Thirty years on the same plateau. In 2023, France devoted 2.18% of its GDP to R&D [1]. The United States and China each crossed the 1 trillion dollar mark in spending [4].

The research tax credit costs 8.06 billion euros in public finances in 2025 [6]. The ANR’s competitive budget represents less than one-sixth of that [12]. The decision point for 2027 is clear: reserve the tax credit for SMEs and mid-market firms, redirect the remainder toward competitive project funding. The alternative is a decline that international rankings will make hard to dispute.

Thirty years on a plateau, and the gap widens

In 2023, R&D conducted on French territory came to 61.5 billion euros, or 2.18% of GDP [1]. That is roughly the same ratio as in 2000. Between 2013 and 2023, domestic R&D spending grew at an average rate of 0.8% per year in real terms. GDP grew at 1.2% [1]. The relative effort is thus declining over the long term.

In 2022, domestic R&D spending reached 58.9 billion euros [2]. The effort represented 2.22% of GDP, as in 2021. Since then, it has fallen.

In 2023, R&D spending declined 0.5% in real terms compared to 2022 [1]. For 2024, provisional data indicate a 1.3% increase in real terms [1]. The effort would reach 63.6 billion euros. But GDP growth neutralizes this gain: the effort remains at 2.18% of GDP, as in 2023 [1].

International comparison is harsh. In 2023, France stands at 2.2% of GDP [3]. South Korea is at 5.0%, Sweden at 3.6%, the United States at 3.5%, Japan at 3.4%, Germany at 3.1%, the United Kingdom at 2.8% [3]. The target set by the European Union in the Europe 2020 strategy was 3.0% of GDP [3]. France has never reached it.

On a global scale, in purchasing power parity, China would have surpassed the United States in 2024 [4]. The two countries lead the race with each spending more than 1 trillion dollars on R&D. The European Union reaches nearly 600 billion [4]. R&D spending increased 3.4% in the United States in 2024. In the European Union, growth stood at 0.4% [4].

The internal structure reveals a second weakness. In 2023, businesses carry out 66% of domestic R&D spending, government and universities 34% [1]. But the private share is highly concentrated.

Large enterprises account for 54% of private R&D spending [1]. Mid-market firms make up 25%. SMEs represent 84% of businesses conducting R&D but account for only 20% of it [1].

Industry calls the shots. Manufacturing accounts for 66% of business R&D [1]. The dominant sectors are miscellaneous industries (13%), automobiles (11%), and aerospace (10%) [1].

Geography is equally concentrated. In 2023, the Île-de-France region represents 39% of French R&D [1]. Auvergne-Rhône-Alpes accounts for 16%, Occitanie for 12% [1].

On researchers, France still holds an honorable rank. It places 6th globally in researcher density per thousand jobs [11]. It attracts more young foreign researchers than the European or OECD average [11].

But between 2012 and 2021, research employment grew faster in the private sector than in the public sector. In the private sector, average annual growth reached 2.3% [11]. In the public sector, it was limited to 1.1% [11]. For researchers alone, the gap was even starker: 3.1% in the private sector versus 1.5% in the public sector [11].

On patents, the signal is more positive in the short term. In 2025, 16,807 applications were filed with the INPI (National Institute of Industrial Property), a 8.7% increase year-over-year [10]. Nine public establishments appear in the top 50 filers, including the CEA and the CNRS in 7th position [10]. But filings are declining in pharmaceuticals (−2.4%) and biotechnology (−9.9%) [10].

The tax credit absorbs public money without guaranteeing breakthrough research

Public research funding rests on two pillars. The first: direct funding of public bodies (CNRS, Inria, universities, CEA) and grant competitions from the ANR (National Research Agency). In 2024, the ANR’s intervention budget stood at 1.24 billion euros, up 3.9% from 2023 [12].

The second pillar is far more substantial. The research tax credit (CIR) is an automatic tax reduction: a company declares R&D expenses and receives a 30% reimbursement, with no selection. Its cost rose from 1.8 billion euros in 2007 to 4.45 billion in 2008 [6]. In 2024, it reached 7.65 billion [6]. The public finance shortfall has exceeded 8.06 billion in 2025 [6].

