Abstract

AI does not generate a distinct, separable tax base. Returns arising from AI, which may be reduced by deductions, tax credits and refundable tax credits, are typically combined with returns on software, proprietary data, organizational capital, intellectual property, market power and complementary investment. A narrowly defined AI tax is therefore administratively costly and likely to discourage productive adoption. This article examines how the European Union can capture economic rents and profits arising from AI without taxing its use. It distinguishes ordinary returns from economic rents, considering the various uses of Pillar One, Pillar Two, digital services taxes, capital-income taxation, cash flow and allowance-for-corporate-equity arrangements, excess-profit taxation, rebalancing of business labor taxes and public participation in publicly financed research and development. These instruments differ in how they identify, allocate and tax mobile AI-related returns between Europe and other jurisdictions. The contribution suggests here that Pillar Two's minimum-rate floor reduces the incentive to reposition mobile profits elsewhere; but cannot allocate residual profit to EU customer and user markets, that Pillar One can play a role here but is not yet operational and that turnover-based digital platforms taxes can only serve as an imperfect interim proxy. The best course of action seems to consist of a modular and layered set of actions: pursuit of enforceable minimum tax, further multilateral negotiations on market- sharing, strengthening of taxation on realized capital income, examining broader rent-sensitive corporate tax reforms and mitigating the excessive burden of taxing too much of workers' labor income. Realized rents and shifted profits should be taxed instead of the algorithm itself.

1. Introduction - Why Even a One Percent Revenue Loss Matters

The most secure starting point is not a thought experiment about all possible mass technological unemployment but a simple example of fiscal mathematics. Eurostat reports that taxes and net social contributions in the EU amounted to €7.281 trillion in 2024.^[1] Total EU general-government revenue was approximately €8.29 trillion, so a one percent decline would have reduced annual receipts by about €82.9 billion.^[2] This is not a civilization-endangering number; quite the opposite, it is unquestionably a fiscally significant loss. What's more, it is not an abstracted European aggregate. Eurostat data show 2024 tax-to-GDP ratios in the article’s focus countries, namely France, Italy and Belgium. They each registered in 2024 an intake of over 42 percent of their gross domestic product.^[3] In economies of such a mass, even minuscule erosions of all-encompassing tax bases turn into budgetary consequences rather than rounding errors. It is the compositional, rather than AI, erosion of tax bases that makes the issue relevant for AI policy. This compositional concern is based on the fact that, according to the European Commission's latest data on the composition of the tax mix, in 2024, the most important sources of taxes in the EU-27 were labor taxes and social contributions, 51.5 percent of the total, followed by consumption taxes at 26.8 percent and capital taxes at 21.6 percent.^[4]

Europe’s heavy reliance on labor and payroll taxation creates fiscal exposure if the composition of part of the value added shifts gradually and unevenly to profits, capital income, royalties, high mark-ups in intangible-intensive sectors and returns to capital ownership and capital gains, income that is more concentrated and more internationally mobile than wages. This does not mean disaster, but just that possibly even moderate compositional shifts might have an influence of fundamental importance for the public finance outcomes, given that European taxation of labor is heavily weighted and that of mobile capital in an environment of much less generous institutional restrictions. There is an equally compelling reason to avoid a naive response, however. A recent firm-level investigation by the European Central Bank demonstrates that while AI technology adoption is expanding, deep integration is only prevalent among a small minority: more than 70 percent of firms surveyed in the euro area reported usage of AI technology in 2025 and only 7 percent reported intensive AI use.^[5] Furthermore, intensive users tend to cluster within high-tech sectors, younger firms and knowledge-intensive sectors^[6] and the benefits from AI derive in significant measure not only from initial adoption but from the integration of AI into core processes, the investment in complementary assets and equipment and organizational overhaul. From a broader perspective, the OECD arrives at a similar bottom line: AI exhibits the qualities of a general-purpose technology; however, its productivity effects are uncertain, varied and reliant upon supporting assets and market configurations.^[7] The intensive AI research and development is concentrated among a handful of leading technology firms and the diffusion of this technology remains uneven across industries. These factors are pertinent to the design of tax policies because they call into question the underlying assumption that it will be easy to make AI profit fully transparent and create a smooth, uniform and administratively straightforward tax base.

That distinction is critical. It has been particularly hard in principle (not just in practice) to establish a distinction between profit generated by AI and profit generated more generally by software, proprietary data, brands, organizational capital, scale, high fixed-cost and low marginal cost businesses, regulation, or simply good managerial skills. The OECD's analysis of MNEs and intangible capital shows that the foreign affiliates of MNEs generate proportionately more income with intangible assets and that that income is more concentrated in ICT, finance, business services, telecommunications and other value chains that are highly intangible.^[8] Much of what a public debate would loosely call AI profit is therefore better viewed as a menu of gains from intangible capital and the organisation of firms, of which AI is one input among several rather than the sole source; field evidence on generative AI points in the same direction: larger productivity gains often require a reorganisation of processes and task reallocation, which means these gains accrue with the tool, the firm and its investments in complementary factors, rather than strictly to the algorithm.^[9] Once this is acknowledged, the road to a more sensible core tax question becomes clearer. It is no longer whether Europe should invent a narrow tax whenever a firm utilizes AI. Instead, the issue becomes whether its current and planned tax instruments can effectively target the tax bases that AI is likely to increase or redefine: standard corporate income, economic rents, residual income attributable to sources of value that may be separately taxed (high-value intangibles), income that has shifted offshore into low-tax jurisdictions, capital gains accruing to founders and investors and value created from customers or end users residing in countries without significant physical presence.

This is also why drawing a line between ordinary return and rent is crucial. If AI makes existing standard corporate activities more productive by bolstering mundane decision-making and administration with AI systems, the additional return from the resulting profit-generating enterprise should, in principle, be no different from returns that result from other types of successful digital-oriented investment. If, in contrast, AI extends market power, data advantages, first-mover scale, control over frontier models, or its sustained advantage in proprietary investment, then some portion of the resulting income may be economic rent, which is potentially a more appropriate object for targeted taxation. Hence, the resulting argument is cautious rather than alarmist. Europe should be cautious in rushing to implement a narrow tax merely because the tax label is AI: such a tax will often be poorly aligned with the underlying economics and will tend to penalize productive adoption of a general-purpose technology too much. But it does not follow that abandoning a narrow AI tax justifies complacency. What it would justify is a different approach. It would justify something that prioritized much stricter anti-shifting rules, more credible minimum taxes, market-jurisdiction allocation when appropriate, broader and more principled rent taxes, suitable capture of capital gains and distributions and a judicious repositioning of tax resources away from over-reliance on labor. The key principle would not be a blanket avoidance of algorithm-specific taxes, but a more careful decomposition of where rents are generated, whether a given jurisdiction has the best case to tax those rents and how to do so in a way that causes the least damage to productive activity and innovative commercialization.

2. Current Approach: What Europe’s Existing Tax Framework Can and Cannot Capture

Today’s European and international solutions build on a real flaw of orthodox international taxation: standard corporate-tax apportionment assumes significant taxing rights follow materially substantial physical presence and that profit can be apportioned by using legal entities, functions, assets and risks.^[10] The digitalization of the economy undermined those priors some time ago, even before the current wave of AI adoption. Large multinational firms can generate enormous revenue from a market without having any factory, shop, or employee there, by locating ownership of code, intellectual property and group financing elsewhere. Those were precisely the concerns addressed by the European Commission's proposals in 2018. One was to define a corporate-tax nexus on the basis of a significant digital presence, in essence, a digital permanent establishment.^[11] The other would levy an interim digital services tax over revenues derived from online advertising, intermediation and user data, aimed at groups earning above €750 million in worldwide revenue and at €50 million in taxable EU digital-services revenue.^[12] Neither succeeded: they both failed in a way due to the absence of unanimity, as many Member States favored instead reaching a solution through the OECD. The conceptual notion that an emerging 2018 turn makes sense remains apparent. If a firm is able to derive value from the monetizing of consumers, users and demand in Europe, without a real physical presence, then market jurisdictions have a substantive argument to claim portions of the tax base. This claim is further reinforced if user participation, network effects and data feedback loops form part of the value creation process. And such claims are becoming increasingly relevant as services become more deliverable by digital means and, where applicable, automatable.

