become a genuinely world-class AI power. The plan called for €1.5 billion in investment by 2022, a network of research institutes, a major supercomputing push, and an ethical framework for AI development that positioned France as both a competitor and a conscience in the global AI race.
Eight years later, the strategy has passed through three distinct phases, survived changes of government, a global pandemic, and severe fiscal pressure, and produced results that are simultaneously more impressive and more mixed than its architects anticipated. France has risen from sixth place in the Global AI Index in September 2024 to fifth in September 2025. It ranks third in the world for AI research and training. It has produced Mistral AI, the only European large language model company capable of competing on the frontier. It hosted the Paris AI Action Summit in February 2025 — the most significant multilateral AI gathering since the UK’s Bletchley Park summit — and announced €109 billion in investment commitments that positioned France as the leading destination for foreign AI investment in Europe for the fifth consecutive year.
And yet. In November 2025, France’s Cour des comptes — the national Court of Auditors — published a detailed assessment of the national AI strategy that praised the research achievements while cataloguing a longer list of persistent shortcomings: slow public sector adoption, insufficient budget execution in the second phase, a widening lag in AI integration among SMEs, and a transformation of public services through AI that had been, in the Cour’s own measured language, “very disappointing.” The gap between France’s headline ambition and its operational delivery is the central tension of its AI story heading into the second half of this decade.
What the Strategy Actually Set Out to Do
France’s national AI strategy has unfolded in three phases, each with a distinct character and a distinct institutional logic.
The first phase, running from 2018 to 2022, was explicitly a research-first investment. The Villani report’s argument was that France had strong AI talent in its grandes écoles and CNRS laboratories but lacked the institutional infrastructure to retain it and commercialise it. The response was a network of four Interdisciplinary Institutes for Artificial Intelligence (3IA) — located in Grenoble, Nice, Paris, and Toulouse — established from 2019 with an initial budget of €80 million. These institutes were designed to connect universities with industry, fund doctoral and postdoctoral research, and produce researchers with both technical depth and commercial awareness. By 2024, the 3IA network had expanded into nine Centres of Excellence — IA Clusters — with combined funding of approximately €350–360 million. The Jean Zay supercomputer at IDRIS in Orsay, launched in 2020 and subsequently upgraded via a €40 million France 2030 investment, became the physical anchor of France’s public compute capacity.
The second phase, announced in November 2021 and running through 2025, shifted the emphasis from research to economic diffusion. The target was to take what the first phase had built — the talent, the compute, the research network — and push it into the broader economy, particularly into SMEs and the public sector. The budget allocated was €1.1 billion. The priority sectors included healthcare, energy, transport, and generative AI. The AI Booster initiative was launched to support SME adoption, and France’s Bpifrance ran calls for generative AI demonstrators and sector-specific projects under the France 2030 umbrella. France counts 16 European Digital Innovation Hubs, several of them AI-focused, as part of the adoption infrastructure.
The third phase was announced by Macron at the Paris AI Action Summit in February 2025, backed by €2.5 billion in direct public investment and framed by the €109 billion in international investment commitments that gave the summit its headline numbers. This phase is explicitly about compute sovereignty, frontier model development, and the kind of industrial AI deployment that could make France not just a research leader but an economic one.
What Is Working: The Genuine Achievements
Research and the Startup Ecosystem
The research investment of the first phase has produced results that are difficult to dispute. France ranks third globally for AI research and training. More than 4,000 French researchers are working on AI. The number of doctoral theses in AI has increased consistently since the strategy’s launch. The 3IA institutes and IA Clusters have created genuine centres of gravity for AI talent in Paris, Grenoble, Nice, and Toulouse that are beginning to resemble the research-to-startup pipelines that made Cambridge and Munich successful technology regions.
The startup ecosystem reflects this. The number of French AI startups has doubled since 2021: more than 1,000 are active in 2025, and they raised nearly €2 billion in funding in 2024. Sixteen French startups valued at over $1 billion incorporate AI into their core value proposition. Mistral AI — founded in April 2023 by Arthur Mensch, Guillaume Lample, and Timothée Lacroix, with alumni from DeepMind and Meta — raised a €1.7 billion Series C in September 2025 at an €11.7 billion valuation, with ASML as the lead investor at 11%. Annual recurring revenue has crossed $400 million, with Mensch guiding toward €1 billion in revenue for 2026, roughly 60 percent from European clients seeking alternatives to American platforms.
