Europe Regulates a Digital Market It Did Not Build

Each month, 83% of European companies’ digital purchases go to American actors. Cumulative spending reaches 264 billion euros per year [1]. The 2027 reckoning is here: continue managing dependence, or finally decide to build the market that regulations presuppose.

Thirty Years of Concentration Reach a Tipping Point

Between 2017 and 2025, the combined market share held by the five largest digital multinationals more than doubled, rising from 21% to 48% of the global market [2]. Their share of total assets climbed from 17% to 35% over the same period [2]. Two structural factors drive concentration in platform markets: network effects and economies of scale.

The more users a platform has, the more it attracts, and the larger it grows, the less its unit costs rise. Massive data ownership now adds a considerable barrier to entry. Jean Tirole formalized this mechanism starting in the 1980s: without regulation, prices become unjustified, new entrants are excluded, and innovation stalls [3].

The rise of generative AI reinforces these concerns. In 2025, Microsoft and Google dominate the value chain and consolidate their positions by partnering with start-ups like OpenAI [4]. Generative AI requires massive computing power, chips, cloud services, talent, and data. All of this is controlled by technology giants [4].

For 2025, Alphabet, Meta, Amazon, and Microsoft announced together more than 320 billion dollars in investment spending, primarily dedicated to data centers for AI [4]. Meta alone targets between 60 and 65 billion dollars, nearly double 2024 [4]. No new entrant can keep pace without the backing of a giant.

On productivity, the gap has been documented for three decades. In 2019, hourly productivity in the eurozone represented 82% of that in the United States, compared to 98% in 1995 [5]. This slippage is largely attributable to underperformance in digital sectors. Lacking investment in the digital assets on which AI relies, its implementation will not deliver the expected productivity gains, according to OFCE [5].

In 2026, digital technology adoption by French companies remains below the European average [6]. The number of information technology specialists is experiencing slow growth there. Digitization of public services stalled between 2020 and 2025 [6].

France is among the most generous countries in tax incentives for research and development, alongside the United Kingdom, according to the OECD [4]. Yet its R&D effort has stagnated over the past twenty years, while it progressed in Germany, the United States, and especially China [4]. The research tax credit costs 7 billion euros per year. It mainly serves to offset the higher labor costs in France than elsewhere, without directing innovation toward digital uses that will drive tomorrow’s productivity [4].

The DMA Sanctions, the AI Act Constrains, But Market Structure Resists

Europe has produced since 2022 a regulatory corpus without global equivalent: the Digital Markets Act (DMA), the Digital Services Act (DSA), the GDPR, the AI Regulation (AI Act). Initial effects are visible. The European Commission imposed a 890 million euro fine on Google under the DMA in 2026 [7].

These sanctions add to a long list of antitrust disputes with the European Union since 2017. Total fines imposed on the group now reach 10.38 billion euros [7]. Apple was fined 500 million euros, Meta 200 million euros [7].

These amounts are impressive in absolute value. Measured against the revenues of the giants in question, they remain marginal [8]. The Commission’s annual report on DMA implementation, published in May 2026, tacitly acknowledges this [8]. For these enterprises, the cost of non-compliance in Europe does not yet exceed the cost of inaction [8].

Regulation after the fact, even if well-conducted, is insufficient when network effects have already locked down the market. This is the structural rupture Tirole identifies: he insists on smarter regulation, not bulkier [3].

In their January 2026 contribution, Tirole, Bergeaud, and Loesekrug-Pietri clarify where the main barrier lies [9]. Europe’s best start-ups cannot flourish sustainably when each country protects its national champions and when administrative barriers prevent the emergence of genuine European champions [9].

The AI Act, gradually taking effect since 2025, reveals another imbalance. European SMEs are required to document, map, and train their teams under threat of theoretically crushing sanctions for their size [8]. The strengthening of Commission powers targets essentially a restricted number of actors, overwhelmingly American.

These actors have the legal teams and resources to absorb the cost of compliance. A regulation designed to constrain giants thus creates two speeds. The economic theory of regulation has documented this trap: complex regulation can reinforce dominant positions by raising the cost of entry for smaller competitors [3].

Fragmentation of the internal market aggravates the problem. In 2026, 60% of obstacles identified in 2002 persist, according to the European Court of Auditors special report [10]. Only 20% of service provision within the EU is cross-border [10]. Information technology services, though immaterial by nature, reached only 15% in 2023 [10].