2.1 Pillar One and Market-Jurisdiction Allocation

A recent analysis by the European Parliament looking at a potential EU digital levy finds that by 2023 digitally deliverable services accounted for 63 percent of EU cross-border services imports, while automated digital services accounted for 33 percent, rising to 39 percent when relevant IPR payments were included.^[13] It also argues that AI is likely to broaden substantially the range of activities that can be produced digitally and supplied from the cloud.^[14] AI therefore intensifies the existing cross-border provider-market problem. Nowhere is the problem of market-jurisdiction EU-wide corporate income tax and VAT/GST attribution and actual collection, in the context of cross-border provision, more profound. The OECD/G20 Inclusive Framework's Amount A under Pillar One was the solution to exactly that: a limited reallocation of residual profit to eligible market jurisdictions.^[15] The OECD factsheets state that Amount A would be implemented just to the biggest, most-profitable multinational groups: those with adjusted revenues above €20 billion and pre-tax profit margin above 10 percent, with segmentation rules in a few cases where the Group overall is out of scope.^[16] Of the profit exceeding that 10 percent level, 25percent of it would be apportioned to eligible market jurisdictions, as defined in the relevant paragraph of the OECD Implementation Framework text, according to sourced revenues.^[17] The general market-revenue nexus would be €1 million, reduced to €250,000 for jurisdictions with GDP below €40 billion.^[18] So the principle is not sectoral; it is a residual-profit-shifting redistribution to markets on broad economic indicator criteria.

Amount A is therefore attractive conceptually for Europe on at least three points: First, it allows the recognition that certain customer/user markets may be value creating without conventional physical presence; secondly, it attempts to tax residual profit rather than routine return and is therefore more conceptually aligned with rent allocation than to normal-source taxation; and thirdly, it provides a multilateral pathway for the phased elimination and standstill of digital services taxes and other similar measures, thus avoiding a proliferation of overlapping national instruments. The OECD factsheets also feature the Distribution and Marketing profits Safe Harbour expressly designed to mitigate double counting of profits where existing rules have already allocated significant profit to the market jurisdiction.^[19] In principle, this renders Amount A preferable to unilateral turnover taxes, since it is profit- and threshold-based and established with the aim of avoiding double taxation, rather than layered on top of existing arrangements without the benefit of that integration. Practically, therefore, the main constraint is the current legal status of Pillar One. The OECD's current Amount A webpage indicates that the wording of the multilateral convention issued in October 2023 has not yet been opened for signature.^[20] This webpage also confirms that the wording embodies the consensus so far reached, though certain issues remain unresolved; footnotes make these clear. The January 2025 co-chair update reported that the agreed text of the MLC had been submitted to the Inclusive Framework for adoption and clarified that adoption of the final text itself did not create any obligation to sign; the decision was to be made separately, according to domestic procedures.^[21] The update explained that‚ since the adoption of the final MLC, discussions in the Inclusive Framework were focused on outstanding issues regarding the Amount B framework. Amount A is neither legally in force nor operational in 2026. By July 2026, therefore, Europe cannot treat Amount A as being implemented or effective.

It is a negotiated but unrealized architecture. This remains an open issue and has kept digital services taxation and digital surcharges on the agenda within Europe. The EU's own resource based on a digital services tax was discussed in a 2026 briefing published by the EU Parliament, which stated that implementation of Pillar One had not materialized and that the absence of a global solution increases pressure for EU action. The briefing also confirms that France, Italy, Spain and Austria are already using domestic digital services taxes or similar levies and that these have generated predictable but limited revenues. In 2024, these states raised, for example, €756 million, €455 million, €375 million and €123 million, respectively. ^[22]

While this is not a fiscally transformative amount, it does tell us that gross-basis digital taxes are administratively plausible. At the same time, it encapsulates the key trade-off involved: a gross-basis DST will raise revenue relatively quickly, but it will do so through a fractured framework of thresholds and reporting duties that threaten to distort the rules of the Single Market. The wider digital-levy debate also illuminates the appeal and risks of unilateral action. The second European Parliament briefing on the 2026 digital-services-tax estimates that a DST limited to the narrow scope of the 2018 Commission proposals would cover about 10 percent of EU cross-border services imports and would yield about €7 billion every year. A broader scope, similar in coverage to the scope of DSTs already adopted in Member States, would cover about 19 percent of cross-border services imports and be expected to raise about €13 billion per year.^[23] An even broader tax, covering automated digital services, including those enabled by AI and from the cloud, including those in cloud-based delivery models, would normally be expected to yield even higher revenue than the 6 percent of the presumably appropriate tax base estimated for the narrower spectrum of digital services that is possible to measure. Nonetheless, this material makes the administrative difficulties clear.

Once the base is broadened enough to approximate where AI-enabled value is really being captured, then what the measure most closely resembles is a broad tax on cross-border digital services, including intellectual property, not a tax on a narrow class of digital platforms. This may be more economically defensible than a purely AI levy, but it is also more politically and legally ambitious and closer to an import tax than the traditional concept of a rent tax.

2.2 Pillar Two and the Minimum-Tax Floor

Pillar Two addresses a different issue. The EU's Minimum Corporate Taxation Directive applies the OECD's 15 percent minimum tax for multinational enterprise groups and large-scale domestic groups generating more than €750 million in revenues.^[24] The European Commission's current guidance page notes that when the effective rate in the jurisdiction is less than 15 percent, a top-up tax will apply through the Income Inclusion Rule, the Undertaxed Profits Rule, or a qualified domestic minimum top-up tax.^[25] The European Commission also explains the policy motivation for the substance-based income exclusion it adopts: to ensure that a fixed level of income is ignored for tax purposes where there is a substantive level of payroll and tangible assets, thereby moving away from those excess profits that result from income shifting strategies and that tend to be rich in intangible assets that are at greater risk for tax planning.^[26] As such, the Pillar Two plan is not a market jurisdiction allocation system. It is an anti-base-erosion floor and, at best, a simple anti-rent-avoidance proxy for highly mobile, low-taxed profit. The OECD's quantitative analysis provides a possible explanation for why Pillar Two is relevant even for the profit shifting associated with AI. It projects that the global minimum tax would cut global shifted profits in half, would cut the share of global multinational profits taxed below the 15 percent minimum tax rate by more than two-thirds and would generate an extra USD 155 billion-192 billion per year in global corporate income tax revenue.^[27]

A report by the European Commission Joint Research Center looks at EU Member States and finds that in the short term, CIT revenue in the EU would increase by approximately 7.1 percent or around €26 billion per year) if all EU countries adopted the minimum-tax rules.^[28] These are important, but the same studies also show the flaw. Pillar Two reduces the return from shifting mobile profit and intellectual-property income into low-tax countries. It does not, on its own, give market countries a higher share of residual profit just because there are customers or users there. So the two pillars are complements, not substitutes: one deals with the allocation; the other deals with the undertaxation and profit shifting. Thus, the argument presented here is that present arrangements are complex and incomplete and a clear indication that the present arrangements are only the beginning of the story: Europe has at present an effective Pillar Two framework. It has a powerful anti-shifting mechanism. Nevertheless, it lacks an effective, treaty-based system of residual profit allocation to markets and it does not possess an agreed-upon, workable common position on the utility of a purely temporary turnover tax in places where Pillar One remains frozen. While some cases can be made for the lack of market jurisdiction in a developing world and if the case is more weak elsewhere, the issues of primacy of source-based taxation, the difficulty of treaty revision and the associated technical complexities, the potential problems of overlap with unilateral measures and the continuing uncertainty surrounding the exact contents and legal effects of the multilateral convention argue that the present arrangements are, at least, incomplete: they impose taxes on some mobile AI-related profits to a greater extent than previously, but are uncertain as to where they will share further those profits amid inherently high user-based source-determined activity.

