OECD Digital Economy and Society Outlook 2026

Chapter 3. Digital technology diffusion and the economy

Abstract

Digital technologies and data are reshaping the economy and society, creating opportunities for productivity gains but also new challenges. This chapter maps current diffusion patterns for artificial intelligence (AI), big data analytics and cloud computing across firms, sectors and society. It explores plausible near-term diffusion paths and examines how information and communication technology (ICT) outputs propagate through supply chains and affect economic activity across sectors. Using new data on the access and usage of ICTs, the chapter’s forward projections and input-output analysis highlight where adoption is concentrated; where co-adoption of technologies is emerging; and how generative AI is changing business practices, as well as patterns of individual use. The chapter concludes with practical policy levers to expand access, strengthen capabilities and govern responsible uptake so that digitalisation can support resilient productivity growth.

 

Key findings

Firm-size disparities persist in AI adoption, while those in cloud computing are expected to converge

• In 2025, cloud computing was the most widely adopted digital technology, used by more than half of all firms and over 80% of large firms across the OECD.

• Adoption of artificial intelligence (AI) remained more limited in 2025, at 20% for all firms and 52% for large firms, while big data analytics occupied an intermediate position.

• Projections to 2030 suggest cloud uptake will start to converge across firms of all sizes, whereas AI adoption will likely remain skewed towards large firms without increased support for adoption.

AI uptake is maturing, with differences across firms and individuals

• Firms primarily use AI for marketing, business administration, finance, and information and communication technology (ICT) security, with large firms leading across business applications.

• The integration of AI complementary technologies remains limited, with combined use of AI, cloud computing and big data analytics concentrated in large firms and knowledge-intensive sectors.

• Among individuals, generative AI use reached 37% in 2025, with higher uptake among younger, highly educated and higher-income groups.

ICT outputs underpin growth, resilience and global digital integration

• ICT sector growth is estimated at 5% in 2025, reflecting a broad-based and robust expansion across the OECD.

• Wholesale and retail trade, professional and technical activities, and finance show the highest total ICT forward linkages, with ICT intensity often exceeding the OECD average by 20-50%, while construction and some service industries rely more on indirect spillovers.

• ICT value creation and sourcing are highly concentrated internationally, with a small number of countries acting as key production and supply hubs, reinforcing both efficiency gains and cross-border interdependencies.

Introduction

Digital technologies and data-driven tools are reshaping how firms compete and how individuals work, learn and consume. Their adoption reflects differences in resources, skills and access to digital infrastructure, which in turn influences where productivity gains and innovation are most likely to occur. Examining these patterns across businesses, sectors and individuals helps identify where policy action can most effectively support digital transformation for innovation, growth and well-being.

This chapter first examines the adoption of cloud computing, artificial intelligence (AI) and big data analytics across firms, highlighting size, sectoral and country-level disparities. It then explores the surge of generative AI (GenAI) among firms and individuals, considering technological choices, applications and socio-economic divides. Finally, the chapter analyses growth dynamics of the information and communication technology (ICT) sector; domestic and cross-border spillovers; and countries’ integration into global digital value chains. It concludes with implications for policy measures that can support uptake for all, workforce readiness and the more efficient use of frontier digital technologies.

Diffusion of data-driven technologies: Current and future trends

The diffusion of frontier digital technologies is increasing, but firm size drives uptake

The diffusion of frontier digital technologies continues across businesses, with differences between large and smaller firms. This section focuses on three key technologies driving business transformation: cloud computing, AI and big data analytics. Cloud computing provides scalable infrastructure and data storage, enabling firms to process large datasets without substantial upfront investment. AI and big data analytics leverage these data to automate processes, generate insights and support decision making.

In 2025, cloud computing was the most widely adopted of the three technologies. More than half of all firms reported using it, with adoption among large firms exceeding 80% across the OECD. AI adoption remained more limited, averaging 20% among all firms and 52% among large firms, while big data analytics occupied an intermediate position. Firm size strongly shapes adoption: larger firms consistently led across all technologies, reflecting greater resources, technical capacity and access to skilled personnel. Smaller firms, while lagging, have embraced cloud computing more readily, highlighting its accessibility and scalability (Figure 3.1).

Country-level patterns reveal substantial variation. Front-runners like the Nordic countries combine high overall adoption, particularly for AI. In contrast, countries such as Türkiye, exhibit more limited uptake across all technologies, with large firms maintaining a clear lead. These differences underscore persistent barriers for smaller firms, including constrained access to finance, digital infrastructure and ICT skills.

Figure 3.1. Cloud adoption remains over twice as high as AI adoption among firms

Adoption rates of cloud computing, big data analytics and AI by firm size, 2025 (or most recent)

Note: For cloud computing and artificial intelligence, the most recent observation refers to 2025 except for Korea (2024), Canada, Japan and Switzerland (2023), and Australia and New Zealand (2022). For big data analytics, the most recent observation refers to 2025 except for Korea (2024), Japan (2023), and Australia and Switzerland (2022).

For countries in the European Statistical System, data for “big data analytics” relate to “data analytics”.

For Australia, data are for the financial year 2021-2022 ending 30 June 2022.

For Brazil, Retail trade (ISIC Rev.4 Section G, Division 47) and Administrative and support service activities (ISIC Rev.4 Section N) are not collected.

For Canada, the North American Industry Classification System is used instead of ISIC Rev.4.

For Japan, data originate from the Japanese National Innovation Survey 2024 and refer to the three-year period 2021-2023.

For Korea, Wholesale trade, except of motor vehicles and motorcycles (ISIC Rev.4 Section G, Division 46) includes Retail trade, except of motor vehicles and motorcycles (ISIC Rev.4 Section G, Division 47).

For Switzerland, data refer to businesses with five or more employees (instead of ten or more) and for “cloud computing”, to proportion of total businesses using (rather than buying) cloud computing services.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

To anticipate the evolution of technology adoption, forward-looking diffusion paths have been constructed using historical data and projections up to 2030 for cloud computing and AI. These trajectories illustrate both the pace of uptake and the development of firm-size disparities. Cloud computing adoption is projected to remain widespread, with smaller firms gradually closing the gap with large firms: while adoption among large firms was already above 80% in 2025, smaller firms are expected to reach near comparable levels by the end of the decade. In contrast, AI adoption is set to accelerate more rapidly among large firms than smaller ones. Historical data show that large firms have consistently led in AI uptake since 2020, and projections indicate that by 2030, adoption among large firms could be nearly 80% higher than the average across all firms (Figure 3.2).

Figure 3.2. Firm-size gaps in adoption are projected to increase for AI but narrow for cloud computing

Projected diffusion paths for cloud computing and AI in the OECD, by firm size, 2010-2030

Notes: Forward-looking diffusion paths are not presented for big data analytics due to changes in survey questions over time. See Annex 3.A for methodology and results.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

Adoption also varies across economic sectors (Figure 3.3). Cloud computing has achieved broad penetration, particularly in information and communication (Section J), financial and insurance activities (Section K), professional, scientific and technical activities (Section M) and real estate activities (Section L). AI remains concentrated in knowledge-intensive and technology-driven sectors, whereas construction (Section F), transportation and storage (Section H) and accommodation and food service activities (Section I) exhibit comparatively low uptake. Big data analytics occupies an intermediate position, with moderate-to-high adoption in most sectors; it is strongest in industries reliant on data and information processing. These patterns highlight how sectoral characteristics, alongside firm size and digital infrastructure, shape both the pace and scope of technology diffusion.

Figure 3.3. Adoption of frontier digital technologies is strongest in knowledge- and data-intensive sectors

Adoption of digital technologies, by economic sector, enterprises with ten employees or more, 2025 (or most recent)

Notes: ISIC Rev. 4 sector codes are as follows: C: Manufacturing; F: Construction; G46: Wholesale trade (except repair of motor vehicles and motorcycles); G47: Retail trade (except repair of motor vehicles and motorcycles); H: Transportation and storage; I: Accommodation and food service activities; J: Information and communication; K: Financial and insurance activities; L: Real estate activities; M: Professional, scientific and technical activities; N: Administrative and support service activities. Data for financial and insurance activities should be interpreted with caution due to the limited country coverage. Averages are based on data from OECD Member countries only.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

Firms are progressively combining frontier digital technologies

Cloud computing, AI and big data analytics are increasingly complementary, with their combined use enabling more sophisticated data processing, advanced analytics and automated decision making. Cloud platforms provide scalable computing and storage capacity, while big data analytics and AI transform raw data into actionable insights and predictions (Sunyaev, 2020_[2]). Previous research suggests that more than half of firms adopting frontier digital technologies deploy multiple technologies simultaneously. This highlights complementarities between AI, cloud computing, big data analytics and other data-intensive tools (Cho et al., 2023_[3]). Examining how firms deploy these technologies in combination, rather than individually, provides valuable insight into the integration of digital technologies and the potential for increased productivity and innovation, particularly among larger firms with the resources to implement bundled solutions (Calvino, Criscuolo and Ughi, 2024_[4]).

