An IMF briefing to EU finance ministers says artificial intelligence could lift European productivity by about 1% over five years, but warns that fragmented markets, strained electricity grids and dependence on U.S. and Chinese technology could determine who captures the gains.

A growth debate moves from laboratories to finance ministries

Artificial intelligence has spent much of the past three years moving through Europe as a technology story: new models, larger computing clusters, regulation, start-up funding and an increasingly public contest with the United States and China. This weekend it moved more decisively into the language of macroeconomics. At an informal meeting of European Union finance ministers in Dublin, an International Monetary Fund background paper framed AI not simply as another digital sector but as a potential source of productivity growth capable of influencing wages, investment, energy demand and the distribution of economic power across the continent.

The headline number is deliberately modest. According to the IMF paper reported by Reuters on September 19, artificial intelligence could raise European productivity by about 1% over five years. That would not amount to an economic revolution on its own, and the estimate is not a forecast of one percentage point of additional growth every year. It is a cumulative medium-term productivity effect under assumptions about adoption, exposure and the capacity of firms to reorganise work around the technology. Yet in an economy where policymakers have spent years worrying about weak productivity, ageing populations and a widening innovation gap with the United States, even a modest increase has attracted the attention of finance ministers.

The paper also carries a more uncomfortable message. The gains are unlikely to arrive evenly. Countries with deeper capital markets, stronger digital infrastructure, higher wages and a larger concentration of professional services are better positioned to adopt AI quickly. Regions with weaker grids, less venture finance, thinner technology ecosystems or a smaller base of firms able to make large intangible investments may lag. What looks at first like a productivity opportunity therefore doubles as a test of whether the EU can prevent a new technology cycle from widening the economic gaps it has spent decades trying to narrow.

The 1% figure is useful — and easy to misuse

The IMF’s estimate deserves careful interpretation because productivity projections around artificial intelligence can easily become inflated into claims that the technology has already transformed economic output. It has not. The Fund’s earlier research on AI and productivity in Europe, published in 2025, estimated a preferred medium-term gain of roughly 1.1% over five years. The exercise was built around task exposure, expected adoption and evidence on the productivity effect of AI tools in occupations where they can either automate or augment work. It was designed to capture incremental effects rather than more speculative long-run changes such as entirely new industries or faster scientific discovery.

That distinction matters. The result describes a plausible one-off lift in the level of productivity as AI diffuses through existing economic activity. It does not establish that Europe will permanently add a full percentage point to annual productivity growth. Nor does it guarantee that firms will capture the theoretical gains. Installation of software is not the same as organisational change. Companies must redesign processes, train workers, integrate proprietary data, create safeguards and often invest in computing systems before an AI tool becomes economically meaningful.

The estimates also vary sharply by country and scenario. Earlier IMF work found that richer European economies could benefit more because their employment structures contain a greater share of professional, managerial and administrative tasks exposed to generative AI and because higher labour costs strengthen the financial incentive to automate or augment those tasks. Lower-income economies may adopt more slowly. In an optimistic scenario cited by the Fund, Norway’s gain could be far above the continental average, while countries such as Romania would see a smaller boost. Those are modelled possibilities, not predictions of national GDP.

The policy implication is less dramatic than the slogans surrounding AI, but more consequential. Europe does not merely need access to capable models. It needs the economic conditions in which thousands of firms can turn those models into cheaper, faster or better production. That is why the Dublin discussion was fundamentally about market structure rather than software features.

Six in ten workers exposed does not mean six in ten jobs disappear

The most politically sensitive figure in the IMF briefing is its estimate that around 60% of workers in advanced European economies are employed in occupations highly exposed to artificial intelligence. Exposure, however, is not synonymous with redundancy. A job can be highly exposed because a large share of its tasks can be assisted by AI, even when the worker remains essential. Lawyers may draft faster, engineers may test more design alternatives, accountants may automate reconciliation, doctors may gain decision-support tools and programmers may generate or review code more efficiently without the occupation vanishing.