3. Alternative Taxation: Taxing AI-Related Profits and Rents Without Taxing AI Use

A robust examination of alternatives must first exclude at least one tempting but unfounded idea: an AI-specific tax base. There is no commonly adopted accounting or legal measure that can, for tax purposes, distinguish AI's contribution to profit from that of software engineering, specialist data, cloud infrastructure, brand capital, distribution channels, organizational design choices, or the strategic advantages that accrue as a result of other investments in scale. The OECD explains how the economic impacts of AI very much depend on pervasive complementary innovation, diffusion and market structure.^[29] Existing research at the firm-level seems to agree: strong value is only realized if AI is embedded in a business through large-scale projects and restructuring. For multinationals, returns on intangible capital are also already inextricably spread across affiliated firms, different tax jurisdictions and various asset categories. The most plausible taxable consequences of the AI transformation, then, are not a profit account for AI in particular, but such things as increased gross margins, royalty streams, high residual profit shares, founder exits, dividend payouts and share prices that anticipate future rents.

3.1 Pillar Two as a Minimum-Tax Floor

The first and strongest alternative instrument, therefore, is not a new AI tax but a more effective implementation of Pillar Two. That has two advantages; it rests on a tax base that exists in law and it directly confronts one of the main weaknesses of AI-intensive business models: the ability to position highly mobile, lucrative intangible incomes in low-taxation jurisdictions. The EU has moved ahead of many peers by transposing the minimum-tax directive for fiscal years beginning in 2024. Just as the 2026 Commission-supported manual for 14 participating Member States demonstrates, considerable variations therefore persist within a shared legislative framework in registration, filing and payment rules, designated entities and centralized compliance.^[30] Better implementation, therefore, not only entails simply sustaining the directive but also improving administrative coordination, information sharing, safe-harbor certainty and domestic capacity so that the floor bites effectively on mobile intangible profits.

Making an argument for such a strengthening is both conceptual and empirical. OECD estimates mean that the global minimum tax substantially reduces profit shifting benefits; the Commission's JRC study found that even taking account of long-term behavioral impacts, the direct tax effects for the EU were sizeable. This is especially relevant where AI-related rents are not such as to flow through to findings of rent in property rental markets, for example, but where these rents are captured in intellectual-property income, other forms of mobile, hard-to-value profit. Furthermore, Pillar Two’s structure already reflects a normative judgment of a sort that is key to this paper's core distinction: the substance-based exemption ensures coverage of a formulaic return linked to tangible assets and payroll, while more residual, intangible-intensive rents are left exposed to the top-up tax. It is far from a perfect approach to distinguish mobile residual income from normal returns and unintended consequences can be avoided simply by moving in that direction. But Pillar Two cannot shoulder the entire policy burden. It does not instruct Europe on which jurisdiction imposes tax on global profit generated by final consumption or user markets, or what have you. It restricts the advantage of declaring profit in a low-tax jurisdiction. Finally, it is not beyond the scope of avoidance via credit-mix, subsidy, classification choice, or under-enforcement induced by internal complexity. To conclude, it is the best tool for AI as a protection of the floor. It can sort out part of the globally mobile profit that drifts without tax; it cannot allocate the resulting tax to the Irish, French, German, Italian, or non-European parent jurisdiction that obtains the revenue from the customer-generating activity.

3.2 Economic-Rent and Excess-Profit Taxation

The second, more serious, family of alternatives is economic rent or excess-profit taxation. The case is supported by the work of Hebous and Mengistu, who define economic rent as returns in excess of opportunity costs and argue that well-designed rent taxes need not be distortionary, thus providing an efficiency argument for rent taxation: the inverse of what we have seen so far.^[31] The IMF identifies two broad operational models for excess profit taxes; one is cash-flow taxation, which effectively affords up-front expensing of investment, thus exempting the normal returns from economic rents while taxing away the above-normal return.^[32] Second, it is a system that envisages an allowance for the normal return through an allowance for corporate equity.^[33] This resembles advanced conventional accounting practices: a notional deduction consistent with what interest payments on debt provide. Instructively, Hebous, Prihardini and Vernon reach a similar conclusion and show that, at least theoretically, excess-profit taxes can be targeted exclusively on those rents thus understood, with the resultant policy remaining non-distortionary. In theory, then, taxing rents and not AI-enabled profit is the superior, elegant countermeasure to the risk of discouraging marginal investment. The problem is that this neat conceptual scheme of rent taxation collides with the administrative disarray of AI production. To even know what an AI rent might be is extremely difficult. Value created by AI is often imprecise, wrapped up with off-the-shelf software, embedded in consulting contracts, subsidized through platform ecosystems, shielded by patent structures, or achieved only as a means to inflated equity prices.

Even separating a normal return from rent within a single firm proves challenging; calculating a firm’s normal cost of capital, accounting for losses and carry-forwards, valuing intangibles under fair-value accounting and establishing an objectively credible method to apportion a corporate group's rent across many different countries all are administratively difficult. In various cross-border contexts, all those difficulties are compounded by group structures and transfer pricing, the strategic place of ownership over the highest-value intangibles. Now that AI tends to foster scale increases and intangible intensification, perhaps the rent values it is easiest to grab will be the most finely-grained and hardest-to-measure. These constraints lead to two conclusions. First, if Europe intends to move towards a system of rent taxation at all, it will need to do so by implementing broad reforms to its corporate tax base rather than by measures targeting an AI rent tax in particular. Extending a safe harbour for normal returns, or proceeding to ad hoc cash-flow reforms would at best generate a tax on AI rents where they exist, but would also capture rents from other activity associated with high levels of intangible investment or distinctiveness, which would arguably be a benefit if the standard engendered by the normative goal in the first place is to attribute tax to rents regardless of the technology that underpins them.

Second, any rent-based tool will still depend on a territorial rule where capital markets operate. A rent tax limited to the territorial sphere of legal taxable events of a corporate entity, or in which an intangible is constituted or is a former company property, will not prevent the market complementarity problem. A rent tax based on destination or formula will provide greater stability against shifting, but will require the implementation of a system akin to formulary apportionment or destination-based cash-flow tax, rather than incremental reform. For Europe, this may be a long-run evolution rather than an immediately deployable tool. A supplemental excess-profit tax is a compromise, somewhere between the ideal and the practical. As the IMF review notes, excess-profit taxes have often been introduced in wartime and during other episodes of acute windfall gains and in such circumstances, a time-limited levy on large profits remains defensible even when baseline calculation is necessarily incomplete.^[34] But that analogy does not transfer well to AI. It is not a once-and-for-all energy-supply shock and the AI industry is not an established tax domain. An ongoing, across-the-economy excess-profit charge designed specifically for AI would be hard to justify. A short-term, wide-angle supplemental excess-profit levy might still make sense if Europe encountered a situation of readily identifiable and concentrated rents in a few fiercely supported or exceptionally protected sectors. But this is a final-resort mechanism, not the core framework to regulate AI rents.