Across the OECD, cloud computing acts as a key enabler of digital transformation in firms. Around 18% of firms rely on cloud services without adopting either big data analytics or AI, while a further 13% combine cloud computing with big data analytics but not AI. Adoption of big data analytics alone, without cloud or AI, is less common (around 10% of firms). At the same time, approximately 12% of firms deploy all three technologies simultaneously, indicating that while multi-technology integration is far from universal, it is no longer marginal. By contrast, firms are not yet leveraging the advantages of combining cloud-based data analytics with AI uptake. That number remains limited (2%), underscoring the importance of technological complementarities (Figure 3.4).

Country-level patterns reveal substantial heterogeneity in the degree of technology integration. The highest shares of firms adopting all three technologies in combination, reaching more than 20%, are observed in Belgium, Denmark, Finland, Korea, the Netherlands and Sweden, pointing to comparatively advanced digital integration strategies. By contrast, in several Central, Eastern and Southern European countries – such as Italy, Poland, Portugal and Spain – firms are more likely to rely on a single technology, particularly cloud computing or big data analytics, rather than integrated combinations. Overall, the data suggest a clear divide between countries where digital tools are increasingly deployed as complementary technologies and those where adoption remains more siloed. These differences point to persistent technological and organisational barriers: combining frontier digital technologies effectively requires not only access to digital infrastructure, but also data readiness, specialised skills and complementary investments in organisational change.

Figure 3.4. Most firms adopt cloud computing alone or with data analytics, while full integration with AI remains limited

Mutually exclusive adoption rates of different combinations of frontier digital technologies, enterprises with ten employees or more, 2025 (or most recent)

Note: The most recent observation refers to 2025 except for Korea (2024), Canada and Japan (2023), and Australia (2022).

For countries in the European Statistical System, data for “big data analytics” relate to “data analytics”. Moreover, data for “cloud computing services” relate to the purchase of sophisticated or intermediate cloud computing services. Sophisticated or intermediate cloud computing services are defined as the purchase of one or more of the following as a cloud computing service: finance or accounting software applications; ERP software applications; Customer Relationship Management (CRM) software; security software applications; hosting for the enterprise’s database; computing platform providing a hosted environment for application development, testing or deployment.

For Australia, data are for the financial year 2021-2022 ending 30 June 2022.

For Brazil, Retail trade (ISIC Rev.4 Section G, Division 47) and Administrative and support service activities (ISIC Rev.4 Section N) are not collected.

For Canada, the North American Industry Classification System is used instead of ISIC Rev.4. Moreover, the Canadian survey only asks about one cloud computing services use case that is considered sophisticated: “Cloud-based environment for application development, testing or deployment.” Furthermore, 2023 data for “big data analytics” relate to “data analytics”.

For Japan, data originate from the Japanese National Innovation Survey 2024 and refer to the three-year period 2021-2023.

For Korea, Wholesale trade, except of motor vehicles and motorcycles (ISIC Rev.4 Section G, Division 46) includes Retail trade, except of motor vehicles and motorcycles (ISIC Rev.4, Section G, Division 47).

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

The distribution across sectors echoes the patterns observed at the country level (Figure 3.5). The information and communication sector (Section J) stands out as the most intensive adopter of frontier digital technologies: more than 86% of firms use at least one of the three technologies, and 39% deploy cloud computing, big data analytics and AI simultaneously. Professional, scientific and technical activities (Section M) follow, with 22% of firms adopting all three technologies. In contrast, construction (Section F), accommodation and food service activities (Section I), and transportation and storage (Section H) display more fragmented adoption patterns, typically centred on cloud computing alone or in partial combinations. These sectoral differences highlight where co-ordinated adoption is already well advanced and where firms may benefit most from targeted support to overcome organisational and skills-related barriers to integration.

Figure 3.5. Information and communication lead in co-ordinated adoption of frontier digital technologies across sectors

Mutually exclusive adoption rates of different combinations of frontier digital technologies, enterprises with ten employees or more, 2025 (or most recent)

Notes: Averages are based on data from OECD Member countries only.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

Co-adoption patterns reinforce earlier observations on the influence of firm size. Large firms are most likely to adopt multiple technologies, with 44% using all three simultaneously. Medium-sized firms appear well positioned for further AI uptake, as over one in five already uses the full set and a further 20% already combine advanced cloud computing services with big data analytics. Small firms also show potential, with 19% already purchasing cloud computing services (see the Statistical Annex), although further upskilling in data analytics and AI would enable them to fully leverage additional advanced technologies.

To complement the analysis of adoption patterns and co-occurrences of frontier technologies, it is useful to examine digital intensity at the firm level. Digital intensity captures not only whether firms adopt individual technologies, but also the breadth and sophistication of their digital capabilities. The following section presents a microdata-based approach that maps firms according to their digital intensity, providing a nuanced view of how advanced technologies are distributed across firms and countries.

A distributed approach to measuring digital intensity

A distributed microdata analysis examines a way of measuring digital intensity by applying a multiple correspondence analysis (MCA) and clustering of ICT Access and Usage survey data. This analysis clusters firms in an MCA factor space according to their use of digital technologies. To that end, it leverages survey data that capture both widespread practices such as having a website, selling online and using simple cloud services like e-mail, and the uptake of intermediate and advanced technologies – including big data analytics, AI and purchasing cloud solutions for computing power and analysis. Digital intensity is defined as adoption relative to the technologies prevalent in each country, resulting in a country-specific digital intensity measure.

Identifying digital leaders

By partitioning firms into clusters of digital intensity, countries with stronger relative ICT presence across the business population can be identified. In this sample, Mexico has the largest share of firms in the most digitally intensive cluster (14%) followed by Sweden (8%) and the United Kingdom (6%), while fewer firms are in the corresponding cluster in Italy (4%) and Canada (3%). Compared with Sweden, Mexico exhibits a more contracted MCA, indicating smaller differences in digitalisation between clusters. Conversely, Canada and Italy display smaller high-intensity clusters combined with broader bases in lower-intensity clusters.

The structure of the business sector, including the presence of a large ICT sector, contributes to a higher digital intensity, as does having many large firms, which are more likely to adopt advanced digital technologies. On average, the top performing cluster across participating countries uses all considered technologies to a greater extent than the others, and in combination. Plotting the cluster shares by technology reveals a widening gap as technologies and their applications become more advanced. In Figure 3.6, AI is notably on the far-right side, indicating on average low uptake across all firms in 2023. However, almost half of firms in the highly digitalised cluster report using AI.

Figure 3.6. As technologies get more advanced, the gap in technology adoption rates between digital intensity clusters increases

Average technology adoption by digital intensity cluster, 2023 (or most recent)

Notes: CC = Cloud computing, ERP = Enterprise Resource Planning, DB = Database, CRM = Customer Relationship Management, BI = Business intelligence, AI = Artificial intelligence. Data refer to Canada, Italy, Mexico, Sweden and the United Kingdom. Data from the United Kingdom refer to 2020.

Source: OECD calculations based on result from a distributed microdata analysis performed by data producers.

Statlink

AI adoption patterns are starting to diverge

Since 2023, GenAI has become widespread due to the rapid diffusion of GenAI tools; more recent data show clear increase in adoption patterns. While this uptake is increasing quickly among OECD Member countries, firms diverge in the complexity of their use. Evidence shows that more advanced applications – such as analysis based on machine learning – place firms further from the centre in the MCA factor space, indicating a higher level of digitalisation. In contrast, simpler use cases like text generation or text analysis, typically associated with GenAI, sit closer to the centre and reflect lower capability requirements. This pattern suggests that AI uptake is beginning to resemble cloud computing uptake: increasingly mainstream but differentiated by levels of sophistication that matter for measuring digital intensity. As adoption broadens, the key distinction becomes how well firms use AI rather than whether they use it.