France’s AI strategy can take partial credit for creating the conditions in which Mistral was possible: the talent pipeline, the research network, the regulatory environment that treated open-source models more favourably, and the culture of technical ambition that the grandes écoles system produces. Whether it deserves more credit than that is a more contested question.
France’s Sovereign Infrastructure Push
The Paris summit’s most durable contribution to the national strategy is likely to be the infrastructure commitments it crystallised. France identified 35 sites capable of hosting AI data centre projects, some up to 1GW in capacity. Macron’s framing was deliberately competitive: France’s decarbonised, abundant electricity supply — underwritten by its nuclear fleet — its streamlined planning procedures, and its central European position gave it structural advantages that justified the ambition.
Mistral AI’s “Mistral Compute” platform, announced in early 2026, features 18,000 NVIDIA Grace Blackwell GPUs deployed across a 40MW data centre in Essonne. Jean Zay, already one of Europe’s most eco-efficient machines — its waste heat warms more than 1,500 homes in the Saclay area — has been extended with AMD Instinct MI430X GPUs and SiPearl European-designed CPUs, a deliberate “silicon sovereignty” signal differentiating the French public stack from pure NVIDIA dependence.
France has also built a distinctive data sovereignty framework through the SecNumCloud certification, operated by ANSSI. The framework requires data to be physically hosted on French territory, but does not exclude foreign cloud providers — Amazon and Microsoft can offer SecNumCloud services if they comply with French security requirements. As one analysis put it, sovereignty is no longer defined by the nationality of the company, but by its ability to meet state requirements. This is a pragmatic and commercially viable approach to sovereignty that has compensated for the relative scarcity of domestic cloud infrastructure.
The Health Data Hub, despite its origins in controversy — French health data was initially stored on Microsoft Azure servers, triggering significant political and legal pushback — has evolved into a more coherent governance model. France is now an active participant in the European Health Data Space and the 1+ Million Genomes initiative, and the OECD’s 2025 country report on France’s AI strategy notes that France’s centralised health system has facilitated robust health data governance and compliance with European regulations.
France’s Role in Shaping EU AI Governance
France has been one of the most consequential single-country voices in shaping the EU AI Act and the broader EU AI Continent Action Plan. French negotiators successfully preserved provisions benefiting French companies: general-purpose AI models below threshold compute requirements face lighter obligations, open-source models are largely exempted from onerous requirements, and sectoral applications in regulated industries face risk-proportionate rather than blanket burdens. The Paris AI Action Summit itself served as the political moment in which the EU AI Continent Action Plan was formally framed, with Ursula von der Leyen using the occasion to announce the InvestAI facility and signal the shift from AI regulation to AI industrial policy.
France established INESIA — the National Institute for the Evaluation and Security of Artificial Intelligence — in February 2025 to co-ordinate national actors on AI security risk analysis, regulatory implementation, and performance testing. The CNIL, France’s data protection authority, has developed an AI division providing pre-market guidance to companies developing AI systems, positioning French regulatory expertise as a commercial asset rather than a purely bureaucratic function.
Where It Is Stuck: The Persistent Gaps
The Cour des comptes report, published in November 2025, is the most authoritative public assessment of where the strategy has fallen short, and it is candid on multiple fronts.
The Adoption Deficit in the Broader Economy
The second phase of the strategy — the €1.1 billion diffusion programme — was meant to take AI from the research institutes and frontier companies into the broader French economy. By the Cour’s own assessment, this has not happened at the intended scale. The priority of supporting business demand for AI solutions received only very modest measures. The expected acceleration and massification of AI in the economy has not taken place. The delay in adapting initial and continuing training programmes to AI has not been made up.
This is not purely a French problem — the EU’s own Apply AI Strategy acknowledges that only 13.5 percent of EU businesses and 12.6 percent of SMEs are using AI technologies — but it is a particularly pointed one for a country that spent eight years building an AI research and startup ecosystem on the explicit premise that it would translate into economic diffusion. France’s 2023 independent AI commission report, “AI: Our Ambition for France,” was already warning that over the past decade, AI integration in French companies had been “slower and shallower overall than in the USA, the UK or Scandinavia.” That warning has not yet been answered.