3.3 Capital-Income and Digital-Services Taxation as Complementary Instruments

Capital gains and dividend taxation are less directly targeted than an effective rent tax, but for some areas, they are closer to the point at which AI-related gains are realized. Much of the economic upside of successful AI deployment will not end up as continuously taxable profit in Europe: instead, it may be capitalized into share values, founders’ wealth, venture-stage exits, acquisitions of start-ups, or distributed to shareholders in high-income households. OECD work on capital gains has found that most OECD countries tax capital gains on realization, frequently at lower rates or with exemptions in comparison to other income, partly on grounds of dynamism, though there is inconclusive evidence that these implementations significantly influence growth.^[35] For Europe, then, this suggests a pragmatic complement rather than a successor. Broad-based and reasonably integrated taxation of capital gains and dividends may allow a share of large potential windfalls of AI to be captured without directly taxing companies that use such AI. However, whether as a consequence of unrealized gains remaining unrealized, individuals or firms leaving, large gross gains being unrealized for a long period of time, or corporations seriously considering exit-tax regimes, it is obvious that this is a less than perfect measure, with obvious mobility and innovation-related drawbacks for founders, entrepreneurs and high-net-worth, high-mobility households. A reasoned plan for capital-income taxation should therefore follow an eminently cautious approach. Europe has legitimate reasons to be cautious about reducing unwarranted differences between taxing labor income and taxing realized capital gains or dividends, particularly where entrepreneurial risk increasingly clearly manifests itself as returns to scalable intangibles. It also has reason to defend doing so by claiming a significant tax benefit where the countries could raise an AI-specific capital-gains tax by precluding any leakage into exit jurisdictions. Yet assigning all AI-related capital gains to a separate tax would reintroduce the main flaw of a narrow AI profit tax. It would blend innovation rents and regular entrepreneurship risk and encourage residence-based tax planning.

Europe would better aim at improving general capital-income taxation and its mitigation than at applying an AI label to it. Turnover taxes and digital levies remain the most tempting unilateral option because they combat one genuine problem head-on: value created and realized in European markets without a tangible taxable footprint. The 2026 European Parliament briefings demonstrate why they remain a politically credible option. Revenues are highly visible; administration is comparatively straightforward and profit-shifting matters less because the dominant base is turnover. Country experience also indicates that revenues have been more resilient than many detractors initially contended. The same 2026 briefings also show why they are a poor long-term basis for taxing AI-related profits. They fail to differentiate between profit streams and rents. They risk taxing low-margin, high-growth firms at the same rate as high-margin market incumbents. They may shift costs onto merchants, online advertisers and end-consumers. They threaten Single Market integration if transposed to their likely multiplicity of national configurations. A more comprehensive levy on automated digital services might be slightly more rational than a narrow Digital Services Tax, but it remains intrinsically a gross-basis tax that by design can only roughly estimate rent capture. That said, the appropriate treatment of DSTs is therefore a matter of prudence. They are not suitable tools for establishing or taxing AI profit. However, they are plausible interim measures to tax market-based value in the absence of treaty-based residual-profit sharing, should national governments be unwilling to wait indefinitely for the aforesaid Pillar One arrangement.^[36] If adopted at the EU level, the most approachable form ought to be EU-wide rather than national, specifically temporary as opposed to perpetual, designed to replace rather than accumulate on top of regional DSTs, in addition to a sunset or modification clause in case a worldwide market-sharing framework eventually functions. But even then, it should be promoted as a digital turnover substitute for market access, rather than a levy on AI itself.

3.4 Labor-Tax Rebalancing and Public Participation

Another set of considerations pertains to the overall fiscal composition. The Commission's own figures demonstrate that Europe remains atypically dependent on the collection of taxes on labor. Such dependence has a particular policy risk: that if AI increases productivity but dampens the growth of the labor-tax base, governments may attempt to offset the effect via increased charges akin to the labor-related burden on firms at the technological frontier that are subject to AI or on labor-saving technologies themselves. This would be unwise. The Commission's own work on taxes repeatedly contends that reallocating more of the burden away from labor and onto environmental taxes and recurrent immovable-property levies can be done fairly and efficiently and its 2024 Annual Report on Taxation found that perhaps, to forestall falling revenues, Europe may need to transition away from labor towards capital and consumption taxes in the medium term.^[37] The key point for the present discussion is thus a cautionary one: Europe should not follow the rest of the world in subjecting itself to a quasi-robot tax that increases the marginal cost of adopting AI. A much less ambitious rebalancing would be more modest and more defensible. It would maintain incentives for an efficient use of AI by mitigating, at the margin particularly where labor-tax wedges are high, the tax penalty on formal employment and routine labor at the margin and only partially fund this mitigation by broadening, in a way less sensitive to adoption, less mobile domestic tax bases, including recurrent property taxes, environmental taxes, enlarged and realised capital-income collection, improved policing of base erosion.

It would remove the built-in, or at least embedded, labor tax cost to AI and, more generally, to more successful intangible deployment, without giving up on the gains characterized by consumer surplus and quality improvements, which, because they do not constitute realized taxable income, should be left untaxed. Finally, there should be modest but serious consideration of public participation in the upside. Where public funds materially reduce the risks involved in follow-on commercialization, ex post taxation is not the only means of securing fiscal capture. The European Innovation Council already stands as an institutional precedent, if one that remains to be refined; its EIC Fund functions as a venture-capital arm, capitalized at more than €4 billion, while EIC Accelerator funding can be combined with substantial equity investment.^[38] A financial stake in this narrower area than general taxation should involve the granting of warrants, royalty-like repayment clauses, convertible instruments, or minority equity holdings, so that direct taxpayers benefit from commercial success rather than socializing losses. This principle is most tenable in cases where support is concentrated, additive, risky and targeted at the development of critical and autonomous AI-related capacities; least tenable where it is applied on an indiscriminate basis, crowds out private resources, or is distorted by political selection. Public upside participation can be an adjunct to taxation; it cannot be an alternative to a rationally designed tax system.

Taken together, these considerations argue for a layered European policy approach. The first layer should be the full, credible administration of Pillar Two, which is the particular solution to the fact that anti-shifting is a pressing and otherwise solvable problem. The second layer should be the persistent effort at a multilateral market-allocation mechanism under Pillar One, without any pretense that its data, much less tax rules, are operational. The third might be the search for broader rent-tax principles, perhaps allowances for normal returns, cash flow, or carefully delimited excess-profit components without AI as a supposed separate object of tax. The fourth should be the targeted bolstering of capital-gains, dividend and exit tax regimes where households are actually experiencing the AI rents. The fifth should be a redefinition of the tax mix, balancing dependence on labor with income from capital and wealth. Once those five layers are in place, the eye-catching idea of a temporally coordinated but geographically limited and administratively manageable EU-wide DST, as something that might initially pick up some of the slack of a below-cost AI tax, might be permissible, if not entirely inevitable. What is obvious is that tax bases affected by AI should be approached through the instrument best matched to that particular phenomenon (profit-shifting, market access, rent, realized wealth, or excessive dependence on labor), rather than through the rhetorically appealing but administratively fragile AI-specific tax.

Conclusion - Tax the Rent, Not the Algorithm

Europe should avoid a narrowly defined tax triggered solely by the use of AI. It is impossible to distinguish AI-related gains from other gains in software, data, brands, organizational capital and complementary investment, nor can it distinguish the label “AI profit” from normal returns or economic rents. The more defensible policy principle is thus to tax the rent, not the algorithm, while treating identifiable local infrastructure and environmental costs separately from internationally mobile profits. No single instrument provides a complete solution. Pillar Two can limit the benefit of locating low-taxed profits in favorable jurisdictions, but it cannot allocate residual profit to European customer or user markets. Pillar One is designed to do so and Amount A is not yet operational. Broader reforms based on the principle of cash-flow taxation or allowances for normal returns are attractive in principle but difficult to administer globally. Capital gains and dividend taxation can capture some realized gains, while a temporary EU-wide digital-services tax may provide a short-term market-based proxy, but turnover taxation can disadvantage low-margin firms and be passed through to consumers. Thus, the most credible European policy requires a layered reform package: enforce the minimum-tax floor, continue multilateral market-allocation negotiations, strengthen taxation of realized capital income, examine broader rent-sensitive corporate-tax reforms and reduce excessive reliance on labor taxation without discouraging productive AI adoption.