The Budgetary Reading of Regulation Without Leverage Effect

The French national roadmap includes 33 measures with a total budget of 18.6 billion euros [6]. Of this total, 11.1 billion are financed by public budgets, or 0.38% of GDP [6]. By comparison, the public deficit reached 152.5 billion euros in 2025, or 5.1% of GDP [11]. The budget constraint is real. It outlines a precise risk: France invests more in regulatory compliance than in productive capacities that would create a national digital industrial base.

The France 2030 plan dedicates 1.5 billion euros to developing AI across all sectors by 2030 [4]. The four American giants alone announced together more than 320 billion dollars for 2025 alone [4]. The gap renders any direct comparison illusory on a free-market trajectory.

The French 2026 budget doubled the tax on digital services. Its rate rises from 3% to 6% of revenue realized on national soil [8]. Its expected yield was around 700 million euros in 2024, then 800 million for 2025 [8]. This amount is considered insufficient relative to profits realized in France by digital giants. A sectoral tax levied without market counterpart finances the concentration in part by increasing operating costs passed on to users, without correcting it.

Building the Market, Not Just Its Rules

Europe lacks scale more than capital. This is the starting point of the January 2026 contribution by Tirole, Bergeaud, and Loesekrug-Pietri [9]. Every debate on competitiveness ends with a recommendation to inject additional billions. The European ecosystem has managed to produce strong start-ups but struggles to transform them into global champions capable of rivaling American and Asian giants [9].

The framework Tirole proposes proceeds in two steps [3]. First, identify the market failure mechanism. Then design the rule that corrects it without creating a new barrier.

On this terrain, the DMA is genuine progress. Two years of application teach that setting clear rules upfront for dominant platforms is worth more than waiting years for a classical antitrust procedure [8]. The question now is one of calibration.

On generative AI, the decade has shifted this technology from the status of promise to that of infrastructure. Tirole estimated in December 2025 that the costs of this technology outweigh the benefits in the absence of stronger regulation [3]. This is a warning against the assumption that technological diffusion would be automatically beneficial without appropriate framework conditions.

Technological progress creates shared prosperity only on the condition that citizens and institutions shape its directions [4]. Carl Benedikt Frey documents this in How Progress Ends (2025): innovation can stop when institutions and monopolies orient efforts toward rent capture rather than inclusive growth [12]. In 2026, European companies lag 10% behind their American counterparts on most generative AI use cases in production [4]. This lag is structural.

Four Levers to Regulate Without Giving Up Building

Complete the single digital market before adding regulatory layers. As long as information technology services represent only 15% of intra-European cross-border provision in 2023, the regulatory arsenal applies to a fragmented market incapable of producing credible competitors [10]. The European Parliament inscribed this diagnosis in its January 2026 resolution on technological sovereignty: concentrated power in the hands of non-European enterprises limits Europe’s capacity to innovate and maintain control of its digital economy [13]. Tirole, Bergeaud, and Loesekrug-Pietri make it an absolute priority: the single market first, industrial policy instruments afterward [9].

Recalibrate the AI Act to avoid sharpening the asymmetry between SMEs and large groups. The full compliance obligations for high-risk systems, originally scheduled for August 2, 2026, were postponed to December 2, 2027 by the European simplification package [8]. This postponement implicitly acknowledges the problem.

Compliance obligations must be modulated according to operator size. Public oversight must concentrate on actors dominating the market with the most powerful models. The European AI Office must be given investigative resources comparable to the Competition Directorate, to handle cases in real time.

Reorient research tax credit toward digital diffusion in SMEs. France spends 7 billion euros per year in tax incentives for R&D for insufficient results, because the tool is untargeted [4]. Refocusing part of this volume toward adoption of digital solutions in mid-sized enterprises would enable productivity gains where potential is strongest. Antonin Bergeaud has championed this recommendation since his award as best young economist of 2025 [4]. It is a reorientation at constant envelope.

Build a public data policy as productive common good. AI will transform all production processes [14]. The creative destruction that Schumpeter theorized is not guaranteed if data remains a concentrated private good, as Axelle Arquié documents in Le double choc de l’IA (2026) [14].