Extending the analysis to 2025 for Italy clearly illustrates the shift towards rapid diffusion of GenAI tools (Figure 3.7). Comparing the placement of technologies between 2023 and 2025 shows how the space contracts as adoption increases. Both AI and advanced cloud services move towards the centre. This indicates less pronounced differences between digital intensity clusters because a larger share of firms uses these tools in some form. This is echoed by the share of Italian firms belonging to the highest-intensity cluster (6% – up from 4% in 2023). Together, these results underscore the importance of distinguishing between basic and advanced AI applications when assessing digitalisation trends.

Figure 3.7. Rapid diffusion of advanced technologies in Italy contracts the MCA factor space

Digital technologies placed in MCA factor space, Italy, 2023 and 2025

Note: CC = Cloud computing, ERP = Enterprise Resource Planning, DB = Database, CRM = Customer Relationship Management, BI = Business intelligence, AI = Artificial intelligence.

Source: OECD calculations based on analysis by Istituto Nazionale di Statistica (Istat).

Statlink

Trends and drivers of AI use by businesses and individuals

The rapid diffusion of GenAI across both firms and households present new opportunities and challenges for economic productivity, equity and policy design. This AI deep dive provides a more granular understanding of adoption patterns, technological choices and purposes of AI use. The section is structured to first examine firms, highlighting what drives AI uptake, which technologies are being deployed and where adoption is accelerating across sectors. This firm perspective provides context for the subsequent analysis of AI use among individuals, where GenAI is emerging as a particularly transformative force.

Business uptake: Drivers, technologies and sectoral patterns

AI adoption among firms continues to expand rapidly, but the pace of diffusion varies markedly across countries. To provide a consistent 2025 snapshot, including for countries where the most recent data are not yet available, this analysis combines official statistics with nowcast estimates. The projections are based on a diffusion framework reflecting a well-established empirical regularity: new technologies typically spread in an S-shaped pattern – gradual initial uptake, followed by acceleration as use cases mature before growth stabilises (Meade and Islam, 2006_[6]). The model draws on observed cross-country trajectories to produce economically grounded estimates where recent observations are missing, while allowing countries to differ in their overall level of adoption. Additional methodological details and accompanying predictive ranges are reported in Annex 3.B.1

Figure 3.8 presents the resulting share of enterprises with ten or more employees using AI in 2025 across OECD Member countries. Solid blue bars correspond to official data from the OECD ICT Access and Usage Database (OECD, 2026_[1]), while hatched bars denote nowcast estimates. In several Nordic countries, more than one in three firms report using AI, whereas in others uptake remains below 10%. These contrasts underscore both the rapid diffusion of AI and the persistence of sizeable cross-country adoption gaps.

Figure 3.8. AI use in enterprises differs substantially across the OECD

Enterprises using AI as a percentage of all enterprises with 10 or more employees, official statistics (full blue bars) and nowcasted point estimates (hatched blue bars), 2025

Notes: Full blue bars point to official statistics while hatched bars refer to nowcasted point estimates. The OECD average includes both official statistics and nowcasted point estimates. See Annex 3.B for additional details on the methodology, along with predictive ranges accompanying the predictions presented above.

Source: OECD (2026_[20]), Data Kitchen (database), https://goingdigital.oecd.org/datakitchen based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

What drives AI adoption in firms

AI adoption among firms can reflect and be driven by a combination of operational efficiency, strategic decision making and customer engagement objectives. Across OECD Member countries, firms most commonly deploy AI for four purposes: marketing and sales; business administration and management; accounting, controlling or finance management; and ICT security. Marketing and sales applications are the most widespread, particularly among large firms. They include activities such as customer profiling, price optimisation, personalised marketing offers, market analysis based on machine learning and chatbots for customer support. Business administration and management applications support strategic decision making, document drafting, forecasting, human resources management and planning. For their part, accounting, controlling and finance management uses are generally less prevalent but show steady growth in certain countries and among larger firms (OECD, 2024_[7]). Lastly, ICT security applications focus on authentication and cyber-attack detection.

Adoption rates vary markedly by country and firm size (Figure 3.9). Large firms consistently report higher uptake across all purposes, reflecting their greater resources and capacity to integrate AI across business functions. In marketing and sales, for example, adoption among large firms exceeds one-quarter in several countries, including Belgium, Finland, the Netherlands, Slovenia and Sweden, while remaining below 10% in many Southern and Eastern European countries. Integration of AI and ICT security is strong in Belgium, Finland, Korea and Slovenia, where 40% or more of large firms adopt the technology. Business administration and finance applications are gradually expanding, especially among large firms in Nordic countries, suggesting a gradual diffusion of AI use beyond customer-facing functions.

Figure 3.9. AI adoption purposes reflect efficiency, strategic and customer-focussed priorities among firms

Adoption rates of top four AI purposes by firm size, 2025 (or most recent)

Notes: The most recent observation refers to 2025 except for Korea (2024), Canada and Switzerland (2023).

The indicator on AI use for marketing or sales measures the share of enterprises employing AI for activities such as customer profiling, price optimisation, personalised marketing offers, market analysis based on machine learning, chatbots using natural language processing for customer support, or autonomous robots for order processing. AI use for business administration or management captures enterprises deploying AI for business virtual assistants to support document drafting, data analysis or strategic decision making. This includes risk assessment, planning or business forecasting, and human resources management tasks such as candidate pre-selection, employee profiling or performance analysis. The use of AI for accounting, controlling or finance management measures the adoption of AI for automated invoice processing, bookkeeping tasks based on machine learning or natural language processing, and the analysis of financial data to inform decision making. The ICT security indicator refers to the application of AI technologies for authenticating users through facial recognition or detecting and preventing cyber-attacks using machine learning.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

From business needs to AI solutions: Technologies in use

The AI technologies deployed by firms closely align with the purposes they serve, highlighting how solutions are tailored to business priorities. The top technologies include AI for analysing written language; AI for generating written or spoken language; speech recognition; and AI for workflow automation or decision support. Text analysis and machine learning are widely used across countries and sectors, underpinning applications in marketing, business administration and financial analysis. Natural language generation and speech synthesis are emerging in areas such as customer support, document drafting and virtual assistants. Workflow automation and decision-support technologies, including software-based robotic process automation, support back-office efficiency and strategic decision making.

Technology adoption is highly correlated with firm size (Figure 3.10). Large firms consistently report higher usage rates across all technologies, reflecting their ability to deploy complex AI solutions and integrate them into existing processes. Among the different technologies, text analysis and natural language generation are the most widely adopted, highlighting the growing importance of language-based technologies in driving digital transformation. Their broad applicability across industries likely reflects lower entry barriers compared to more specialised AI systems, underscoring the need for policies that support responsible deployment and skills development in this area. Workflow automation technologies are also widely used, supporting both operational efficiency and higher-level decision making. These patterns suggest that firms are not only meeting current operational needs but are also investing in scalable AI solutions that can transform multiple business functions.

Figure 3.10. AI technologies deployed align with business priorities and firm sizes

Adoption rates of top four AI technologies by firm size, 2025 (or most recent)

Notes: The most recent observation refers to 2025 except for Korea (2024), Canada and Switzerland (2023).

AI technologies used by enterprises: the indicator on AI technologies performing analysis of written language captures the use of methods such as text mining, sentiment analysis and other text analytics. The generation of written or spoken language measures enterprise use of AI for natural language generation, chatbots or speech synthesis. Speech recognition refers to AI technologies converting spoken language into machine-readable format. Finally, AI technologies for automation or decision support cover the use of AI-based software for robotic process automation, workflow automation, and decision-support systems designed to assist or streamline enterprise processes.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

Sectoral patterns and the surge in AI uptake

AI adoption accelerated rapidly across sectors between 2023 and 2024, with growth rates exceeding 50% in most industries (Figure 3.11). This surge points to a broad-based expansion of AI use, likely reflecting heightened awareness of AI’s potential to improve efficiency, support decision making and enhance customer-facing activities. Growth remains strong through 2025, although at a slower pace across most sectors, suggesting a transition from an initial surge towards more gradual diffusion.