This article was prepared as an independent research contribution following the conference Inequalities in Longevity, held at Fondazione Giorgio Cini in Venice on 3–4 July 2026. It develops themes presented and discussed at the conference but is separate from the official conference proceedings.

Unless expressly stated otherwise, this publication has not been commissioned or endorsed by Fondazione Giorgio Cini. The analysis, interpretations, and conclusions are those of the author(s) and do not necessarily reflect the official positions of Fondazione Giorgio Cini, Swiss Institute of Artificial Intelligence (SIAI), or their respective affiliates.

References

[1, 3] Eurostat (2025) EU and euro area tax-to-GDP ratio up in 2024. Luxembourg: Eurostat.

[2] Eurostat (2025) Euro area government deficit at 3.1% and EU at 3.2% of GDP. Luxembourg: Eurostat.

[4] European Commission (2026) Data on Taxation Trends. Brussels: Directorate-General for Taxation and Customs Union.

[5] European Central Bank (2026) Survey on the Access to Finance of Enterprises in the euro area: Fourth quarter of 2025. Frankfurt am Main: European Central Bank.

[6] Chaloupka, D., Lalinský, T. and Lopez-Garcia, P. (2026) What separates firms that use AI intensively from firms that don’t? Frankfurt am Main: European Central Bank.

[7, 29] Filippucci, F., Gal, P., Jona-Lasinio, C., Leandro, A. and Nicoletti, G. (2024) ‘The impact of artificial intelligence on productivity, distribution and growth: Key mechanisms, initial evidence and policy challenges’, OECD Artificial Intelligence Papers, No. 15. Paris: OECD Publishing.

[8] Cadestin, C., Jaax, A., Miroudot, S. and Zürcher, C. (2021) ‘Multinational enterprises and intangible capital’, OECD Science, Technology and Industry Policy Papers, No. 118. Paris: OECD Publishing.

[9] Calvino, F. and Fontanelli, L. (2023) ‘A portrait of AI adopters across countries: Firm characteristics, assets’ complementarities and productivity’, OECD Science, Technology and Industry Working Papers, No. 2023/02. Paris: OECD Publishing.

[10] OECD (2022) OECD Transfer Pricing Guidelines for Multinational Enterprises and Tax Administrations 2022. Paris: OECD Publishing.

[11] European Commission (2018) Proposal for a Council Directive laying down rules relating to the corporate taxation of a significant digital presence, COM(2018) 147 final. Brussels: European Commission.

[12] European Commission (2018) Proposal for a Council Directive on the common system of a digital services tax on revenues resulting from the provision of certain digital services, COM(2018) 148 final. Brussels: European Commission.

[13, 14, 23] Amaro, F. and Picciotto, S. (2026) Possible EU own resource based on a digital levy: Cross-border services trade, digital transformation and tax implications. Brussels: European Parliament.

[15] OECD (2026) Reallocation of Taxing Rights to Market Jurisdictions. Paris: OECD/G20 Inclusive Framework on Base Erosion and Profit Shifting.

[16, 18, 19] OECD (2023) Multilateral Convention to Implement Amount A of Pillar One: Overview and Factsheets. Paris: OECD/G20 Inclusive Framework on Base Erosion and Profit Shifting.

[17] OECD (2023) Explanatory Statement to the Multilateral Convention to Implement Amount A of Pillar One. Paris: OECD/G20 Inclusive Framework on Base Erosion and Profit Shifting.

[20] OECD (2026) Multilateral Convention to Implement Amount A of Pillar One. Paris: OECD/G20 Inclusive Framework on Base Erosion and Profit Shifting.

[21] OECD/G20 Inclusive Framework on Base Erosion and Profit Shifting (2025) Pillar One Update from the Co-Chairs of the Inclusive Framework on BEPS. Paris: OECD.

[22, 36] Thomadakis, A. (2026) Could a digital services tax become an EU own resource? Revenue potential, policy trade-offs and strategic options. Brussels: European Parliament.

[24] Council of the European Union (2022) ‘Council Directive (EU) 2022/2523 of 14 December 2022 on ensuring a global minimum level of taxation for multinational enterprise groups and large-scale domestic groups in the Union’, Official Journal of the European Union, L 328, pp. 1–58.

[25] European Commission (2026) Minimum Corporate Taxation. Brussels: Directorate-General for Taxation and Customs Union.

[26] OECD (2021) Tax Challenges Arising from the Digitalisation of the Economy: Global Anti-Base Erosion Model Rules under Pillar Two. Paris: OECD Publishing.

[27] Hugger, F., González Cabral, A.C., Bucci, M., Gesualdo, M. and O’Reilly, P. (2024) ‘The Global Minimum Tax and the taxation of MNE profit’, OECD Taxation Working Papers, No. 68. Paris: OECD Publishing.

[28] Brun, L., Pycroft, J., Speitmann, R., Stasio, A.L. and Stoehlker, D. (2025) The impact of the global minimum tax on corporate tax revenues: Evidence for EU Member States. Seville: European Commission Joint Research Centre.

[30] European Commission (2026) Manual for MNE Groups on Global Minimum Tax (Pillar Two) Compliance Obligations. Brussels: Reform and Investment Task Force.

[31, 32, 33] Hebous, S. and Mengistu, A. (2024) Efficient Economic Rent Taxation under a Global Minimum Corporate Tax. IMF Working Paper WP/24/57. Washington, DC: International Monetary Fund.

[34] Hebous, S., Prihardini, D. and Vernon, N. (2022) Excess Profit Taxes: Historical Perspective and Contemporary Relevance. IMF Working Paper WP/22/187. Washington, DC: International Monetary Fund.

[35] Hourani, D. and Perret, S. (2025) ‘Taxing capital gains: Country experiences and challenges’, OECD Taxation Working Papers, No. 72. Paris: OECD Publishing.

[37] European Commission (2024) Annual Report on Taxation 2024: Review of Taxation Policies in the European Union. Luxembourg: Publications Office of the European Union.

[38] European Innovation Council (2026) European Innovation Council 2026 Work Programme. Brussels: European Commission.

Abstract

Data centres are increasingly presented as an attractive tax base, given their capital intensity, immobility and local demands for electricity, land, water and other infrastructure. The article argues, however, that the physical location of servers cannot automatically identify the wider economic rents generated by AI intrinsic to the wider AI value chain. These value streams are dispersed across geographies and layers in an international value chai- from semiconductors, hardware and energy, through cloud orchestration, foundation models and application software, to proprietary data and ultimately to adopting entities.  These locations may differ substantially from the physical location of the server and are linked through contractual, licensing, transfer-pricing and corporate ownership arrangements which, together, separate compute geography from the location of accrued economic rent and associated profit. The article contends that host jurisdictions do have a legitimate claim to recover for identifiable land, grid, water, environmental and other infrastructure costs, but do not automatically acquire taxing rights over all model, cloud, softwar and market-based rents.  It proceeds to differentiate the locational characteristics of training and inference workloads and to note how mixed workloads weaken the case for a broad facility levy. Concluding that Europe's key regulatory dilemma is to balance legitimate cost-reflective charges with an interest in growth and innovation, the article advocates taxing site-specific infrastructure and environmental burdens locally while addressing wider rents through separate profit-allocation and jurisdictional rules.