The AI Booster programme, run through Bpifrance, is the primary SME support vehicle, providing advisory support and subsidised access to AI tools for companies with between 10 and 2,000 employees. France’s 16 European Digital Innovation Hubs, several of them AI-focused, have been refocused as AI experience centres under the EU’s December 2025 reforms. But the gap between these instruments and the scale of the adoption challenge — hundreds of thousands of French SMEs making investment decisions about AI with limited information, limited skills, and genuine regulatory uncertainty — remains significant.
Public Sector Transformation: A Troubled Record
The Cour des comptes described the transformation of public action through AI as having been “very disappointing,” despite isolated initiatives. There are exceptions worth noting: the Ministère des Finances and the DGFIP have implemented AI-driven tools for fraud detection, undeclared property identification, and public spending control. Air France has used its Prognos AI-driven predictive maintenance system since the early 2000s, with more recent applications in baggage handling and eco-piloting. The France Travail employment agency has run a collaboration with Mistral AI on public service automation.
But these remain isolated proof points rather than systemic transformation. The broader challenge — getting AI into procurement processes, administrative decision-making, service delivery, and regulatory functions across hundreds of public bodies — requires not just technology but governance frameworks, civil service capability, and institutional appetite for change that no amount of infrastructure investment automatically provides. The AI Cafés programme, which aims to reach 2 million French citizens with AI literacy by 2027, addresses the awareness gap but not the deployment one.
The Investment Pledges: Ambition and Reality
The €109 billion in investment commitments announced at the Paris summit generated enormous international attention. The number requires careful reading. Much of it came from a single commitment by Abu Dhabi’s MGX sovereign wealth fund of €50 billion — a pledge with no legal obligation to arrive on any particular timeline. By November 2025, nine months after the summit, the Cour des comptes and independent analysts had confirmed only approximately €26 billion in concretely committed investment, though the Élysée maintained that €90 billion was “secured.” The Choose France summit in June 2026 added a further €93 billion in pledges — including €45 billion from SoftBank for a 3.1GW data centre capacity project in Hauts-de-France — though, again, these are announced rather than contracted figures.
This is not to dismiss the numbers. The pipeline of genuinely planned and under-construction AI infrastructure in France is real and growing. But the gap between summit announcements and binding commitments is a structural feature of French AI strategy communications that independent analysts have consistently noted, and it matters for how policymakers and investors should interpret the headline figures.
Reskilling: Targets Outrunning Reality
The France 2030 programme set an objective of 100,000 additional workers per year trained in AI applications by 2025. Early execution has been slower than this target, though the grandes écoles executive programmes at Polytechnique, HEC, and others have collectively reached tens of thousands of working professionals. The CIFRE contract system — which embeds PhD researchers within companies while funding their doctorates — has been expanded for AI, targeting approximately 1,500 AI CIFRE contracts per year. The AI Skills Academy, launched at EU level in 2025, is being integrated into France’s national upskilling architecture.
The deeper challenge is not the volume of training, but its distribution. AI skills remain heavily concentrated in the Paris tech ecosystem and the grandes écoles alumni network. Whether the reskilling investment reaches workers in manufacturing, agriculture, healthcare, and local government — the sectors where AI adoption would have the broadest economic and social impact — is an open question that France shares with every major AI strategy in Europe.
The Governance Architecture: Stronger Than It Looks
One of the less-discussed achievements of the French AI strategy is the governance infrastructure it has built around a very complex policy problem. The national AI co-ordinator role (“Monsieur IA”) has provided interministerial coherence that more fragmented systems lack. The CNIL’s AI division has developed genuine pre-market guidance capacity. INESIA, established in February 2025, adds a dedicated security evaluation function. The Conseil National de l’IA et du Numérique (CIAN) — the rebranded National Digital Council — provides a high-level forum for industry, civil society, and government dialogue.
France relies on sector-specific rules and the EU AI Act rather than a single domestic statute, which gives it flexibility but also means that enforcement architecture is still being assembled. The government is finalising a decree to designate national competent authorities under the AI Act, with debate pointing toward a multi-authority model: CNIL overseeing fundamental rights, and sector-specific regulators such as ARCOM (media) and ACPR (financial services) handling conformity assessments in their domains.