Data produced by public services, hospitals, transportation, and national education represent an underexploited strategic asset. A structured data-opening policy, framed by the General Data Protection Regulation and overseen by the National Data Protection Commission (CNIL), would enable European AI models without reproducing the logic of private extraction [15]. The CNIL and Competition Authority worked jointly in March 2025 on links between data protection and competition in generative AI development [15]. This joint work is the right level of institutional articulation, provided it is given power to act.

The reckoning posed for 2027 is this: continue regulating a market Europe did not build, or decide to build the market its rules presuppose.

Sources

[1] Asteres, Cigref, Numeum, study on digital purchasing by European companies, cited in Studeria.fr, August 2026, https://www.studeria.fr/articles-de-blog/couts-ia-generative-budgets-dsi-souverainete-numerique (accessed 10/10/2026).

[2] UNCTAD, “Highly concentrated digital markets put consumers at risk: here’s how to change course,” July 2025, https://unctad.org/news/highly-concentrated-digital-markets-put-consumers-risk-heres-how-change-course (accessed 10/10/2026).

[3] Jean Tirole, “We need smarter regulation in the digital age,” HEC Paris, February 2025, https://www.hec.edu/en/school/news/jean-tirole-s-urgent-call-smarter-regulation-digital-age (accessed 10/10/2026).

[4] National Productivity Council, A Changing World: Productivity, Competitiveness at the Pace of Generative AI, fifth report, April 2025, https://www.strategie-plan.gouv.fr/files/files/Publications/2025/2025-04-14%20-%20CNP%20-%20Rapport/CNP-2025-Cinqui%C3%A8me-Rapport-15avril.pdf (accessed 10/10/2026).

[5] OFCE, “French productivity slippage and underperformance of ICT-intensive sectors,” Policy Brief, November 2024, https://www.ofce.sciences-po.fr/blog2024/fr/2024/20241118_SBPG/ (accessed 10/10/2026).

[6] European Commission, France’s National Report on the Digital Decade 2026, June 2026, https://digital-strategy.ec.europa.eu/fr/factpages/frances-2026-digital-decade-country-report (accessed 10/10/2026).

[7] Usine Digitale, “Record fine of 890 million euros against Google under the DMA,” July 28, 2026, https://www.usine-digitale.fr/big-tech/google/nouvelle-phase-de-confrontation-entre-lue-et-les-geants-americains-de-la-tech (accessed 10/10/2026).

[8] European Commission, Annual Report on the Implementation of the Digital Markets Act (COM(2026) 247 final), May 21, 2026, https://www.concurrences.com/en/bulletin/news-issues/may-2026/the-eu-commission-publishes-its-annual-report-on-the-implementation-and (accessed 10/10/2026).

[9] Antonin Bergeaud, André Loesekrug-Pietri, and Jean Tirole, “Europe Does Not Lack Capital: It Lacks Scale,” Project Syndicate, January 6, 2026, https://www.project-syndicate.org/commentary/eu-technological-innovation-suffers-from-lack-of-market-scale-and-incentives-by-antonin-bergeaud-et-al-2026-01 (accessed 10/10/2026).

[10] European Court of Auditors, special report on European digital sovereignty, cited in IT Social, March 2026, https://itsocial.fr/cloud-infrastructure-it/cloud-infrastructure-it-articles/souverainete-numerique-europeenne-les-initiatives-se-multiplient-mais-le-marche-reste-fragmente-et-le-retard-se-creuse/ (accessed 10/10/2026).

[11] INSEE, National Accounts of Public Administrations, 2025-2026, https://www.insee.fr (accessed 10/10/2026).

[12] 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 10/10/2026).

[13] European Parliament, resolution on European technological sovereignty and digital infrastructure (2025/2007(INI)), January 22, 2026, https://oeil.europarl.europa.eu/oeil/fr/procedure-document-summary/pdf?id=1884418 (accessed 10/10/2026).

[14] Axelle Arquié, Le double choc de l’IA: emploi et fiscalité, 2026, https://shs.cairn.info/publications-de-axelle-arquie–111394 (accessed 10/10/2026).

[15] CNIL, Annual Report 2025, July 2026, https://www.cnil.fr/sites/default/files/2026-07/rapport_annuel_2025.pdf (accessed 10/10/2026).