Retail trade (Section G, Division 47) experienced the fastest growth in 2023-2024 (over 90%), followed by wholesale trade (Section G, Division 46), professional, scientific and technical activities (Section M) and administrative and support services (Section N), all recording growth rates above 70%. In 2025, growth moderates across all sectors but remains relatively strong in construction (Section F), accommodation and food services (Section I), and wholesale trade (Section G, Division 46). The latter all record rates above 50%, indicating continued diffusion beyond early adopters.

While AI adoption continues to rise, its growth rate is slowing in several sectors. Growth in information and communication (Section J), while strong in 2023-2024, slows more markedly in 2024-2025, consistent with earlier and more mature adoption patterns.2 Administrative and support service activities (Section N), retail trade (Section G, Division 47) and professional, scientific and technical activities (Section M) similarly shift from rapid initial uptake to more moderate growth, suggesting that early experimentation is giving way to consolidation. Transportation and storage (Section H), real estate activities (Section L) and manufacturing (Section C) also record steady but lower growth rates, pointing to a more gradual embedding of AI into existing business processes. Overall, these patterns indicate that while AI adoption continues to expand across the economy, the pace increasingly reflects sector-specific maturity, operational readiness and the scope for further integration.

Figure 3.11. AI adoption surged in 2024 but has moderated in 2025 with strong disparities across sectors

Adoption growth rates of AI by sector, enterprises with ten employees or more, 2023-2024 and 2024-2025

Note: Growth rates are based on data from OECD Member countries only.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

The sharp acceleration in 2024 partly reflects the rapid uptake of general-purpose GenAI tools. Tools such as ChatGPT, Gemini and Copilot have lowered entry barriers by offering ready-to-use solutions for a wide range of business functions (Calvino, Haerle and Liu, 2025_[8]). Evidence shows that enterprises use these tools not only to generate written content but also to generate pictures, videos and sound – 10% of European firms report such uses in 2025 (Eurostat, 2025_[9]). Yet many firms continue to face challenges that constrain broader adoption. In the European Union in 2025, 80% of firms do not use AI and only 11% of non-users have considered using these technologies in the future (Eurostat, 2025_[9]). Among enterprises considering use of AI, the most frequently cited obstacle is lack of in-house expertise. Firms also report concerns related to data protection and privacy, uncertainty about the legal and regulatory implications of AI use, perceived high implementation costs and difficulties in integrating AI systems into existing infrastructure3 (OECD, 2026_[1]). These findings suggest that policies to promote AI diffusion need to address a broad set of constraints simultaneously, combining skills development, clarity on legal and data protection frameworks, and support for integration costs, particularly for firms with limited digital capabilities.

Individuals’ use of generative AI: Purposes and key divides

The growing role of GenAI is particularly important. Not only is GenAI likely driving firm adoption, it is also reshaping how individuals engage with AI in their daily lives. This dual influence makes GenAI a natural bridge between business and personal use, which is the focus of this subsection.

Generative AI spreads widely among individuals, driven mainly by private activities

GenAI tools are now widely accessible, and their usage extends well beyond professional contexts. Across the OECD, around 37% of adults reported use of GenAI tools in the last three months (Figure 3.12), with adoption exceeding 45% in several Nordic and smaller European countries. Use is most common for private purposes, reported by around 30% of individuals on average, compared to 19% for professional activities and 11% for formal education, such as school or university. This pattern indicates that GenAI is increasingly embedded in everyday life, not only in work or learning environments.4

Figure 3.12. About 40% of individuals use generative AI in the OECD, mostly for private purposes

Share of adults (aged 16 to 74) using any generative AI tools to create content in the last three months by purpose of use, 2025 (or most recent)

Notes: For Colombia, data refer to 2024 with a reference period of 12 months. For Japan, data refer to 2024 and no reference period is specified. In addition, data relate to individuals aged 20 to 69.

The indicator on generative AI (GenAI) use captures the purposes for which individuals employed tools such as ChatGPT, Copilot, Gemini, LLaMA, Midjourney or DALL-E to create new content – including text, images, programming code or videos – over the past three months. Use for private purposes reflects engagement with GenAI in personal activities, such as hobby projects, creative writing or leisure applications. Professional purposes capture the adoption of GenAI in work-related tasks, including content creation, coding assistance or data analysis. Use for formal education refers to applications of GenAI in school or university contexts, such as completing assignments, learning or research activities.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

Younger, highly educated and higher-income individuals are more likely to use generative AI

According to the data, the use of GenAI is strongly associated with age, with younger individuals using these technologies at a higher rate. The share of individuals aged 16 to 24 who reported using GenAI is on average 68%, and the share for students aged 16+ reached 76%. The adoption rates decline significantly with age across all countries, with the largest differences observed between the 16-24 and 55-74 age groups. In these cases, the OECD average gap reaches 54 percentage points (p.p.), ranging from 26 p.p. in Colombia to over 67 p.p. in the Slovak Republic.

Moreover, individuals with a tertiary education report higher usage rates than those with only compulsory education. The average gap of 22 p.p. reflects greater awareness and access to AI for more educated individuals. Similarly, adoption of the technology is also correlated with income. On average, individuals in the highest income quintile reported usage rates at around 22 p.p. higher than those in the lowest quintile. Together with the strong age and education gradients, this suggests that exposure to GenAI is concentrated among younger, more educated and higher-income individuals, who are more likely to use these tools. Usage is also generally higher among men than women, although the gap across sexes is much smaller, at around 4 p.p., and it varies considerably across countries, from slightly favouring women in Estonia, Slovenia, Lithuania and Croatia, to over 9 p.p. higher for men in Ireland (Figure 3.13).

These findings are broadly consistent with those of the OECD Truth Quest Survey (2024_[10]), which shows that younger adults (those aged 18-24) and those with a tertiary education are also the most optimistic about the impact of AI on their lives. After students aged 16 and over, GenAI is most favoured by those in employment (41%) and unemployed (37%) – groups with strong connections to the jobs market. In contrast, retired and other inactive groups use the technology less, at 13%. Existing evidence further suggests that GenAI can have both augmenting and displacing effects in the labour market: occupations with higher exposure to GenAI tend to experience income gains, while those more exposed to automation face greater risks (Engberg et al., 2025_[11]; Johnston and Makridis, 2025_[12]). As GenAI becomes common practice in more workplaces, these gaps between AI users and non-users are likely to widen, stressing the need for on-the-job AI training, and policies that support awareness and educational initiatives for AI adoption.

Selected countries that collected data for both 2024 and 2025 display an average growth rate of 64%, showing that adoption is still increasing strongly. The share of those aged 55-74 adopting the technology more than doubled over the year, while adoption among those aged 16-24 increased by 41%. The shares of those identified as belonging to unemployed, retired and other groups further from labour markets also more than doubled, while adoption among students increased by 32%.

Figure 3.13. Disparities in AI use are greatest by age

Differences in the share of adults (aged 16 to 74) using any generative AI tools to create content in the last three months, 2025 (or most recent)

Notes: For Colombia, data refer to 2024 with a reference period of 12 months. For Japan, data refer to 2024 and no reference period is specified. In addition, data relate to individuals aged 20 to 69. Gaps are defined as the difference in usage rates between men and women; young (aged 16 to 24) and older people (aged 55 to 74); individuals with high levels of educational attainment and individuals with low levels of educational attainment; and individuals residing in households in the fifth quintile of the household income distribution and individuals in households in the first quintile.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Statlink

Groups report a variety of reasons for not using GenAI. The most common factor, cited by around 36% of individuals in the OECD, is a lack of perceived need. Smaller shares report barriers linked to awareness and capabilities: about 6% indicate they are not aware of GenAI tools, while 9% cite a lack of knowledge on how to use them. Concerns related to privacy, security or safety are less frequently reported, affecting around 4% of individuals (OECD, 2026_[1]).

Countries are increasingly considering how to incorporate AI into their national curriculum. Sweden funded universities to develop short courses to strengthen AI skills among professionals. It also tasked higher education institutions to increase educational offerings related to AI and make efforts to integrate AI elements into the education programmes where relevant (OECD.AI, 2025_[13]). Few education systems prepare students to engage properly with AI and to understand how to use it safely. Consequently, the OECD and the European Commission are joining forces to develop an AI Literacy framework. This will outline the knowledge, skills and attitudes to prepare students in primary and secondary education (OECD/European Union, 2025_[14]; Schleicher, 2025_[15]).