1. Introduction - Why Data Centres Are an Attractive but Incomplete Tax Base for AI

The attraction of a city-based and physical tax base for data centers is straightforward. If AI diminishes aspects of the tax base that have traditionally fallen on labor income, its employment, its plant and equipment or its labor-intensive processes, then the tangible, capital-intensive, polluting, geographically fixed over the short to medium term and state-plannable nature of the data center infrastructure seems to offer a recognizable alternative handle-one that is easier to notice, measure and charge than less tangible digital profits. In a digital, and indeed AI, age, where value appears to be drifting partly across borders via codified lines of code in the cloud, the data center tower seems to have prima facie appeal as an analog solution. However, that intuition is incomplete. The tangible placement of data center infrastructure can justly underpin certain local taxation and fee-recovery rights, but it does not secure a rightful claim in law or economy as to the entire AI-generated economic rent.^[1]

The trade-off is that a data center is just one node in a layered and globally distributed value chain.^[2] OECD's recent benchmarking of AI supply shows chips, data centers, clouds, models, connectivity and end users form part of the same supply chain, which has a structure characterized by large fixed costs, scale economies, vertical integration and bottlenecks not exactly aligned with taxes across national frontiers.^[3] OECD analysis shows that leading AI firms increasingly operate across several layers of the infrastructure stack, while countries differ markedly in their specialization across compute, cloud, models and applications. A host country can thus earn property taxes, construction-related activity, operator and contractor wages, electricity payments and utility revenues from hosting the data center, while much larger rents go elsewhere to chip design, semiconductor fabrication, hardware suppliers, cloud hyperscalers, model vendors, integration software, proprietary data owners, adopting firms and final investors.

A single charge against the coal in the mines, or those contracts, or those silicon wafers, involves other geographies of rent. For computing, geography has to determine where the servers actually run. For energy geography, the question is where electricity, capacity, storage, cooling and transmission lines are supplied from and reinforced. For labor geography, there is the question of where the engineers, technicians, managers, AI researchers, software engineers, facility technicians, clients and end-users are. For intangibles, it is where legally protected models, patents, the software program, or the data rights are owned or exercised. For corporate, it is where the subsidiary posts its contract, invoices and reports. For markets, it is where the paying users or consumers are. Each separate geography may imply a participant in value creation, a legal interface and a political policy base. Entry into the data center, as if it encapsulated all those six geographies simultaneously, erodes distinctions maintained for inextricable reasons by international tax law, by utilities regulation and by industrial economics.

This differentiation is clearly observable in tax doctrine.^[4] According to the OECD Model Tax Convention, a server may qualify as a fixed place of business under certain conditions, but a website by itself has no taxable presence and ordinary hosting arrangements do not usually bring the server into use for the business tenant at the hosting.^[5] Where a permanent establishment does exist, the location of the hosting service does not alone tell how much profit is attributable to it. Transfer-pricing rules draw another difference; the legal owner of an intangible property does not by itself receive the residual return if the development, enhancement, maintenance, protection and exploitation of that intangible property are carried out or controlled elsewhere.^[6] Put otherwise, neither the server nor the nominal owner of the source code is sufficient to define the locus of AI rent.

The truth is therefore more subtle and the mainstream rhetoric is simply too coarse to capture what is realistic. Data-center taxation is feasible when it recovers identifiable local costs, when it accounts for local externalities and when it only taxes the incremental operating income of the data center that is genuinely attributable to it. It ceases to be analytically sound and may even prove distortionary, when it is believed to be a surrogate for the broader economic rent created along the entire AI value chain. The server has a place in the matrix, but indispensability should not be confused with platform dominance. Sound policy must distinguish he difference between taxing local environmental and physical infrastructure externalities and taxing facility profits and claiming taxing rights over rent extracted from cloud, model, software, data and IP ownership and design and capturing the tax domain of the jurisdiction in which paying users live. The rest of the paper explores this nuance in terms of the AI value stack, the multiple geographies of AI, the justifiable claims of host jurisdictions, the shortcomings of data facility taxation as a proxy for AI rent and an application to the particular strategic dilemma now facing Europe.

2. The AI Value Stack: Seven Layers of Value Creation and Rent Capture

The first is that an incomplete tax proxy for the data center exists because AI value is not generated on a lone asset but on an entire stack. Closest to the value chain are semiconductors. The need for powerful AI capabilities pushes for GPU and AI accelerators, high-bandwidth memory, foundry services such as TSMC’s and manufacturing capacity, all of which are characterized by high capital intensity and high concentration and differentiation.^[7] Previous OECD research on AI infrastructure already demonstrated the fact that the GPU market is highly concentrated, that complementary software (e.g., CUDA) is integral and that specialized upstream suppliers, including TSMC, ASML, SK Hynix and Samsung, occupy critical positions without necessarily participating in consumer cloud application markets.. Whenever compute becomes limiting, or advanced-generation chips are constrained due to supply problems, then a large share of rent is specific to this layer, rather than the later data centre.^[8]

The second layer includes servers, networking, storage and the construction of the data center.^[9] This is the layer most visible to local decision makers since it encompasses the buildings, racks, cables, switchgear, site preparation and complex engineering works. It is also the layer most often called upon in the eyes of many by public debates about the locus of AI. Yet even here, the economic footprint is quite equivocal. Some value is gained on-site in terms of civil engineering works, construction services and routine operations. Much of the value in the provision of advanced AI flows to globally active equipment vendors, engineering consulting firms and integrated infrastructure suppliers, whose earnings do not remain local. OECD studies emphasize that modern AI systems are reliant not just on an anonymous warehouse of servers but on advanced interconnects, resilient networks and digital backbone infrastructure -a good proportion of which is owned or coordinated by the world's most major technology players with a global presence. The physical plant thus remains a requirement, but is often not the primary owner of the space of strategic value, which renders the shell itself valuable.

The third layer includes electricity, the grid, access, cooling, land and water.^[10] Here, the argument for local public benefits is strongest. Power is almost always the largest operational expense of running a data center; heating or cooling loads and power densities are rising dramatically for AI-centric data centers; and according to the International Energy Agency (IEA), it may not be long before a single AI server rack consumes maximum electrical power similar to that of dozens of households.^[11] OECD analysis concludes that these data centers consume enormous amounts of electricity, often excessively consume resources like water and require large, costly cooling infrastructure. Electricity supply and water supplies and discharge systems are geographically sticky because the location of grid connection points, substations and water will be explicitly local. This lets hosts claim the value for the environmental costs and burdens of the local infrastructure and resources, even if the rent flows up the stack to many global owners.

The fourth layer is the cloud-computing and orchestration services.^[12] Economics is now no longer about physical raw materials. Rather, they are about having control over the shared capacity, interfaces, middleware, scheduling, managed services and ecosystem access. Since the OECD shows that the largest hyperscalers are especially well positioned to claim large proportions of the AI cloud market, that combined hyperscaler market shares are consistently very large across national and regional studies and that a growing proportion of large AI firms go to market primarily via partnerships through which cloud providers provide both capital and compute, then it follows that much of the product is in fact the contract, not the server hall in Finland, Ireland, or Virginia; while large proportions of the quasi-rent can be claimed where the racks are in another country and the customer never actually sees its servers.

The fifth layer is the foundation models and model APIs.^[13] Frontier-model developers can capture rents through scarce expertise, safety fine-tuning, brand, switching costs, first-mover benefits, proprietary evaluation pipelines and the ability to commercialize a pretrained model over numerous downstream applications. However, these companies are not necessarily completely vertically self-reliant. BIS evidence demonstrates that private frontier model companies like OpenAI and Anthropic have limited direct participation in compute and cloud markets and utilize close partner and infrastructure arrangements. This matters for tax analysis because the model rent may not sit with the data-center operator or with the customer-facing user, but with a third party who licenses or provides access to the model via cross-border contracts, which are themselves possibly routed by different connected firms.