The central economic question is the balance between substitution and complementarity. Where AI replaces routine tasks, demand for certain roles could decline or hiring could slow. Where it augments workers, output per employee can rise, potentially increasing wages or allowing firms to expand. The same technology can produce both effects inside a single occupation. Administrative staff may spend less time composing standard documents and more time handling exceptions. Analysts may automate first-pass research while devoting more hours to judgment, client interaction and verification.

That ambiguity explains why the distributional risk is so important. Workers whose tasks are complemented by AI may capture part of the productivity gain through higher wages or improved career prospects. Workers whose tasks are easily automated and whose skills are less transferable may bear more of the adjustment cost. The effect will depend on labour-market institutions, training systems, bargaining power and the speed with which expanding sectors absorb displaced workers.

Europe enters this transition with advantages and constraints. Stronger social insurance can soften disruption, and public training systems can support reskilling. But rigidities that make it difficult for workers to move between regions, professions or employers can slow adjustment. Housing shortages in productive cities, inconsistent recognition of professional qualifications and fragmented pension or benefit systems all matter because technological change is easier to absorb when people can move toward new opportunities.

For finance ministers, that turns AI policy into labour-market policy. The fiscal debate is not just about subsidising chips or data centres. It is also about whether education systems, employment services and social protection can help workers adapt quickly enough for productivity gains to translate into broad income growth rather than a narrow increase in returns to capital and scarce technical skills.

Europe’s single market is still incomplete where AI needs it most

The IMF’s most conventional recommendation may also be the hardest to implement: complete the single market. Europe has spent decades removing barriers to the movement of goods, yet services, capital, data and professional activity remain more fragmented than the legal architecture of the EU might suggest. That fragmentation is especially costly for digital companies, which often need scale before they become globally competitive.

An AI start-up can technically sell software across borders from its first day, but regulation, procurement rules, language, taxation, sector-specific licensing and access to local datasets can still raise the cost of expanding across 27 member states. The result is that a company founded in a large U.S. market may reach a continental customer base more easily than a European rival attempting to scale across national systems. For mature companies, fragmentation can also delay the diffusion of AI because technology vendors must adapt products and compliance processes to multiple national environments.

The IMF argues that a deeper single market could spread adoption and its benefits more evenly. In practical terms, that means reducing barriers to cross-border services, harmonising standards where possible and making it easier for firms to operate at European rather than national scale. The objective is not deregulation for its own sake. It is to lower the fixed cost of deploying a technology whose economics reward large markets and repeated use.

This is closely related to the long-running European competitiveness debate. Former European Central Bank president Mario Draghi and the European Commission have repeatedly identified fragmentation as a brake on investment and innovation. AI gives that argument new urgency because the technology is unusually dependent on scale: large pools of data, computing capacity, specialised talent and capital can reinforce one another. If those inputs remain divided by national markets, Europe may own excellent research while allowing much of the commercial value to accumulate elsewhere.

The difference between invention and diffusion is crucial. Europe does not need every company to build a frontier model. Most productivity gains will probably come from ordinary businesses using AI to improve logistics, engineering, finance, customer service, manufacturing, research, administration and design. The single market determines how quickly those applications can move from a successful pilot in one country to widespread use across the bloc.

Capital is the second bottleneck

Artificial intelligence is capital intensive in ways that do not fit comfortably with Europe’s traditional financial structure. Frontier models require massive computing expenditure, while many smaller AI companies depend on intangible assets such as software, algorithms, proprietary data and intellectual property. Those assets can be difficult to value and are poor collateral for bank lending. The financing model therefore favours deep equity markets, venture capital and investors willing to tolerate high failure rates.

Europe has strong banks but thinner venture and growth-capital markets than the United States. Promising companies frequently find early financing at home and then turn to U.S. investors as their capital needs rise. Some eventually shift parts of their operations or listing ambitions across the Atlantic. That pattern can weaken the domestic ecosystem because the largest returns, executive networks and future financing decisions migrate with the companies.

