France is going through a potentially decisive period in its history. While the troubled financial context plunges the country into political uncertainty, the rapid rise of artificial intelligence presents both an additional challenge and a sudden opportunity. Properly harnessed, it could help rebuild our economy, strengthen our sovereignty, and offer new and promising prospects for our citizens. To achieve this, it is imperative to combine political vision and technological expertise. At a time when we must decide on the best path to take for the next five years, I found it necessary to take stock of our actions within the framework of the national strategy for artificial intelligence (SNIA).
The first phase of the SNIA (€1.85 billion, 2018-2022) aimed to mobilize research actors around what is called an 'ecosystem,' by bringing together strengths, eliminating redundancies, and structuring research hubs that are thematically relevant and subject to coordinated governance: 4 3IA Institutes, 180 research chairs, more than 300 doctoral programs, and a large supercomputer (Jean-Zay).
The second phase of this strategy, endowed with €1 billion by the France 2030 funds, was launched in mid-2022, and I have been coordinating it since January 2023. Its main focus is the 'dissemination of AI in the economy.' AI can indeed be incorporated into most of our production processes, thus exerting a transformative power over the entire national economic fabric. It is not limited to a mere 'deep tech' and sectoral issue but represents a global lever for competitiveness and growth for France.
The majority of the work in phase 2 was initiated from 2023, following a plan established in 2021. The strategic choices of sectors made at that time concerned embedded AI, frugal AI, and trustworthy AI, complemented by a massive development of AI training offerings.
At the beginning of 2023, during the 'ChatGPT buzz,' France had only one player positioned on generative foundation models: LightOn. Our technological dependency was almost complete. The priority of the first half of the year was to reallocate the SNIA budget to create a new axis dedicated to generative AI. These actions were carried out under time constraints to enable France, at a minimum, to remain competitive technologically and in terms of strategic autonomy. Successful, they allowed us to position ourselves in 2024 among the leaders in the race for generative AI:

  • We moved from one to over a dozen world-class generative AI players between early 2023 and late 2024: Mistral, H, Kyutai, Photoroom, Poolside, Dust, Gladia, etc.
  • France has become the leading European nation in terms of AI investment attractiveness, which had never happened in the digital sphere before. Fundraising also doubled, reaching €3.2 billion in 2023.
  • This European leadership in generative AI attracts the best international AI players who decide to establish themselves in France. In 2024, OpenAI, Google, and Tata Sons joined Cisco, Meta FAIR, Fujitsu, HPE, IBM, Intel, Microsoft, NaverLabs, Samsung, SAP, Uber.
  • This quantitative assessment is accompanied by a strong coverage of the technical field, with Mistral for text generation, Photoroom for images, Kyutai for voice, Aive for videos, H for text-to-action and agent deployment, Dust and LightOn for model adaptation to needs, Artefact and Capgemini for their dissemination in our companies, Giskard, Kili technology and Prism Eval for their evaluation, Pruna AI and ZML for their optimization, etc. Numerous application domains are also covered, such as Aqemia, Bioptimus, Nabla, or ReciTAL for health, Comand.ai, Chapsvision, or Thales for defense and security, Valeo for mobility, and the OpenLLM France collective for education. Generative AI for engineering (automatic code generation with Poolside and Software Heritage, CAD parts generation with Dassault Systèmes, etc.) or sciences (protein and molecule generation by companies mentioned above, material generation with Entalpic, etc.) is also a key area where France has been able to establish itself.
This remarkable positioning concerning the rest of Europe, and in absolute terms in terms of value chain coverage, is also observed on the technical sectors deemed priorities:
  • Regarding embedded AI, French hardware players are among the best positioned in Europe (STMicroelectronics, Sipearl, Kalray, Vsora, Flex.ai, Dolphins Designs, Hawai.tech, Menta, Prophesee, Greenwaves, Upmem, the DeepGreen collective, etc.).
  • In terms of frugal AI, twelve local governments are engaged in developing and deploying energy and data-efficient AI to support their ecological transition.
  • Regarding trustworthy AI, France now holds European leadership (winner of each of the five Testing and Experimentation Facilities for AI tenders, coordinator of the only project aiming to prefigure Union Testing Facilities for AI Act compliance evaluation, the only EU member state to be a member of the AI Safety Network and thus to have an AI Safety Institute). Furthermore, the integrated partnership between the OECD and the Global Partnership on AI, initiated by France in 2023, has become the main international AI governance body.
The effort also focused on multiplying our AI research and training capacity, supported by open-source AI software libraries mobilizing large collectives (Scikit-learn initiated by Inria and now the most used worldwide, DeepGreen, etc.), as well as on accessing all enabling resources for state-of-the-art AI development, notably:
  • Supercomputing: Generative AI training now relies on significant computing infrastructures (supercomputers, cloud) mostly owned by American and Chinese Big Tech. Therefore, it was decided to expand Jean-Zay's capacity, strengthen its support engineering team, whose quality is recognized by the entire ecosystem, and establish an exascale-class supercomputer at TGCC (CEA) as part of the European EuroHPC initiative. Moreover, to secure the presence of AI-optimized GPUs on national territory, we recruited national (Scaleway, OVHCloud, Outscale, Eclairion) and international private investors (€7 billion announced at Choose France 2024). Atos/Eviden remains the main European integrator of public supercomputers, thus ensuring a sustainable market.
  • Data: France holds European leadership in data pooling for AI, including the aspect specific to defending European languages and cultures, through the European ALT-EDIC structure bringing together 19 EU member states. Nationally, the SNIA also supports the preparation of large test databases from our national archives (INA, BnF, Fondation Software Heritage, etc.) for AI model training, in compliance with European data protection and intellectual property laws. Within the Francophonie, we have launched an initiative to build large French-language databases at the Villers-Cotterêts summit in October 2024.
In summary, France is now recognized as Europe's foremost hub for generative AI, having risen from the thirteenth place in 2023 to fifth place in 2024 in the Global AI Index.
We have achieved this result by combining the two approaches usually referred to as 'bottom-up' and 'top-down.'
On the one hand, programmatic consistency to exploit targeted developments over the long term, allowing us to be the first to arrive at applications considered the most commercially and economically advantageous. But also to maintain top-level expertise across the entire spectrum, thus preserving full freedom of action.
On the other hand, flexible management to seize and make the most of, and therefore second, but at lower cost, all the opportunities in an astonishingly prolific sector.
In the current political mindset, we prioritize the reactivity, creativity, and proposal strength of private enterprise, without yielding to any 'laissez-faire.' Our official governance structure is streamlined and fueled by the reflection and discernment of several steering committees in which our best specialists participate at the highest level of the State.