The sixth layer comprises data, software, applications and distribution. This is where narrow model capacity can translate into industry-specific commercial value. Firms that own or maintain proprietary industrial data, customer workflows, embedded software, enterprise integration capacity and distribution channels may be able to claim economic returns that aren't readily reducible to the raw compute or frontier model. Evidence on national AI ecosystems indicates that economies without leading positions in compute or frontier models may nevertheless specialize in downstream applications, software and distribution. This is an important antidote to data-center-tax narratives. Even if the most intensive compute activities are thousands of kilometers away, a significant portion of the value that can be effectively appropriated may be claimed by those firms that translate the models to the relevant markets, own the relevant workflow, or maintain the route to the customer. The seventh layer is the adopting firm and its final customers.^[14] Economically, not all of the value of AI is captured as taxable profit by upstream digital companies. Some of the value is transferred down as lower prices, better/cheaper products, better quality, less downtime, faster design cycles, higher yields, or newer output in the adopting company. Some accrue to end consumers. Some is earned by low-cost producers of an imported model, on an imported cloud, who build a factory, hospital, supply chain, or professional service that is more productive. So the largest social or private gain from AI adoption can potentially be realized outside the data center and outside the model vendor company. The AI value chain, thus, does not end when an individual facility powers up a GPU cluster; economically, it ends where the adoption shifts output, margins, quality and market power.

In principle, these seven levels can also see their rents shift over time.^[15] During times of chip shortage, rents may accrue to upstream suppliers of design, foundries and memory. When power-rather than efficiency-becomes the critical bottleneck, tremendous rents could go to the few rare grid access providers or the handful of players capable of quick energization. When cloud monopolies tighten their grip in the coordination layer, tremendous rents might eventually go to those few providers who link platforms together and those firms that lock in entire ecosystems of users. Where model interface measures become commoditized and where open alternatives flourish, these rents may migrate to all but the earliest stages of industrial applications, to industrial data, or to the distribution layer. Where adoption diffuses to all firms, the distribution of rents may shift again toward those firms that best reorganize production itself. The available evidence therefore, suggests that policy should not assume that current rent-holders will necessarily hold power forever because their facility is eventually heavily taxed and therefore fixed in place.

3. Six Geographies of AI Value: Compute, Energy, Labor, Intellectual Property, Corporate Structure and Markets

Since the AI value stack is layered, the production of AI services and products also occurs across multiple geographies that are only partially overlapping.^[16] Compute geography specifies site of server and accelerator installation and job execution; energy geography specifies site of electric power and grid capacity, backup, cooling and transmission reinforcement; labor geography identifies where model engineers, site of facility operators, where associated professionals work, site of people working with the output; intellectual-property geography identifies wher ownership of models, software, patents, datasets, or licenses; corporate geography specifies the location of the parent and subsidiary contracting, invoicing, funding and profit-reporting site; market geography specifies the location of consumers and end users of AI products and services. The location of the economic value stems not from any one of these, but from the interaction of all six. Compute geography and energy geography are not the same thing, even on the same site. The IEA finds that the siting and expansion of data centers is highly dependent upon local grid variables, queue times and capacity to bring large loads online.^[17] The OECD finds that the availability of water and cooling can influence siting decisions. A Finnish or Swedish site may be selected because its temperature, electrical mix, water availability and permitting process will enable it to support inference or training loads efficiently. Such a choice says a lot about local infrastructure scarcity and local system costs, but it does not mean that Finland and Sweden become the sole source of the firm's model, customer base, or reported profits.

Labor geography and market geography often tend to differ even more from compute geography. The team working on the initial development of the model might be based in California, London, Paris, or Berlin. The operations team administering and updating the site might be based in Finland. The customer success team might be based in Dublin or Warsaw. The engineers actually using the system might come from Bavaria and the head of production might come from Poland. The organization paying for the service might be a German manufacturer, but its customers will be spread across the rest of the European Union. In terms of taxation, these differences matter because production, consumption and business activity can all stay within the jurisdiction withoutsubstantial local compute capacity. In terms of economics, they matter because the productivity effects of artificial intelligence are powered by organizing a firm once it adopts the technology – not by the servers in the data center being cooled down.

Another level of stratification was created between intellectual-property geography and corporate geography. In many cases where commercial AI services are sold into the EEA, the services may be contracted through Irish entities with respect to the service, even if the technology is created elsewhere. OpenAI's European terms state that when European EEA and Switzerland residents seek to contract with OpenAI, customers that reside within the EEA or Switzerland shall enter into the terms of the agreement with OpenAI Ireland Ltd; whereas the contractor-entity schedule for Google Cloud states that for much of EMEA, the contracting-entity name is Google Cloud EMEA Limited, stated in Dublin.^[18] The mere presence of such foreign locations (for services, contracting entities, or even ultimate parent companies does not necessarily reflect where profits are ultimately taxed and potentially where profits will be taxed, even if no activity is directly linked to holding that specific location), but the location of an Irish contracting entity remains a possibility for a European customer where the model is created in the US, run on servers within another Member State, and at the same time, under an ultimately U.S. parent company. For representative factors such as who is legally invoiced, where the server is located, where the ultimate parent company is located, and who is the beneficial owner, the distinctions are difficult to discern.

The hypothetical commissioned by a German manufacturer, using an American model, via a contracting Irish entity, running inference in a Finnish data center, on chips designed in the United States, fabricated in Taiwan, applying German industrial data to improve the output of a Polish factory but still owned by a parent corporation in Germany, is economically and legally plausible. Each element is a described pattern elsewhere in the current AI economy. Google publicizes Hamina in Finland as one of its live data-center locations.^[19] OpenAI and Google both contract some of their European work through Irish entities. Publications by the OECD demonstrate that chip design, fabrication, cloud infrastructure, models, and applications are themselves geographically distinct and frequently located in separate jurisdictions. The real legal point is not to claim the example as exhaustive or universal, but rather, to demonstrate that it is theoretically, legally, and economically feasible for the one unit of AI-enabled value to pass through no fewer than six jurisdictions before taxing rights must be allocated among jurisdictions.

Once that multi-jurisdictional chain is recognized, the boundaries of a facility tax are clear.^[20] An Irish or German tax on the value generated in a data center in Ireland or Germany might be justified to the extent that the data centers are the means by which land-use authorizations, access to the grid, cooling infrastructure, ecological regulation, connected transport and communication links, and the public services and streets of a neighborhood are supplied. An Irish, German, or other national taxing the profits generated by servers within a particular permanent establishment or facility business is probably justified to the extent that they are collecting from the operation of the servers taxes on the resources used by the servers, be they the services of a Polish, U.S., Taiwanese, or other chip designer or the running costs of an Irish contract manufacturer. But the home country of the server is not in and of itself in a position to determine what part of the overall AI surplus should go to the Taiwanese foundry, the Irish contractor, the German consortium, the American chip home, and the European end user.

4. What Data-Centre Host Jurisdictions Can Legitimately Tax and Charge

By untangling the broader geography of value in AI, the fiscal claim of host jurisdictions is not any weaker but rather clearer. There is a compelling normative argument for a data-center host to charge for what it supplies in practice, and for adjusting for the additional burdens that the facility places on the local systems.^[21] These norms are familiar: benefit taxation, user charging, recovering costs, and Pigovian correction. If a data center occupies scarce land, necessitates rezoning, consumes public infrastructure, and its operation strains constrained grid capacity, the host jurisdiction can legitimately charge to recoup the costs. This claim carries its greatest force when it is implementation-neutral, can be relatively quantified, and is defined in relation to the immediate, local facility rather than certain purported rights over the rents of the distant AI stack. That logic clearly leans towards natural land-use controls, property taxation, building- and civil engineering-related charges, and local authority permitting fees. A server hall is a building on land. It may necessitate road access, public safety supervision, flood or fire planning, and utility cooperation. The IEA points out that data centers can become operational in between 2 and 3 years, which on its own indicates a significant and concentrated development process, even if the broader energy system required to support them proceeds at a slower rate. There is therefore nothing conceptually unusual about considering the facility in its initial form to be no different than other large industrial or warehousing assets when it comes to land, building, and property taxation. These are not levies on AI rent in the abstract; they are levies on land-use and built environment infrastructure on a site-specific capital asset.