The IMF has argued that a more integrated European capital market would help channel the continent’s large pool of savings toward risky technology investment. The issue is not a shortage of money in aggregate. European households and institutions hold enormous financial wealth. The problem is intermediation: too little of that capital reaches high-growth companies that cannot be financed through conventional debt.

This is why the EU’s long-discussed capital-markets project now intersects with AI policy. Harmonised insolvency rules, deeper cross-border investment funds, more liquid public markets and better treatment of intellectual property on corporate balance sheets are not normally described as technology policy. Yet they could be as important to Europe’s AI competitiveness as subsidies for computing hardware.

The danger is a split system in which public money finances infrastructure while the most successful private companies remain dependent on foreign capital. That would leave taxpayers carrying part of the investment risk without guaranteeing that the highest-value businesses scale inside Europe. A durable AI strategy therefore requires a financing structure in which private European capital can participate from seed funding through global expansion.

The electricity bill is becoming a macroeconomic variable

The most immediate physical constraint is power. AI is digital at the user interface but industrial in its infrastructure. Training and operating large models requires data centres filled with advanced processors, cooling systems, networking equipment and backup power. The IMF paper told ministers that Europe’s data centres already consume roughly 3% of the continent’s electricity and that demand is set to rise as AI use expands. Major clusters around Frankfurt, London, Amsterdam, Paris and Dublin are particularly exposed to local grid pressure.

The European Commission, drawing on International Energy Agency data, says data centres account for about 1.5% of global electricity consumption, around 415 terawatt-hours annually, and projects that figure could more than double toward 945 terawatt-hours by 2030. Europe, meanwhile, has set an objective of tripling data-centre capacity by 2035. Those two ambitions — more computing and a cleaner, affordable electricity system — can be compatible, but only if investment in generation, transmission and grid flexibility keeps pace.

That makes electricity prices a competitiveness issue. A model trained or served in a region with abundant, reliable power has a structural cost advantage over one operating where grid connections are delayed or wholesale prices are persistently high. Europe’s industrial debate has already been dominated by energy costs since the Russian invasion of Ukraine disrupted the continent’s old gas model. AI adds another category of large, relatively concentrated electricity demand.

The pressure is not uniform. Some regions can absorb new data centres with renewable generation, nuclear capacity or strong interconnectors. Others face congestion that forces grid operators to delay connections or invest in reinforcement. A data centre may be economically attractive to a city because it brings construction, tax revenue and strategic infrastructure, while simultaneously competing with housing, industry or electrified transport for scarce network capacity.

The IMF’s recommendation to deepen energy-market integration and invest in cross-border grids therefore belongs at the centre of the AI debate. Compute cannot be scaled independently of electricity. Europe’s ability to convert artificial intelligence into productivity will depend partly on whether electrons can move across borders as efficiently as data is supposed to.

Renewables help, but geography still matters

Europe’s energy transition gives the AI build-out a potentially cleaner foundation than a purely fossil-fuelled expansion. Eurostat reported on September 18 that renewable sources produced 54.1% of EU electricity in the second quarter of 2026. Solar accounted for the largest share of renewable generation, followed by wind and hydropower. That is a substantial structural shift in the power system and offers technology companies a growing pool of low-carbon electricity.

Yet the continental average conceals enormous national differences. Eurostat said Latvia generated 97.7% of its electricity from renewables in the quarter, Denmark 94.3% and Croatia 92.2%, while Slovakia, Czechia and Malta recorded much lower shares. Data-centre investment is also concentrated in places chosen for connectivity, customers, cloud ecosystems and existing infrastructure, not simply where renewable generation is highest.

This mismatch creates a planning challenge. Building large computing clusters where grids are already constrained can increase the need for transmission, storage, flexible demand and new generation. Moving data centres toward regions with abundant electricity can reduce power pressure but may raise latency, connectivity or labour-market costs. The economic optimum is therefore not obvious.