Environmental considerations have been integrated systematically into France’s AI policy in a way that distinguishes it from most peer nations. Environmental criteria are embedded in calls for AI projects. All France 2030 training and research projects must address AI’s environmental dimension. The Jean Zay supercomputer’s waste heat reuse model has become a reference point for sustainable compute design. This is not merely symbolic: France’s nuclear electricity supply gives it genuine structural advantages in building AI infrastructure with lower lifecycle emissions than coal or gas-dependent alternatives.
What Phase Three Needs to Deliver
The third phase of France’s AI strategy — now operational — has its priorities reasonably well defined by the lessons of what came before. The Cour des comptes made ten recommendations in its November 2025 report, the most consequential of which are: establishing a dedicated secretariat for cross-ministry co-ordination; using public procurement strategically to foster AI growth rather than defaulting to foreign platforms; accelerating AI integration in education at all levels; and setting clear, measurable targets for business AI adoption against which the strategy’s success can be evaluated.
The infrastructure investments coming online through 2026 represent a genuine opportunity. Mistral’s Essonne cluster, the Jean Zay extension, the Grenoble Gigafactory groundbreaking, and the broader data centre pipeline add up to a sovereign compute ecosystem that did not exist in anything like this form three years ago. If access to that infrastructure can be directed toward the French SME base — through the AI Booster, the Digital Innovation Hubs, and the sectoral flagships of the EU’s Apply AI Strategy — then the link between research excellence and economic diffusion that Phase Two failed to forge might yet be built.
The copyright question also requires resolution. French publishers and authors filed suit against Meta in the Paris Judicial Court in March 2025, alleging unauthorised use of their works to train generative AI models. The EU-level text-and-data mining debate remains unresolved, with France’s creative industries — one of its most globally significant economic sectors — watching closely. How France navigates the tension between protecting its creative economy and enabling its AI developers to train on the data they need will be a recurring challenge in this phase.
France as a European AI Leader: The Bigger Picture
Ranked second globally and first in Europe in the Oxford Insights Government AI Readiness Index, France has consolidated a position that was far from guaranteed when the Villani report was published. It is the leading destination for foreign AI investment in Europe for the fifth consecutive year. It has produced the continent’s only frontier-scale large language model company. Its research institutions remain world-class. Its regulatory contributions have shaped the EU framework in ways that benefit the European AI ecosystem as a whole.
France’s “AI for Humanity” strategy was one of the first national AI strategies published anywhere in the world. Five phases and eight years later, the version 2.0 framing of Phase Three is an acknowledgement that the ambition of the original plan — research excellence translating seamlessly into economic leadership — was more complicated to execute than the 2018 moment of enthusiasm suggested. The foundations are strong. The adoption gap is real. The infrastructure investment is accelerating. Whether this phase can finally close the distance between France’s research capability and its economic AI deployment is the central question of the next five years.
As the Cour des comptes concluded: it is on this condition that France, in close co-operation with the European Union and local authorities, will continue on its path to excellence and succeed in embracing all the dimensions that the AI revolution is set to affect, in the service of the common good and with a view to guaranteeing national sovereignty. The ambition is unchanged. The delivery mechanisms are better than they were. The pressure to demonstrate results rather than announce them has never been higher.
Key References and Further Reading
- France National AI Strategy Overview — EU Digital Skills and Jobs Platform
- Make France an AI Powerhouse (February 2025) — Élysée Palace
- The National Strategy for Artificial Intelligence: Court of Auditors Report (November 2025) — Cour des comptes
- AI: Our Ambition for France — Independent Commission Report (2023) — French Government
- Progress in Implementing the EU Coordinated Plan on AI: France (2025) — OECD
- France Bolsters National AI Strategy With NVIDIA Infrastructure (2026) — NVIDIA Blog
- AI in France: Betting on Sovereignty Rather Than Racing for the Most Powerful Models — Oxford Insights
- France as a European Leader in AI: Lessons for Poland — Casimir Pulaski Foundation
- France’s AI Sovereignty Push: Mistral, SecNumCloud and Sovereign Infrastructure — Introl
- AI Watch: Global Regulatory Tracker — France — HSF Kramer
- Macron Announces €93 Billion at Choose France Summit (June 2026) — France 24
- The EU AI Continent Plan: Europe’s Bid for Sovereignty — AI Insider
