AI and productivity : a source of calamity or prosperity ?

The topic of artificial intelligence (AI) has been influencing market trends for several years. From microcomponent manufacturers to software developers, the ongoing transformation of economies is a narrative punctuated by both hope and fear. The impact on the workforce, the potential for automation to replace human labour and the need for substantial investment in computing and energy infrastructure are all factors that are gradually making the technological and digital revolution visible and measurable. The speed at which this new technology is being adopted, combined with apparent productivity gains, raises questions about the scale of the transformation in the production of goods and services. Beyond this initial channel through which the technological shock is transmitted, a second area of application remains to be explored: the capacity of innovation to generate new ideas, which makes the economies of scale associated with the production and sharing of knowledge virtually infinite. This note provides an overview of the current state of economic research on the impact of this innovation on growth, the place of human capital within the production structure of economies, and the productive capacity of actors adopting this new wave of innovation.

The impact of innovation on productivity: the cornerstone of economic development

Technological and organisational innovations have consistently shaped the development of the modern economy. From the industrial deployment of the steam engine and the standardisation of international trade to electrification and digitalisation, not to mention the introduction of Taylorism and Kanban, the development of the global economy has continually increased the efficiency of the production process.

Figure 1 illustrates productivity gains in the US, the eurozone and Japan since the late 19th century. Deep learning, or AI, is therefore part of a long line of technical innovations that will shape the creation of added value through their use and integration into current production processes, or the creation of future modes and relationships of production. However, Figure 1 shows that the labour productivity growth rate varies across economic regions. This suggests that the efficiency of production methods stems from a combination of investment, research, legal and regulatory frameworks, as well as the diversity of production ecosystems in the US, Japan, and the euro area.

As Aghion and Bunel (2024) point out, the innovation shock brought about by AI is likely to be subject to a transmission lag similar to that observed during the introduction of electricity or digitalisation. This delay is due to the inherent inertia of the physical and intellectual structures that are required to assimilate innovation, such as incompatible machinery and resources that must be adapted to facilitate adoption.

Figure 1: Labour productivity since 1890(left) and recent trends(1999 = 100), Bergeaud, A., Cette, G. and Lecat, R. (2016)

Source: LSEG Eikon and NIM Solutions. Data as of 15/03/2026

The impact of AI on future value creation depends on current and future investment, the adoption rates of these technologies, and the reorganisation of production structures. Economies that respond effectively to these challenges are likely to experience significant productivity gains, thereby strengthening their dominant position.

Massive investments are required, even though their future profitability remains an unproven question

Adapting existing production structures to the technological revolution embodied by AI therefore relies primarily on investment. This includes investments in computing capacity, the efficient use of energy resources to power this capacity, as well as research and development of new applications capable of harnessing this capacity to generate efficiency gains in production processes (for example, automation and robotics in industry or services). Figure 2 illustrates the relatively heterogeneous investment trends in information and communication technologies or intellectual property products across different economies since the mid-1990s. Furthermore, the data suggest an acceleration in US investment in information and communication technology (ICT), in 2025, alongside a relative decline in euro area investment in intangible capital.

Figure 2: Nominal investment in intellectual property products (IPP) and information and communication technology (ICT), 1995= 100, local currency

Source: LSEG Eikon and NIM Solutions. Data as of 15/03/2026

Some fear that there is excessive investment in both tangible and intangible capital, particularly among so-called ‘hyperscalers’. In 2025, US technology companies invested a total of $380 billion, and forecasts for 2026 currently stand at over $700 billion. These figures inevitably raise two questions: that of optimal capital allocation and, ultimately, that of the dynamics of return on investment, which could come under downward pressure. These questions encapsulate the doubts surrounding market capitalisations and earnings growth forecasts for companies in the sector. At this stage, we view these investment needs as inherent to a ‘classic’ capital-intensive technological revolution. Companies undergoing these substantial transformations are likely to reap the benefits of their future oligopolistic position.

To understand the implications for the aggregate production function of economies, it is necessary to appreciate how labour and capital would be reallocated at the sectoral level in the event of rapid and widespread adoption of innovation. Aghion, Jones, and Jones (2017) analyze the presumed link between increasing automation and the capital intensity of value creation. While it is certain that automation leads to increasing returns to capital at the expense of labour in certain sectors, the authors invoke Baumol’s law, or the “disease of costs,” to argue that at the aggregate level, despite near-total automation, the share of returns to capital in value added would be far from erasing returns to labour. Sectors with low exposure to productivity gains (eg health care, education, arts, and culture) attract labour displaced by productivity gains in other sectors, particularly industrial sectors. Depending on the cross-elasticity of demand, this labour force generally sees its wages rise in the same proportion as in sectors with high productivity gains. The question, then, is which sectors and tasks are likely to be most exposed to AI. Although this is difficult to determine, several academic studies have sought to use survey data to assess the extent to which different sectors are exposed to AI and to draw conclusions about the future of work.

Will the role of labour gradually decline?

Acemoglu (2024) argues that the automation of tasks, particularly in labour-intensive sectors such as low-value-added services, will widen the gap between returns on capital and returns on labour, to the benefit of capital. Consequently, it is essential to determine how exposed companies and their sectors are to AI in order to understand where productivity gains might materialise. These figures are the subject of legitimate debate in the current literature. Figure 3 is taken from a recent attempt to measure this exposure. It shows the proportion of exposed firms and a negative exposure rate. This variable is characterised by a key distinction arising from the use of AI: certain tasks can be replaced by AI, while others can be ‘enhanced’. The negative exposure rate is calculated by dividing the number of responses in the survey sent by the authors to 750 business leaders regarding ‘tasks replaced’ by AI by the number of responses regarding ‘tasks improved’ by AI.

Figure 3: Sectoral exposure and replacement rates (Baslandze et al., 2026)