This amount exceeds the ANR’s intervention budget by more than six times [6][12]. And the breakdown reveals the device’s logic. SMEs form 81% of tax credit beneficiaries but represent only 30% of declared spending and 31% of the credit [9]. The bulk benefits large enterprises, which would have incurred these expenses anyway.

The result in 2025: large enterprises receive massive reimbursements for R&D they would have funded without aid, while researchers in public laboratories struggle to secure project funding [6][9]. The mechanism favors applied R&D in established sectors (automotive, aerospace, pharmaceuticals). Fundamental research and breakthrough technologies come second.

The 2020 research programming law was meant to correct the underfinancing of public bodies. The balance sheet is mixed. The planned increase for 2025 was 500 million euros more than 2024 [7].

The 2025 budget bill concretized only about one-third of it [7]. Of the 501 million planned for the three affected programs, less than 169 million was appropriated [7]. Universities were asked to draw on their reserves to contribute to budget savings [8].

AI redistributes the cards by amplifying capacity gaps

The rupture is technological and structural. Generative AI changes both the content of R&D and the distribution of its fruits.

Artificial intelligence carries considerable growth potential, capable of automating both tasks and the production of ideas [13]. But its positive effects are hampered by excessive market concentration. Insufficient competition among innovators kills innovation [13]. The report co-chaired by Philippe Aghion in March 2024 defends a sovereign AI strategy with a “France & AI” fund of 10 billion euros, investments in research, training, and computing capacity [13]. This document remained without direct political follow-up.

In 2023, France devoted 1.8 billion euros to R&D in artificial intelligence in businesses [17]. This represents 4.5% of domestic business R&D spending [17]. SMEs devote a proportionally larger share than large groups: 7.7% of their R&D spending versus 4.2% for mid-market firms and 3.5% for large enterprises [17]. Their absolute weight remains marginal.

The question of how gains are distributed is distinct from that of wealth creation. The gains of technical progress accrue to elites far more than to workers, according to the analysis of Acemoglu and Johnson [14]. Generative AI, in this reading, focuses on substituting workers rather than complementing them. Call center automation sometimes proves less productive than human labor—simply cheaper [14]. This orientation is an institutional choice, not a technological inevitability.

Established interests constitute the most durable obstacle. Concentration of enterprises, lobbying power, and rent-seeking suppress competition and R&D investment [15]. History shows a recurrent tension: when institutions fail to adapt to technological change, stagnation follows [15]. A landscape of large enterprises benefiting from the tax credit, little exposed to competition on the AI frontier, could exactly reproduce this pattern.

Reserve the tax credit for SMEs and mid-market firms, redirect large enterprises’ share toward the ANR

The architectural decision consists of two tiers. The tax credit ceases to be a universal declarative credit: it becomes a right reserved for SMEs and mid-market firms, which form 81% of its beneficiaries [9]. The share captured by large enterprises, roughly 70% of the credit [9], must shift toward the ANR in the form of competitive project funding. It becomes the endowment of a peer review body, open to any entity able to demonstrate it works on the scientific frontier: a company, a laboratory, a mixed consortium. The brief “France Invents, Others Industrialize” details the legal boundary of this architecture.

Two things change in nature. Who decides. And who has the right to access public money.

The choice is clear-cut. A large enterprise that declared 50 million euros in automotive R&D no longer has an automatic claim on the Treasury. It competes before a peer jury, like a biotech SME or a CNRS laboratory. In 2025, the opposite situation prevails [6][9].

What this closes is real. Sectors that live off the current tax credit (automotive, aerospace, established pharmaceuticals) lose a predictable rent and a balance sheet advantage [6][9]. The finance departments of large groups can no longer budget the credit as an accounting line item. Enterprises that invested in R&D only because the state reimbursed 30% without conditions will exit the game. That is the intended result.

What this opens is symmetrical. The ANR success rate rises mechanically. Public scientific careers become fundable again. SMEs, which represent 84% of businesses conducting R&D but capture barely 31% of the tax credit [9], retain the tax credit and now compete at the same ANR table as large groups. Regions like Occitanie or Auvergne-Rhône-Alpes can anchor local consortiums without going through the Parisian headquarters of current beneficiaries [1].