Beyond adoption rates: ICTs as drivers of economic growth and resilience

ICTs are central to modern economic activity, supporting productivity growth, enabling innovation and shaping how firms organise production, deliver services and connect to global markets. While indicators of ICT adoption provide valuable insights into how widely digital tools are used, they offer only a partial view of the economic role of ICTs. A fuller assessment can be achieved based on how ICT value is created, how it circulates across sectors and borders, and how countries position themselves within increasingly interconnected digital production systems.

This broader perspective is particularly relevant in the context of heightened attention to economic resilience and supply-chain interdependence. ICTs are both engines of growth and key channels of international integration. Their production and use rely on global networks, simultaneously shaping countries’ exposure to external shocks and developments. Policymakers can enrich their understanding of ICT performance by moving beyond adoption and diffusion to consider growth dynamics, domestic spillovers5 (i.e. indirect use of ICT value added embodied in intermediate goods and services) and international sourcing structures.

Estimates for 2025 point to continued ICT growth across the OECD, with signs of convergence

Timely estimates of ICT sector performance are essential to understanding shifts in the digital economy. While official statistics provide a reliable baseline, they are often subject to publication lags. Complementary short-term indicators, such as nowcasts based on online search and demand-side data, provide a more immediate view of sector performance and emerging trends (OECD, 2024_[16]; Umana Dajud, 2024_[17]).

Estimates for 2025 point to a continued growth in ICT value-added relative to 2024, with an OECD average of 5% highlighted in red in Figure 3.14. While this aggregate reflects a generally resilient expansion, substantial variation persists across countries. Growth is strongest in Lithuania (9%), Hungary (9%) and Germany (8%). However, several previously high-growth countries from 2024, Finland (11%), Portugal (9%), France (9%) and Czechia (8%), are projected to moderate, suggesting a partial normalisation following strong short-term expansions.

Comparing 2025 with 2024 predictions, the distribution of growth is more compressed, with fewer extreme outliers and a convergence towards the OECD average. Some countries, such as Germany, Hungary, Mexico, Norway and Sweden see an increase relative to 2024, while others, including Czechia, Finland, France, and Portugal, experience a slowdown. At the lower end, Italy (1%), Canada (2%), Switzerland (2%) and the Slovak Republic (1%) are projected to see modest growth. Mid-tier countries, including Belgium (5%), the Netherlands (5%), Norway (5%) and the United States (5%), remain close to the OECD average, reflecting broad-based, moderate expansion.

Figure 3.14. ICT sector growth in 2025 points to broad convergence across the OECD

Predicted (full bars and white diamonds) and observed (grey diamonds) ICT sector growth rates, 2024 and 2025

Notes: Vertical bars show 2025 nowcasted point estimates derived from TiVA-based optimal pooling. Overlapping white diamonds indicate 2024 nowcasts while grey diamonds refer to official statistics. Full country-level predictions for all 2025 (or most recent), with 90% confidence intervals and historical missing years, are reported in the complementary technical paper (OECD, 2026_[18]).

Source: OECD (2026_[20]), Data Kitchen (database), https://goingdigital.oecd.org/datakitchen based on OECD (2026_[19]), STAN Database; OECD (2026_[24]), TiVA Database; and Google Trends (dataset) (accessed from 1 to 20 January 2026).

Statlink

This greater convergence echoes the patterns of AI diffusion, which suggests that the exceptionally rapid acceleration in AI adoption observed in 2023-2024 has given way to a more mature and incremental phase in 2025, as firms move from initial deployment to optimisation and integration (OECD, 2026_[21]).

ICT outputs are critical enablers of economic activity across global value chains

The economic contribution of ICTs extends far beyond firms and sectors that directly produce or adopt digital technologies. ICT goods and services are widely used as intermediate inputs across the economy, supporting production in manufacturing, market services and the public sector. Through these supply-chain connections, the direct purchase of ICTs influences productivity, costs and innovation. At the same time, they have an indirect influence on these factors as ICT-enabled inputs are passed along production networks.

Analysis of OECD Inter-Country Input-Output (ICIO) Tables highlights substantial cross-country differences in the extent to which economic activity relies on ICT outputs (Figure 3.15).6 These linkages can be broken down into two components. Direct ICT forward linkages capture the immediate use of ICT goods and services by downstream sectors (e.g. firms purchasing software, data services or telecommunications). For their part, indirect ICT forward linkages capture spillovers transmitted through supply chains, where ICT-intensive inputs are embedded in other intermediate goods and services used throughout the economy.

In countries such as Germany, ICTs play a particularly strong supporting role, with a large share of this contribution occurring indirectly through suppliers and production networks. This suggests that ICTs are deeply embedded across German value chains, amplifying their impact beyond direct users. The United States and the United Kingdom also show a high reliance on ICT outputs, combining strong direct use with sizeable indirect spillovers transmitted through supply chains. Even in countries where the overall reliance on ICT outputs is more modest, indirect linkages remain substantial. This indicates that ICTs operate as general-purpose technologies: improvements in ICT production or adoption can generate multiplier effects across the economy, while digital bottlenecks can constrain performance well beyond digitally intensive sectors.

Figure 3.15. ICT spillovers represent a significant share of forward linkages in most countries

Total and direct ICT forward linkages, 2022

Notes: Total forward linkages measure the use of ICT outputs by non-ICT sectors, combining both direct consumption (direct ICT forward linkages) and indirect spillovers (indirect ICT forward linkages) through upstream supply chains. Country-level totals exclude ICT sectors and are normalised by the OECD average of non-ICT sectors, so the plotted values are dimensionless.

Source: OECD calculations based on OECD (2026_[22]), Inter-Country Input-Output Tables, https://www.oecd.org/en/data/datasets/inter-country-input-output-tables.html. StatLink contains more data.

Statlink

At the sectoral level, wholesale and retail trade (Section G), followed closely by professional, scientific and technical activities (Section M), show the strongest reliance on ICT outputs (Figure 3.16). In these sectors, ICTs are used intensively and directly to manage complex production processes, logistics and information flows. These sectors combine high direct purchases of ICT services with strong spillovers7 (i.e. indirect use of ICT value added embodied in intermediate goods and services) transmitted through supply chains.

Conversely, sectors such as food products (Section C, Divisions 10 to 12), electricity supply (Section D), real estate (Section L) and machinery manufacturing (Section C, Division 28) make more limited direct use of ICT goods and services but benefit substantially from indirect ICT inputs embedded in intermediate goods and services supplied by upstream firms. For example, construction (Section F) and real estate (Section L) rely heavily on ICT-enabled inputs such as engineering services, even if their own ICT purchases are relatively modest.

This pattern reveals that the economic importance of ICTs extends well beyond digitally intensive sectors. Productivity growth in sectors such as construction (Section F) increasingly relies on the digital sophistication of their suppliers and service providers. As a result, the returns to ICT investment are not confined to adopting firms but accrue across interconnected value chains. This implies that digital transformation strategies should focus on both firm-level adoption and strengthening complementary assets that enable diffusion throughout production networks.

Figure 3.16. Reliance on ICT goods and services varies significantly across sectors

Top 15 non-ICT sectors by direct and indirect ICT forward linkages, OECD average, 2022

Notes: ICT forward linkages measure the total output generated across sectors by one unit of ICT sector output. Values are normalised by the average forward linkage of non-ICT sectors, so the plotted values are dimensionless. A value greater than 1 indicates that the sector is more dependent on ICT inputs than the average non-ICT sector, while a value less than 1 indicates lower dependence.

Source: OECD calculations based on OECD (2026_[22]), Inter-Country Input-Output Tables, https://www.oecd.org/en/data/datasets/inter-country-input-output-tables.html.

Statlink

The United States and the People’s Republic of China drive ICT value creation in the global economy

While ICTs support activity through domestic supply chains, their impact increasingly extends across borders. The international dimension can be examined by considering where ICT value is created and how it supports demand abroad. Domestic ICT value added embodied in foreign final demand captures the extent to which ICT production in one economy ultimately contributes to consumption and investment elsewhere.