There is also a timing problem. A data centre can often be designed and built faster than a major transmission line. If governments approve computing projects without parallel grid investment, bottlenecks can emerge before new capacity arrives. If they impose blanket restrictions, they risk pushing investment to other continents. The policy challenge is to coordinate permits, grid connections, generation and local planning so that digital infrastructure does not become another source of scarcity.

Europe’s renewable progress gives it a strategic opportunity: an AI sector powered increasingly by low-carbon electricity could be both an industrial asset and part of the energy transition. But that advantage will exist only if the grid can deliver the power where and when computing facilities require it.

The EU is spending heavily on computing sovereignty

Brussels has already responded to concerns about technological dependence with a more interventionist infrastructure policy. This month the EU launched a process to establish up to seven large AI “gigafactories” intended to provide advanced computing capacity for start-ups, scale-ups, researchers, public institutions and companies. The initiative envisages up to €10 billion in European and national public funding alongside at least €20 billion in private investment, for more than €30 billion in total financing.

The economic logic is straightforward. Access to advanced processors has become a strategic input comparable to laboratory capacity in biotechnology or fabrication plants in semiconductors. If European companies cannot secure enough computing power at competitive prices, they may be forced to build products on infrastructure controlled by foreign hyperscalers or move workloads to other regions. Publicly supported capacity is meant to lower that barrier and create a base on which domestic companies can experiment and scale.

The harder question is whether infrastructure alone creates competitiveness. Europe can build powerful machines without necessarily building globally dominant firms. The commercial ecosystem also needs model developers, cloud platforms, applications, specialised data, skilled engineers and customers willing to adopt new tools. A publicly funded compute cluster that is difficult to access, poorly connected to private capital or separated from strong research networks could become expensive capacity rather than an engine of productivity.

Supporters argue that this is precisely why the facilities are being designed as part of a wider network that includes existing AI factories and national innovation programmes. The objective is to create a ladder from research to deployment. Critics will still ask whether governments can allocate computing resources efficiently and whether subsidies risk protecting weak business models.

Those debates should not obscure the strategic shift. Europe no longer treats cloud and AI infrastructure as an ordinary private-sector utility. It increasingly sees compute as part of economic security, industrial policy and technological sovereignty. The IMF’s warning about foreign dependence reinforces that political direction even as it raises the bar for proving that public investment produces measurable economic returns.

Dependence on U.S. and Chinese technology is more than a trade issue

The IMF paper warned that Europe risks creating a new strategic dependency because the United States and China dominate the development of leading AI models and much of the surrounding technology stack. That dependency is broader than the nationality of a chatbot. It includes processors, cloud infrastructure, model weights, developer tools and the corporate platforms through which businesses increasingly consume AI services.

Reliance on foreign suppliers is not automatically harmful. International specialisation is a normal feature of an efficient economy, and Europe itself depends on exports of advanced machinery, pharmaceuticals, vehicles and other high-value products. Attempting to reproduce every layer of the AI supply chain domestically could waste capital and slow adoption.

The strategic concern arises when dependence is concentrated in a small number of suppliers and the technology becomes embedded in critical functions. A bank, hospital, manufacturer or government agency that builds processes around a proprietary foreign model may face switching costs, data-governance questions and exposure to decisions made outside European jurisdiction. Export controls on advanced chips have already demonstrated how geopolitical policy can affect access to computing hardware.

Europe’s response is therefore likely to combine openness with selective capacity building. The goal is not necessarily technological autarky. It is to ensure that European companies have credible alternatives, that critical infrastructure can operate under European rules and that the continent retains bargaining power in a market dominated by a few global firms.

This is also an economic question about where value accrues. If European companies become heavy users of AI but most spending flows to foreign cloud providers and model developers, productivity may rise while a large share of profits and intellectual property accumulates abroad. Domestic applications can still create value, but the structure of the supply chain will determine how much of the AI dividend remains inside Europe.

