Skip to content
Marsanne Studio
← The AI watch

Weekly digest ·

AI adoption rising among SMEs, but concerns over costs and control of agents

AI adoption by SMEs is progressing in France and Spain, but remains uneven across company sizes and is often limited to isolated uses. At the same time, several large companies are tightening their AI budgets and struggling to demonstrate a return on investment, while IT departments are losing control of the AI agents deployed by business teams.

Key developments

  1. AI adoption by SMEs is progressing, but remains uneven and often superficial

    In France, a Banque de France survey conducted in early 2026 among around 7,000 companies shows that two-thirds of companies with at least 20 employees use generative AI, but only 31% of companies with 20 to 49 employees use predictive AI, compared with 60% of large companies. The France Num barometer indicates that 40% of micro-businesses and SMEs use at least one AI solution, up from 26% in 2025, but only 11% use AI to automate tasks. In Spain, AI use rose from 21.1% to 33.6% in one year according to the INE, with a marked gap between large companies (74.4%) and micro-enterprises (21.4%). The OECD confirms this two-speed adoption across its area: 20% of companies compared with 52% of large companies.

    Sources: Maddyness (1, 2), Brief IA, El Español, Funcas Blog, Silicon.fr

  2. The return on investment of AI remains hard to demonstrate for most companies

    According to a PwC study cited by the head of Anaplan, 81% of French executives believe AI has no impact on their revenue yet, even though more than two-thirds of French companies say they use it. In the United States, Torsten Slok, chief economist at Apollo Global Management, states that AI has generated no financial benefit for companies outside the technology industry; an international study shows that 90% of 6,000 executives surveyed have seen no productivity gains over three years. Gartner notes that faulty strategic calculation on AI is now among the top three emerging risks identified by finance departments, with only 17% of chief financial officers seeing measurable gains on more than half of their AI projects.

    Sources: Solutions Numériques, Developpez.com, IT for Business

  3. Several large companies are tightening their AI spending

    Apple has reportedly cut the daily token budget allocated to some engineers from $300 to $200, according to Bloomberg. British fintech Revolut says it has cut its inference costs tenfold by replacing the Anthropic and OpenAI APIs with the Chinese open-weight model GLM, for around 2,000 engineers. An Ocient study of 500 mostly American companies shows that only 38% see a return on investment from their AI projects and that three-quarters are seeing their costs rise; a quarter of decision-makers have already stopped unprofitable projects. Major software vendors, including Oracle, Microsoft and Salesforce, are meanwhile shifting to consumption-based pricing for AI agents.

    Sources: Xataka, IT for Business (1, 2), LeMagIT

  4. IT departments are losing control of AI agents deployed internally

    A Dataiku study conducted by The Harris Poll among 685 CIOs in eight countries shows that 81% of them do not control AI agents created outside approved channels, and that 84% see employees creating them faster than IT can oversee them. Two-thirds of companies have at least 51 agents active in production, but only 28% of CIOs systematically measure their value and 21% track their costs in real time. In France, only 27% of CIOs are certain they could trace an AI agent's decision for a regulator. A TCS study confirms that industrialisation remains marginal: only 8% of organisations consider their AI platforms fully deployed.

    Sources: Alliancy, LeMagIT

  5. The effects of AI on employment remain limited, despite targeted restructuring

    According to a Banco de España study, 89% of Spanish companies have not seen a significant effect of AI on their workforce over the past three years. In the United States, a KPMG survey of 400 CEOs of large companies shows that nearly nine in ten plan to increase headcount over the next three years, while 53% of them anticipate targeted job cuts linked to AI. More marked restructuring is affecting some technology companies: Canva announced a reduction or restructuring of 22% of its workforce, and Oracle France a plan affecting 194 positions, while 70% of the code produced by Oracle is now generated by AI.

    Sources: Infobae, Journal du Net, elEconomista, Siècle Digital, Silicon.fr

Figures of the week

  • 81 %

    Share of CIOs who do not control AI agents created outside approved channels

    685 CIOs surveyed in 8 countries

    « Selon une étude Dataiku publiée le 29 octobre, 81 % des directeurs des systèmes d'information interrogés n'exercent aucun contrôle sur les agents créés hors des environnements approuvés ou sans déclaration préalable. »

    Alliancy , from Étude Dataiku / The Harris Poll

  • 40 %

    Share of French micro-businesses and SMEs reporting use of at least one AI solution in 2026

    French micro-businesses and SMEs, 2026

    « Selon le baromètre France Num 2026, publié en septembre, 40 % des TPE-PME françaises déclarent utiliser au moins une solution d’IA, contre 26 % en 2025. »

    Maddyness , from Baromètre France Num 2026

  • 33.6 %

    AI adoption rate among Spanish companies in 2025

    Spain, companies with 10 or more employees, 2025

    « En solo doce meses, su uso ha pasado del 21,1% al 33,6%, el mayor salto registrado desde 2022. »