In 2022, global ICT value embodied in foreign final demand was highly concentrated among a small number of economies (Figure 3.17). The structure of these cross-border flows can be represented as a network, where nodes correspond to countries and directed links capture the value added generated in one economy and ultimately absorbed in another’s final demand. This network perspective highlights both the scale of leading contributors and the intensity of bilateral value-added corridors that anchor the global digital economy.

Two economies dominate: the United States and the People’s Republic of China (hereafter “China”). China is the single largest supplier of ICT value added embodied in foreign demand, including USD 26.3 billion flowing to the United States alone. Major additional corridors link China to Japan (USD 10.0 billion), Korea (USD 3.9 billion) and Chinese Taipei (USD 2.5 billion), underscoring the central role of East Asia in ICT manufacturing.

The United States is both a major supplier and a major destination within the network. It supplies substantial ICT value added to key partners, including Canada (USD 11.3 billion), China (USD 10.4 billion) and Japan (USD 10.4 billion), while also absorbing large inflows from North American and European partners, such as Canada (USD 7.9 billion), Germany (USD 7.5 billion) and Mexico (USD 7.0 billion). These dense trans-Pacific and North American corridors illustrate the regional clustering and interdependence that characterise global digital production networks.

The composition of these flows reflects complementary specialisations. The outward contribution of the United States is largely driven by ICT services, including software development, digital platforms and advanced technological solutions. Conversely, China’s prominence reflects its central role in ICT manufacturing and hardware production, supported by large-scale and deeply integrated supply chains. Together, these patterns show that global ICT value creation is not only concentrated but organised around a limited number of powerful production hubs connected by high-volume bilateral value-added channels.

Figure 3.17. A few economies drive most global ICT value added absorbed abroad

Domestic ICT value added embodied in foreign final demand, top 20 economies, 2022

Notes: The figure shows domestic ICT value added embodied in foreign final demand, capturing cross-border flows of ICT value through final consumption and investment. Node colour indicates the dominant ICT sub-sector. Node size reflects total ICT value added exported. The distance between nodes indicates the relative strength of ties, and link thickness the magnitude of bilateral value added flows. Please refer to OECD (2025_[23]) for the methodology for deriving the TiVA indicators.

Source: OECD calculations based on OECD (2026_[24]), TiVA Database, http://oe.cd/dk-tiva.

Statlink

OECD Member countries vary in how they integrate into global digital supply chains

While some countries stand out as major suppliers of ICT value to global markets, all countries rely to some extent on ICT inputs produced abroad. The share of foreign ICT value added embedded in domestic ICT demand provides insight into how deeply countries are integrated into international digital production networks. It also sheds light on the exposure of countries to cross-border supply dynamics.

Across the OECD, the degree of international ICT input integration varies widely but has increased overall between 2010 and 2022 (Figure 3.18). Smaller and more open countries tend to rely more heavily on foreign ICT inputs, reflecting their roles as assembly hubs or specialised service providers within global value chains (Johansson and Olaberría, 2014_[25]; OECD, 2021_[26]). Larger countries generally rely less on foreign inputs, supported by broader domestic production bases, although none are fully insulated from international linkages. These differences reflect a combination of factors, including countries’ industrial structure, the balance between ICT manufacturing and services, and their position within global value chains.

Figure 3.18. The dependence on imported ICT inputs varies significantly across the OECD

Foreign ICT value added as a share of domestic ICT demand, 2010 and 2022

Notes: Foreign ICT value added is measured as the share of domestic ICT demand that originates abroad. ICT demand includes both intermediate and final use of ICT goods and services. Please refer to OECD (2025_[23]) for the TiVA indicators’ methodology.

Source: OECD calculations based on OECD (2026_[24]), TiVA Database, http://oe.cd/dk-tiva.

Statlink

ICT sourcing is highly concentrated around a small number of strategic hubs, reflecting both efficiency gains and deep interdependencies

The structure of ICT sourcing provides critical insight into how OECD Member countries navigate global production networks. Concentration in sourcing – the extent to which foreign ICT inputs originate from a limited number of partner economies – often arises from efficiency gains and scale effects. However, it also defines a country’s exposure to regional or partner-specific supply shocks.

In 2022, China emerged as the single largest source of foreign ICT value added for OECD Member countries (Figure 3.19). For many OECD Members, China accounted for a substantial share of foreign ICT inputs, particularly in hardware-related segments, where production benefits from scale and deeply integrated manufacturing ecosystems. In several countries, including Australia, Japan, Korea and Mexico, more than one-quarter of foreign ICT value added originated from China, highlighting its central role in global ICT production.

Other major countries also play an important role as sources of foreign ICT value added, reflecting complementary areas of specialisation. The United States is a key supplier for several OECD Member countries, particularly in high-value digital services and advanced ICT solutions, while Germany plays a central role within European regional production networks. These patterns underscore the importance of long-standing, trusted partnerships in shaping ICT sourcing structures.

Figure 3.19. China is the leading source of foreign ICT value added for most OECD Member countries

Share of foreign ICT value added accounted for by the largest supplier, 2022

Notes: The figure shows, for each OECD Member country, the share of foreign ICT value added accounted for by the single largest foreign supplier. ICT value added refers to value generated in foreign ICT sectors and embodied in domestic ICT demand. The figure highlights concentration in sourcing, not overall reliance on foreign ICT inputs. Please refer to OECD (2025_[23]) for the TiVA indicators’ methodology.

Source: OECD calculations based on OECD (2026_[24]), TiVA Database, http://oe.cd/dk-tiva. StatLink contains more data.

Statlink

From a policy perspective, high concentration in ICT sourcing does not automatically imply vulnerability. In many instances, it reflects stable, high-efficiency supply relationships that accelerate the diffusion of frontier technologies. However, the stark contrast between the concentration in hardware manufacturing and the relative diversity in IT services suggests that policy efforts to enhance supply-chain resilience should be sector specific. For OECD Member countries, maintaining reliable and trusted international partnerships remains the cornerstone of a resilient and interconnected digital economy.

Maximising the benefits of digital technologies for the economy and society

This chapter highlights how frontier digital technologies and ICT outputs are reshaping business activity, individual behaviour and economic performance, while also exposing persistent gaps. Cloud computing has become widely adopted across firms, forming a foundation for complementary digital tools. Conversely, AI and big data analytics remain concentrated among larger, knowledge-intensive firms, with full integration across all three technologies still limited. Growth in AI adoption surged in 2024 but moderated in 2025, and projections suggest that uptake may continue to be skewed towards large firms. GenAI is rapidly diffusing among individuals, with younger, highly educated and higher-income users leading use.

Beyond uptake, ICT goods and services underpin productivity, innovation and competitiveness across the economy. Forward and horizontal linkages illustrate that indirect use of ICT outputs embodied in intermediate inputs and transmitted through supply chains (i.e. ICT spillovers8) extend well beyond direct users, supporting other sectors and strengthening interconnections within the ICT ecosystem. These effects are amplified by international integration: OECD Member countries rely to varying degrees on foreign ICT inputs, while sourcing structures and the concentration of suppliers. Notably, China plays a central role in ICT manufacturing, while other major players like the United States make complementary contributions in IT services, and shape exposure to global production networks and the diffusion of ICT value across borders.

Policy responses must therefore adopt a systemic perspective that goes beyond promoting uptake of individual technologies. Expanding access to cloud services, alongside targeted support for AI and big data analytics, can help narrow firm-size gaps, foster co-ordinated adoption of technology bundles and accelerate enterprise-level transformation. At the same time, policies should recognise the international dimension of ICTs, supporting stable and trusted cross-border partnerships, and mitigating risks arising from concentrated sourcing of ICTs.

Strengthening digital infrastructure, interoperability and open standards can amplify ICT spillovers, benefitting both lagging sectors and smaller firms. Taken together, these approaches can enhance productivity and innovation potential. At the same time, they can ensure that digitalisation gains are more broadly shared across firms, sectors and society, and that countries remain resilient and well positioned within global digital production networks.

Annex 3.A. Methodology for the diffusion path forecasts and regression tables

The diffusion paths are estimated using ordinary least squares (OLS) regressions on logistic-transformed adoption rates, a standard approach for modelling S-shaped technology diffusion curves. For each technology and firm size, the logit of the OECD average adoption rate was regressed on year. This model set-up ensures consideration of country-specific, time-invariant factors influencing adoption.