Regulation can protect trust — and still carry an economic cost

Europe’s AI policy is inseparable from regulation. The bloc has sought to establish rules around safety, transparency, data protection and the use of high-risk systems while other jurisdictions have moved more cautiously. Supporters argue that clear rules can create trust and prevent harmful deployment. Businesses may be more willing to use AI if they know liability, governance and data standards. Consumers and workers may also accept the technology more readily when safeguards are credible.

The economic risk is that compliance costs fall unevenly. Large companies can hire legal teams and build extensive governance systems; small firms may struggle with the same fixed costs. Regulation can also reduce the number of tasks or sectors in which AI can be deployed, lowering potential productivity gains. Earlier IMF research estimated that European and national rules could reduce projected AI productivity benefits substantially under assumptions in which exposure falls in regulated tasks, occupations and sectors. That was a scenario analysis, not a measured loss already visible in the economy.

The more useful question is therefore not whether Europe should regulate AI, but how dynamically the rules can be calibrated as the technology changes. Requirements designed for today’s models may become outdated quickly. Rules that are too vague can deter investment because firms cannot price compliance risk; rules that are too rigid can freeze business practices before the technology has matured.

There is also a competitive dimension. If European companies face higher compliance costs than foreign rivals, they may scale more slowly. But if European standards become globally influential, firms that meet them early could gain an advantage in regulated industries. The history of data protection shows both possibilities: Europe can set global norms, but compliance can also impose significant costs.

The IMF’s message to finance ministers is essentially that regulation has macroeconomic consequences. Governance choices influence adoption, and adoption influences productivity. The task is to protect citizens without unintentionally preventing the diffusion of technologies that could raise living standards.

The divide could run between regions as much as between countries

National averages can hide a more granular geography of winners and losers. AI activity tends to cluster where universities, capital, skilled workers, data centres and large corporate customers are already concentrated. London, Paris, Munich, Amsterdam, Dublin, Stockholm and other established technology or financial hubs can attract further investment because each new company benefits from the ecosystem built by those already present.

That agglomeration effect can magnify regional inequality. A successful AI cluster raises demand for highly skilled labour, office space, power and housing. Local wages and tax revenues may rise, but so can rents and infrastructure pressure. Regions without such clusters may see fewer direct benefits even if companies there use AI tools purchased from elsewhere.

The EU has long used cohesion policy to counter geographic divergence, but digital technologies complicate the model because their most valuable assets are mobile and intangible. A semiconductor plant can be placed in a particular region with a subsidy. A software company can relocate key staff or intellectual property more easily. The policy response therefore has to focus on capabilities — universities, digital networks, energy supply, skills and finance — rather than simply attracting a single flagship investment.

Remote access to AI may spread some benefits more widely. A small manufacturer in a peripheral region can use cloud-based design, forecasting or translation tools without being located near a major technology hub. That is one reason the diffusion of applications matters more than the location of frontier-model developers alone.

Still, the IMF’s warning about uneven gains is credible because adoption itself depends on management quality, workforce skills and access to complementary capital. Regions where firms are smaller, less digitised or financially constrained may receive the technology later. Without targeted support for diffusion, AI could reinforce Europe’s existing core-periphery pattern rather than narrow it.

Productivity gains will be judged by investment and wages, not demonstrations

The next phase of Europe’s AI story will be measured less by the capabilities of individual models and more by conventional economic indicators. Are companies investing more? Are workers producing more per hour? Are real wages rising? Are new firms scaling? Are established businesses becoming more competitive in export markets? Are public services delivering better outcomes at lower cost? Those questions determine whether AI becomes a broad productivity technology or remains concentrated in a narrow technology sector.

Early corporate experimentation is easy to observe; economy-wide productivity is not. Firms may spend heavily for several years before reorganisation produces measurable gains. Historical general-purpose technologies often required complementary investment in infrastructure and management before they changed national productivity statistics. AI could follow a similar pattern.