Allocation authority passes from tax authorities to the ANR, whose intervention budget quadruples at a stroke [12]. It is an institution that changes scale. The orientation of technical progress is an institutional choice, not a necessity [14]. Concentrating decision-making in a peer review body rather than in the tax declaration of a large group poses this choice.

Sources

[1] MESR-DGESIP/DGRI-SIES, “The State of Higher Education, Research and Innovation in France,” no. 19, 2025 edition, https://publication.enseignementsup-recherche.gouv.fr/eesr/FR/T923/l_effort_de_recherche_et_developpement_en_france/ (accessed 09/08/2026).

[2] MESR-SIES, “France’s Domestic R&D Spending 2022,” Flash Note no. 23, September 2024, https://www.enseignementsup-recherche.gouv.fr/sites/default/files/2024-09/nf-sies-2024-23-34670.pdf (accessed 09/08/2026).

[3] Eurostat, “Total R&D expenditure 2024: EU statistics,” December 2025, https://education.newstank.fr/article/view/425350 (accessed 09/08/2026).

[4] OECD, “Foundations for Growth and Competitiveness 2026,” DIRD data 2024 in PPP, April 2026, https://www.oecd.org/content/dam/oecd/fr/publications/reports/2026/04/foundations-for-growth-and-competitiveness-2026_f68a156b/df51b240-fr.pdf (accessed 09/08/2026).

[5] OECD, “Science, Technology and Innovation: OECD Perspectives 2025,” October 2025, https://www.oecd.org/content/dam/oecd/fr/publications/reports/2025/10/oecd-science-technology-and-innovation-outlook-2025_bae3698d/a7207a31-fr.pdf (accessed 09/08/2026).

[6] François Ecalle, Fipeco, “Public spending in favor of research and innovation,” updated 03/09/2025, https://www.fipeco.fr/fiche/Les-d%C3%A9penses-publiques-en-faveur-de-la-recherche (accessed 09/08/2026).

[7] Senate, Report on the 2025 Budget Bill, “Research,” 2024, https://www.senat.fr/rap/a24-149-51/a24-149-51_mono.html (accessed 09/08/2026).

[8] Senate, Report on the 2026 Budget Bill, “Higher Education,” 2025, https://www.senat.fr/rap/l25-139-323/l25-139-323_mono.html (accessed 09/08/2026).

[9] Senate, Report on the 2024 Budget Bill, “Reimbursements and Tax Reliefs (CIR),” 2023, https://www.senat.fr/rap/l23-128-327/l23-128-3278.html (accessed 09/08/2026).

[10] INPI, “2025 Patent Filers Rankings,” March 2026, https://www.inpi.fr/a-la-une/palmares-2025-des-deposants-de-brevets (accessed 09/08/2026).

[11] MESR-SIES, “The State of Scientific Employment in France,” 2025 edition, https://www.enseignementsup-recherche.gouv.fr/fr/l-etat-de-l-emploi-scientifique-en-france-2025 (accessed 09/08/2026).

[12] ANR, “2024 Budget,” https://anr.fr/fr/lanr/nous-connaitre/budget/ (accessed 09/08/2026).

[13] Philippe Aghion, “Resetting the Innovation Clock: Endogenous Growth through Technological Turnover,” 2025, https://www.assemblee-nationale.fr/dyn/17/comptes-rendus/cion-eco/l17cion-eco2526012_compte-rendu.pdf; Report “France & AI” co-chaired with Anne Bouverot, March 2024 (accessed 09/08/2026).

[14] Daron Acemoglu and Simon Johnson, Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity, PublicAffairs, 2023; analytical summary, MIT Economics, December 2024, https://economics.mit.edu/news/daron-acemoglu-what-do-we-know-about-economics-ai (accessed 09/08/2026).

[15] Carl Benedikt Frey, How Progress Ends: Technology, Innovation, and the Fate of Nations, Princeton University Press, 2025, https://press.princeton.edu/books/hardcover/9780691233079/how-progress-ends (accessed 09/08/2026).

[16] MESR-SIES, “R&D in Artificial Intelligence in Businesses,” in EESRI no. 19, 2025, https://publication.enseignementsup-recherche.gouv.fr/eesr/FR/T603/la_r_d_en_intelligence_artificielle_dans_les_entreprises/ (accessed 09/08/2026).