Forecasts extending to 2030 were derived by applying the inverse logit function to the predicted logit values from the fitted models. While these forecasts provide an indication of potential future trends based on historical data, the actual diffusion of fast-growing technologies can be influenced by unforeseen market dynamics, rapid technological advancements and evolving policy landscapes. Therefore, these projected pathways should be interpreted as informed estimates rather than precise predictions. The estimated specification can be found below:

Where , is the intercept, represents the coefficient for year, indicating the average annual change in the log-odds of adoption, while is the error term.

Table 3.A.1. Estimated diffusion paths for cloud computing and AI

OLS regressions, OECD average, by firm size, 2010-2025 (or most recent)

	All firms (+ 10 employees)		Large firms (+ 250 employees)
	Cloud computing	AI	Cloud computing	AI
Year	0.1485*** (0.0070)	0.2747*** (0.0410)	0.2065*** (0.0068)	0.2994*** (0.0406)
R-squared	0.9783	0.9181	0.9894	0.9314
Observations	12	6	12	6

How to read the table: The dependent variable is the logit-transformed adoption rate, i.e. , where is the adoption rate in country in year . Robust standard errors are in parentheses. *, ** and *** denote statistical significance at the 10%-, 5%- and 1%-level, respectively.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Annex 3.B. Methodology for nowcasting AI uptake in enterprises in 2025

This annex describes the statistical framework used to generate 2025 nowcasts of AI adoption for enterprises with ten or more employees in countries in which official data are not available to date. The approach is designed for settings in which historical time series are short and uneven across countries. It combines a flexible diffusion model grounded in the innovation literature with a cross-country borrowing mechanism that stabilises estimation under limited data availability. Model performance is assessed through rolling out-of-sample validation, and uncertainty ranges are reported to reflect predictive precision.

The generalised logistic curve

The diffusion of new technologies typically follows an S-shaped trajectory: gradual early uptake, a phase of acceleration and eventual saturation. While the standard logistic function captures this pattern, it imposes symmetry around the inflection point. To allow for greater flexibility, the analysis employs the generalised logistic (or Richards) curve (Richards, 1959_[27]), which introduces an additional shape parameter governing asymmetry. The functional form is:

Where  is the AI adoption rate in year ;  is the saturation level (upper asymptote);  is the growth rate (steepness of diffusion);  is the inflection point (year of maximum marginal growth); and  governs the asymmetry: when  , the function collapses to the standard logistic curve;  implies earlier acceleration and slower convergence;  implies later acceleration and faster convergence.

The additional flexibility provided by is particularly relevant in the context of AI, where adoption patterns differ substantially across countries due to variations in industrial structure, digital infrastructure, skills and policy frameworks, for example. The Richards specification allows for heterogeneous diffusion speeds and take-off dynamics, while maintaining a coherent theoretical structure consistent with the innovation diffusion literature (Tsoularis and Wallace, 2002_[28]; Rogers, 2003_[29]; Meade and Islam, 2006_[6]).

Data and estimation

The analysis uses annual data from the OECD ICT Access and Usage Database (OECD, 2026_[1]) beginning in 2017. The focus is on enterprises with ten or more employees. Only countries with at least two observed data points are retained for nowcasting, ensuring a minimal empirical basis for estimation.

Because country-level time series are short, the first step consists of estimating a global reference trajectory. A Richards curve is fitted to the mean adoption rate by year across the OECD. This pooled estimation yields global parameters , and , which capture the typical diffusion pattern across OECD Member countries. The saturation parameter is capped at 50%, calibrated based on observed maximum adoption and expected remaining diffusion potential.

For countries with fewer than four observations, full country-specific estimation of all Richards parameters is not statistically robust. A borrowed approach is therefore applied, wherein the global shape parameters ( , , ) are retained while a country-specific scaling factor  is estimated to adjust the effective saturation level. The underlying intuition is that while the pace and timing of adoption may follow a common pattern across countries, the ultimate level of adoption can vary according to national circumstances, which the scaling factor captures efficiently.

Performance assessment and uncertainty quantification

Model performance is evaluated using a rolling one-step-ahead out-of-sample Root Mean Square Error (RMSE), obtained by re-estimating the model up to year , predicting year and repeating iteratively. Across all evaluated countries, the Richards borrowed specification produced lower or more stable out-of-sample errors than alternative logistic models. In particular, standard logistic variants generated implausible trajectories and large forecast dispersion, whereas the Richards borrowed model maintained economically coherent dynamics and bounded uncertainty. This superior predictive performance supports its selection for the final predictions.

Alternative estimation techniques were explored, including machine-learning approaches such as random forests and gradient boosting. Their application proved challenging given the short length of the available time series (at most seven annual observations per country). With limited temporal depth, these data-driven methods tended to overfit, producing unstable out-of-sample predictions. By contrast, the parsimonious structure of the Richards borrowed model, anchored in a theoretically grounded diffusion function and borrowing strength across countries, delivered more robust and interpretable results. This approach was therefore preferred for its combination of flexibility, stability and economic interpretability.

Each prediction for 2025 is accompanied by a predictive interval constructed as:

Where  is the point estimate, is the out-of-sample RMSE and  represents years beyond the last observation. The term   imposes a horizon-dependent penalty that progressively widens intervals as the forecast horizon extends, reflecting the increasing uncertainty inherent in longer-range extrapolation. Bounds are truncated at 0 and 100 to respect natural adoption limits. These intervals account for model error, parameter uncertainty and extrapolation risk, and should therefore be interpreted as predictive ranges rather than strict statistical confidence intervals.

Estimation results

Table 3.B.1 presents the 2025 nowcasts of AI uptake among enterprises with ten or more employees for selected OECD Member countries, together with model-based measures of predictive uncertainty.

Table 3.B.1. Nowcasts of AI uptake in enterprises for selected OECD Member countries

Richards borrowed diffusion model, enterprises with ten or more employees, 2025

	Point estimate (%)	RMSE (p.p.)	Lower bound (%)	Upper bound (%)
Australia	7.6	1.7	4.4	10.8
Canada	12.5	0.7	11.4	13.6
Colombia	23.8	1.4	20.9	26.7
Japan	13.0	2.5	8.8	17.2
Korea	33.5	12.4	14.9	52.1
Switzerland	18.5	3.6	12.4	24.6
United States	17.5	4.2	9.2	25.8

Notes: Bounds are derived from out-of-sample Root Mean Square Error (RMSE), scaled by a horizon-dependent factor to account for growing uncertainty with forecast distance. The relatively high RMSE for Korea reflects a sharp acceleration in observed AI uptake between 2021 and 2023, which generates larger one-step-ahead forecast errors in the rolling validation exercise.

Source: OECD calculations based on OECD (2026_[1]), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage.

Annex 3.C. Methodology for ICT linkages and results

This annex describes the approach used to compute ICT forward linkages based on the OECD Inter-Country Input-Output (ICIO) Tables (OECD, 2026_[22]). The objective is to capture the extent to which output from ICT-producing industries is used as an input in the production of goods and services across countries. This, in turn, allows tracing the role of ICT as an enabler of activity throughout global value chains.

The ICIO framework

Forward linkages are estimated using the Ghosh (supply-driven) input-output model (Ghosh, 1958_[30]). Let = inter-industry transaction matrix across all countries and sectors (with elements , i.e. deliveries from sector in country to sector in country ); and = vector of total outputs by country-sector, with elements . The output coefficients matrix is defined as:

Where each entry shows the share of output from sector in country delivered as input to sector in country . The Ghosh inverse is then given by:

Where is the identity matrix. traces how a unit change in the supply of one country-sector propagates through downstream production chains globally.

ICT forward linkages

ICT forward linkages are obtained by focusing on the ICT-producing sectors identified in the OECD ICIO classification (C26 “Computer, electronic and optical products”, J61 “Telecommunications”, J62T63 “IT and other information services”) across all countries.

For each year , country and non-ICT sector :

• Direct forward linkage (immediate purchase of ICT outputs):

  

• Total forward linkage (direct and indirect dependence, via global production chains):

  

• Indirect forward linkage:

  

References

[4] Calvino, F., C. Criscuolo and A. Ughi (2024), “Digital adoption during COVID-19: Cross-country evidence from microdata”, OECD Science, Technology and Industry Working Papers, No. 2024/03, OECD Publishing, Paris, https://doi.org/10.1787/f63ca261-en.