    El Español , from Encuesta sobre el uso de TIC y del comercio electrónico, INE

  • 90 %

    Share of US CEOs planning to increase headcount over the next three years

    400 CEOs of large US companies with annual revenue above $500 million, KPMG survey published on 15 October 2025

    « Près de 9 sur 10 prévoient d'augmenter leurs effectifs au cours des trois prochaines années. »

    Journal du Net , from KPMG

  • 20 %

    Share of OECD companies (10 or more employees) using AI in 2025

    Companies with 10 or more employees, OECD area, 2025

    « En 2025, 20 % de l’ensemble des entreprises de la zone OCDE (de dix salariés ou plus) utilisent l’IA, contre 52 % des grandes entreprises. »

    Silicon.fr , from OCDE, Digital Economy and Society Outlook 2026

In France and Spain

France

Beyond adoption, several signals illustrate the growing role of AI in French organisations. At Michelin, AI agents contribute to the more than €100M in annual AI-related gains, and are now used by around a hundred IT staff out of 2,000, compared with about fifty pioneers previously. The Numeum-EY Top 250 of French software vendors shows growth slowing to 5.2% in 2025, in a context where agentic AI is becoming the top technology priority for nearly one vendor in two. The La Boussole des Héros barometer shows that 38% of French mid-sized companies cite AI among their investment priorities, but only 13% have deployed it across several functions. Groupe Positive has also acquired Humanlinker, a French specialist in AI-driven prospecting.

Sources: IT for Business, Alliancy, Le Journal des Entreprises (1, 2)

Spain

Sage has launched a €3.5 million fund in Spain to finance AI projects in SMEs and professional firms, while only 9% of Spanish SMEs use AI according to the INE, against a national average adoption rate of 4.1% according to the IndesIA barometer in Andalusia. The Cámara de España offers grants that can cover up to 85% of the first €7,000 invested in digital transformation. Even among large companies, 60% do not yet know where to direct their AI strategy, according to Deloitte España. Nearly six in ten Spanish companies consider themselves poorly prepared for attacks specifically targeting AI systems.

Sources: MuyComputerPRO, El Confidencial, El Español, El Debate, Expansión

Internationally

United States

Several large American companies are tightening their AI spending: Apple has reportedly cut the daily token budget allocated to some engineers from $300 to $200, and is considering a further reduction, according to Bloomberg. At Microsoft, one employee spent $28,000 on tokens in a month, whereas Nvidia, by contrast, encourages consumption. Uber says that only four people booked a ride through an AI agent in the busiest week of the last quarter, illustrating that agentic AI use among customers is still marginal. At Shopify, the internal agent River now writes up to half of merge requests, which led its chief executive to warn about the lack of verification. For a French or Spanish executive, these examples show that even the most advanced companies are tightening costs and questioning the reliability of agents before rolling out their use more widely.

Sources: Xataka, L'Usine Digitale, TechCrunch, Unite.AI

China

This week, Chinese companies appear mainly as technology suppliers to Western companies. Z.ai's open-weight GLM model has been adopted by British fintech Revolut, which hosts it in-house to reduce its inference costs. An article on Forrester's 2027 predictions also mentions the diversification of AI accelerators beyond Nvidia, citing Alibaba Cloud and Huawei among the Chinese players being tracked, as well as an incident at AWS in China as an illustration of the risk of outages linked to AI-generated code.

Sources: IT for Business (1, 2)

Rest of the world

In the Middle East, Abu Dhabi and OpenAI have signed four agreements covering energy, administration and investment, while Adnoc announces it is becoming the first energy company to deploy OpenAI Frontier. British law firm A&O Shearman is investing $1.1 billion in AI over five years. In Argentina, a study by Seeds and IDEA shows that 88% of professionals already use AI at work, but only 34% of companies have a formal usage policy. In Mexico, Google reports that 82% of SMEs already see AI-related sales.

Sources: ActuIA, BFM Business, Financial Times, El Cronista, El Ecosistema Startup

What's changing

Over the past five weeks, the share of articles on AI governance has risen from 6–8% to 11% this week, and return on investment is becoming a more prominent keyword (7%, compared with a previous maximum of 4%). The share of Data-driven studies has fallen (21%, compared with 27% the previous week), while Analyses have increased (24%, the highest level of the five weeks). The share devoted to large companies has decreased slightly (28%, compared with 31–36% in the previous four weeks), while that of SMEs remains stable at around 19%. The banking-insurance-finance industry accounts for a larger share (19%) than two weeks ago (13%), while manufacturing has fallen (6%, compared with 8–12% before).

Comparison with the previous 4 weeks

Method

Digest written on 9 November 2026 by an AI model (Claude Sonnet 5, by Anthropic) from the 160 articles published in the AI watch for 2 – 8 November 2026. Each paragraph links to the entries it draws on, and every number in the text was found in one of those articles before publication. The figures of the week are given with the exact sentence from the article that provides them. Digest translated from French by Claude Sonnet 5.5.

Previous weeks