That lag creates a political challenge. Governments are committing public money to computing infrastructure and training while voters are being asked to tolerate disruption in labour markets and local energy systems. If promised benefits remain abstract, support may weaken. Policymakers will need credible metrics that distinguish genuine productivity improvements from the simple growth of technology spending.

The same applies to companies. Investors increasingly expect AI strategies, but purchasing model access is not a strategy. Firms must show that tools reduce costs, raise revenue, shorten product-development cycles or improve quality. Some projects will fail. Others may generate value only after business processes are redesigned.

This is why the IMF’s restrained 1% estimate may be more useful than spectacular forecasts. It sets a benchmark that is economically meaningful without assuming instant transformation. If Europe can capture even that gain while broadening investment and wages, it would matter. If adoption remains concentrated in a few firms and regions, the headline number will conceal a much less successful transition.

A modest dividend could still be strategically large

Europe’s growth problem makes small productivity improvements valuable. Ageing populations limit the expansion of the workforce, while high public debt in several large economies constrains fiscal stimulus. In that environment, lasting gains in output per worker are one of the few ways to raise living standards without relying on ever greater borrowing or population growth.

Artificial intelligence will not solve those structural problems by itself. The IMF’s estimate is a reminder of the scale: around 1% over five years is meaningful, but it is not enough to erase Europe’s investment gap, demographic pressures or energy disadvantages. The technology becomes more important because it interacts with those challenges. Faster research could improve industrial competitiveness. Better automation could offset labour shortages. More efficient public administration could reduce fiscal pressure. But each benefit depends on complementary reforms and investment.

The costs are equally interconnected. If data-centre demand worsens power scarcity, AI could raise costs for households and industry. If automation concentrates income among capital owners and scarce specialists, it could intensify political resistance. If European firms depend entirely on foreign infrastructure, productivity gains could coexist with a larger strategic trade deficit in digital services. If regulation fragments the market, the continent could pay for safety without capturing scale.

That is why the Dublin discussion matters beyond the technology sector. It places AI inside the core economic choices facing Europe: integration versus fragmentation, investment versus fiscal restraint, openness versus strategic autonomy, and innovation versus social protection.

The most credible outcome is neither a sudden productivity miracle nor mass technological unemployment. It is a gradual, uneven transformation in which some firms and countries move quickly, others lag, and policy determines how widely the gains spread. Europe has enough capital, talent and industrial depth to benefit substantially. The question is whether its institutions can move at the speed required to turn technical capability into economic performance.

The next test is implementation

For now, the IMF paper is a warning and a map rather than a verdict. Its numbers describe potential outcomes based on assumptions about technology, adoption and policy. The actual European AI dividend will be determined by decisions that are much more concrete: how fast a data centre can connect to the grid, whether a start-up can raise growth capital, whether a manufacturer can deploy software across multiple countries, whether a worker can retrain, and whether a company can comply with regulation without abandoning a useful project.

Those decisions will accumulate. A delayed grid connection in Dublin, a financing round that moves a company to California, a procurement rule that blocks a cross-border supplier or a training programme that successfully shifts workers into higher-productivity roles may appear local. Together they shape the continental outcome the IMF is trying to estimate.

The timing is important because Europe is already committing large sums to AI infrastructure. The Commission wants more data-centre capacity and up to seven new gigafactories; member states are competing for investment; companies are integrating generative systems into daily work. The policy window is therefore not theoretical. The architecture being built now will influence costs and competitive positions for years.

Finance ministers left Dublin with no single lever capable of delivering the promised productivity gain. They instead face a portfolio of less glamorous tasks: integrate markets, mobilise private capital, expand grids, improve skills, keep regulation adaptable and manage the distributional consequences of automation. None offers the instant visibility of a new model launch. All may prove more important to Europe’s prosperity.

The IMF’s central message is ultimately one of conditional optimism. AI can lift European productivity, but technology alone will not deliver the dividend. Europe must build the economic system around it — and do so without allowing the pursuit of digital competitiveness to create new fractures in its labour markets, energy networks or strategic autonomy.