[32] Calvino, F. et al. (2024), “A sectoral taxonomy of AI intensity”, OECD Artificial Intelligence Papers, No. 30, OECD Publishing, Paris, https://doi.org/10.1787/1f6377b5-en.

[8] Calvino, F., D. Haerle and S. Liu (2025), “Is generative AI a General Purpose Technology?: Implications for productivity and policy”, OECD Artificial Intelligence Papers, No. 40, OECD Publishing, Paris, https://doi.org/10.1787/704e2d12-en.

[3] Cho, J. et al. (2023), “What’s driving the diffusion of next-generation digital technologies?”, Technovation, Vol. 119, p. 142177, https://doi.org/10.1016/j.technovation.2022.102477.

[11] Engberg, E. et al. (2025), “Artificial intelligence, tasks, skills, and wages: Worker-level evidence from Germany”, Research Policy, Vol. 54/8, p. 105285, https://doi.org/10.1016/j.respol.2025.105285.

[9] Eurostat (2025), “Use of artificial intelligence in enterprises”, https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises#:~:text=In%202025%2C%20among%20EU%20enterprises,protection%20and%20privacy%20(48.83%25) (accessed on 4 March 2026).

[30] Ghosh, A. (1958), “Input-output approach to an allocation system”, Economica, Vol. 25/97, pp. 58-64, https://doi.org/10.2307/2550694.

[25] Johansson, Å. and E. Olaberría (2014), “Long-term Patterns of Trade and Specialisation”, OECD Economics Department Working Papers, No. 1136, OECD Publishing, Paris, https://doi.org/10.1787/5jz158tbddbr-en.

[12] Johnston, A. and C. Makridis (2025), “The labor market effects of generative AI: A difference-in-differences analysis of AI exposure”, SSRN, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5375017#paper-references-widget.

[6] Meade, N. and T. Islam (2006), “Modelling and forecasting the diffusion of innovation – A 25-year review”, International Journal of Forecasting, Vol. 22/3, pp. 519-545, https://doi.org/10.1016/j.ijforecast.2006.01.005.

[21] OECD (2026), “AI use by individuals surges across the OECD as adoption by firms continues to expand”, https://www.oecd.org/en/about/news/announcements/2026/01/ai-use-by-individuals-surges-across-the-oecd-as-adoption-by-firms-continues-to-expand.html (accessed on 2 February 2026).

[20] OECD (2026), Data Kitchen (database), https://goingdigital.oecd.org/datakitchen (accessed on 3 November 2026).

[1] OECD (2026), ICT Access and Usage Database, https://oe.cd/dx/ict-access-usage (accessed on 25 February 2026).

[22] OECD (2026), Inter-Country Input-Output Tables (dataset), https://www.oecd.org/en/data/datasets/inter-country-input-output-tables.html (accessed on 12 January 2026).

[19] OECD (2026), Structural Analysis (STAN) Database, https://www.oecd.org/en/data/datasets/structural-analysis-database.html (accessed on 24 June 2026).

[24] OECD (2026), Trade in Value Added (TiVA) Database, http://oe.cd/dk-tiva (accessed on 12 January 2026).

[23] OECD (2025), Guide to OECD Trade in Value Added (TiVA) Indicators, 2025 edition, OECD, Paris, https://stats.oecd.org/wbos/fileview2.aspx?IDFile=2143f34e-6feb-41a9-abaf-cb52132608c4.

[7] OECD (2024), “Fostering an inclusive digital transformation as AI spreads among firms”, OECD Policy Briefs, No. 8, OECD Publishing, Paris, https://doi.org/10.1787/5876200c-en.

[16] OECD (2024), OECD Digital Economy Outlook 2024 (Volume 1): Embracing the Technology Frontier, OECD Publishing, Paris, http://dx.doi.org/10.1787/a1689dc5-en.

[10] OECD (2024), “The OECD Truth Quest Survey: Methodology and findings”, OECD Digital Economy Papers, No. 369, OECD Publishing, Paris, http://dx.doi.org/10.1787/92a94c0f-en.

[26] OECD (2021), “Global value chains: Efficiency and risks in the context of COVID-19”, OECD Policy Responses to Coronavirus (COVID-19), OECD Publishing, Paris, https://doi.org/10.1787/67c75fdc-en.

[18] OECD (2026), Advances in nowcasting the growth rate of the ICT sector, OECD Publishing, Paris, https://doi.org/10.1787/e7d36875-en.

[13] OECD.AI (2025), “The government strengthens AI skills among professionals and students”, https://oecd.ai/en/dashboards/policy-initiatives/the-government-strengthens-ai-skills-among-professionals-and-students-5388 (accessed on 24 September 2025).

[14] OECD/European Union (2025), Empowering learners for the age of AI: An AI literacy framework for primary and secondary education (Review draft), OECD/European Union, https://ailiteracyframework.org (accessed on 24 September 2025).

[27] Richards, F. (1959), “A flexible growth function for empirical use”, Journal of Experimental Botany, Vol. 10/2, pp. 290-301, https://doi.org/10.1093/jxb/10.2.290.

[29] Rogers, E. (2003), Diffusion of Innovations, 5th Edition, The Free Press, New York, https://teddykw2.wordpress.com/wp-content/uploads/2012/07/everett-m-rogers-diffusion-of-innovations.pdf.

[15] Schleicher, A. (2025), “Global perspectives on education and skills”, OECD Education and Skills Today, https://oecdedutoday.com/new-ai-literacy-framework-to-equip-youth-in-an-age-of-ai/ (accessed on 29 April 2025).

[31] Statistics Austria (2025), “Around a quarter of the population rate their knowledge of AI as good”, Statistics Austria, Vienna, https://www.statistik.at/fileadmin/announcement/2025/05/20250515KI_AnalyseEN.pdf.

[2] Sunyaev, A. (2020), “Cloud computing”, in Internet Computing, Springer, Cham, https://link.springer.com/book/10.1007/978-3-031-61014-1.

[28] Tsoularis, A. and J. Wallace (2002), “Analysis of logistic growth models”, Mathematical Biosciences, Vol. 179/1, pp. 21-55, https://doi.org/10.1016/S0025-5564(02)00096-2.

[17] Umana Dajud, C. (2024), “Nowcasting the growth rate of the ICT sector”, OECD Digital Economy Papers, No. 362, OECD Publishing, Paris, http://dx.doi.org/10.1787/eb4938a0-en.

Notes

← 1. As with any predictions, uncertainty increases with nowcast distance from the last observed data point. Annex 3.B reports predictive ranges alongside each nowcast and provides full methodological details. These ranges should be interpreted as indicative bounds reflecting historical forecast error and extrapolation risk, rather than precise confidence intervals.

← 2. These patterns are broadly consistent with recent findings showing that AI intensity is highest in knowledge- and data-driven sectors such as IT services, media and telecommunications, and lowest in traditional manufacturing and construction industries (Calvino et al., 2024_[32]).

← 3. The data refer to the proportion of businesses not using AI due to perceived high costs. The following options were included in the survey as reasons that AI is not considered useful for business operations: i) a lack of relevant expertise in the enterprise; ii) an incompatibility with existing equipment/software or systems; iii) difficulties with availability or quality of the necessary data; iv) concerns regarding violation of data protection and privacy; v) a lack of clarity about the legal consequences (e.g. liability in case of damage caused by the use of AI); and vi) ethical considerations.

← 4. Despite the widespread use of generative AI, public understanding and confidence remain limited. An Austrian study shows that 73% of individuals report having little to no knowledge of AI, and only around one-third express a positive attitude towards its increasing use in society (Statistics Austria, 2025_[31]). This cautious outlook is mirrored in the OECD Truth Quest Survey, which finds that in 2024, only about 30% of adults believed that AI would have a positive impact on their lives (OECD, 2024_[10]).

← 5. ICT spillovers refer to the indirect use of ICT outputs through production networks. In the context of Inter-Country Input-Output Tables (OECD, 2026_[22]), they capture ICT value added embodied in intermediate goods and services that are used as inputs across sectors throughout the economy. Forward linkages are estimated using the Ghosh (supply-driven) input-output framework (Ghosh, 1958_[30]). Further methodological details are provided in Annex 3.C.

← 6. Additional methodological details are provided in Annex 3.C.