Grid-connection costs, transmission and distribution upgrades, and capacity-reservation charges are defensible if they correspond to system costs.^[22] One must make a special case here because data centers are capable of creating system costs that extend well beyond their fenced perimeter. For instance, IEA analysis of Europe shows waiting times for grid connections averaging over two years, core European hubs experiencing seven-to ten-year queues, and the European Union needing some policy coordination simply to match project pipelines with existing electricity infrastructure.^[23] ACER reports that in 2024 EU transmission system operators spent 4.3 billion on various remedial actions to address grid congestion.^[24] If a facility needs a dedicated network line, a new substation, flexible connection arrangements, or network reinforcement, then the host system is not just hosting a digital service but securing limited grid capacity. Costing that scarcity with market forces is internally consistent and can help guard against other ratepayers being burdened by implicit cross-subsidy.

Water use, cooling, carbon emissions, externalities of backup generation, and environmental monitoring are other justifiable local compensations. OECD analysis emphasizes that many AI data centers are water-dependent in cooling requirements and that water access is a cause of location; it also concludes that cooling technologies are changing, so local policy should focus on actual resource use rather than a blanket anti-data-center approach. The EU's second-generation data-center energy-performance framework has explicitly included monitoring of water usage once the facility becomes operational. Carbon or local air-pollution taxes are equally justifiable if on-site generation or backup-generator testing, or local particulate matter emissions impinge on the inflowing community. Again, the relevant tax or charging base is not the global AI but the physically present load and its measurable externalities. Ireland is an example of how such a claim can be meaningful. Central Statistics Office Ireland has stated that data centers were responsible for 5 percent of metered electricity consumption in 2015, 22 percent in 2024 and 23 percent in 2025.^[25] At these levels, the host-country validity of concerns over the adequacy of the grid, capacity planning, limited locations, and fair distribution should be clear. Such concerns are clearly not merely rhetorical; they are indicative of how a significant proportion of an actual national electricity system should be allocated. Slightly smaller-scale data-center concentrations could cause similar host country concerns in other parts of the world. This justifies the prediction that the charging of infrastructure and pollution costs is not merely a pretext for taxing intangible profit flows, but an actual attempt to deal with the substantial, geographically distributed consequences of digital loads that require substantial infrastructure.

European law and policy now increasingly mirror this logic of host jurisdiction precisely. The recast Energy Efficiency Directive requires Member States to cover reporting by significant data centers, and the Commission's 2024 delegated regulation established the database and KPIs for that reporting.^[26] The Commission's data-center performance page states that the database collects information on energy performance and water footprint for sites with significant energy consumption, and earlier Commission documents established the reporting threshold at built-in (IT) power consumption of at least 500 kW.^[27] The very same framework makes clear that the legal regime in place, not yet evolved into an EU tax instrument, is essentially a transparency and monitoring one. It is not a mature EU tax instrument even now, and no attempt should be made to categorize it as such.

As of mid-2026, the Commission is preparing a Data Center Energy Efficiency Package that would include baseline data collection, a rating scheme, and an electronic label, followed by work on minimum performance standards.^[28] The Commission's March 2026 call for evidence outlined the proposed rating regulation as a follow-up to the 2024 delegated regulation and the Commission's broader energy efficiency roadmap, which identified minimum standards in a consultation phase.^[29] In other words, EU reporting and collecting obligations are in place, but the rating-label regime and any mandatory minimum performance standards are a work in progress, not a finalized, fully functioning tariff and regulatory scheme. This distinction is central to accurate analysis. Existing standards and labels could influence costs and investment signals, but they should not be conflated with implemented fiscal measures or an EU decision to extract rents from global AI by taxing server venues.

The correct policy position, therefore, becomes the disciplined rather than anti-host one. Data-center hosts could legitimately charge for land, property, building, and connection, congestion, environment, monitoring, and facility-flexibility profits, impose locational conditions, and negotiate obligations of system-integration (such as use of waste-heat or flexible loads) where appropriate. However, these charges should be aimed at as site-specific public financing tools, not policy rhetoric used as an improper substitute for taxing the whole of AI. After the financial purpose is changed from recovering local costs to capturing the entire surplus of global AI, this data center is no longer an appropriate tax base, and a comparatively inappropriate substitute.

5. Why Physical Server Location Is an Incomplete Proxy for AI Rent

The most significant logical error in the data-center tax proposal is a category error. A data centre is a physical input into AI, but it cannot be the only residual claimant on the model, be the sole owner of a customer relationship in AI, or stand to gain from the returns to AI adoption. In many commercial arrangements, it is a cost center, an infrastructure asset, or an inception node. The rents that really do matter may not be sitting on a facility but on a cloud platform distributing finite compute time among users, on a model supplier turning a boundary API into revenue, on a software company leveraging AI in the enterprise's processes, or on an in-house company sharpening its margins and output.

This is further underscored by demand heterogeneity. Data centers are not solely for AI use, and even the abrupt current spate of activity in AI does not change the multipurpose character of the infrastructure it is built upon.^[30] The IEA estimates that accelerators (of which the majority are AI accelerators) constitute nearly half of the net growth in global data-center electricity use by 2030; conventional data-center servers still account for just under a fifth, and the remaining part is accounted for by data-center infrastructure and other IT equipment.^[31] All three main types of data centers: enterprise, colocation, and server provider, and hyperscale (including AI-driven workloads of any type) contribute to the overall demand increase. Any facility-level tax base attracted by AI rent will therefore almost certainly appear as some combination of AI and non-AI workload within a mixed-use facility (absent a transparent way of individualizing them in meters, which the public debate does not seem to assume); the potential end result could be a tax on non-labor digital infrastructure with only partial links to frontier-AI business phenomena.

Training and inference also differ in ways that weaken the argument for a simple facility proxy.^[32] OECD analysis describes training as extremely expensive, requiring high-density clusters, specialized accelerators, high-bandwidth networking, power, and cooling. Inference inevitably also consumes hardware optimized for energy efficiency (or latency). But as services reach the point of commercial scale, they harness device-to-edge latency effects, and cloud providers also benefit from developing optimum local or regional infrastructure, for example, for security reasons. Carnegie's analysis of data-center competitiveness shows that for major compute investments, time to power, project speed, and the availability of infrastructure may dominate over even huge differentials in local taxes or electricity prices.^[33] In fact, these two facts point in different local directions. Frontier-model training may often be best concentrated where power, chips, financing, and permitting are all readily available. Inference, by contrast, needs to be assembled along the geographies where latency, customer proximity, security, and institutional conditions are optimum. A single, uniform facility tax may therefore not be adequate to capture the value spread across all AI workloads.

Legal complications extend the physical/taxable location divide. The OECD Model Tax Convention supports three propositions that are relevant here: a website by itself is not a place of business, a hosting arrangement will generally not involve a server being made available to an enterprise, and even where a server or related facility does constitute a permanent establishment, that alone will not determine the division of profit attributable to it. Further, the business conducted on the server must be analyzed on a case-by-case basis; attribution is a separate determination. Modern cloud architectures heighten this importance because a cloud client accessing AI may never own, lease, or have managerial control over the specific server involved in generating a model response. The transfer-pricing rules introduce a second form of incompleteness. The 2022 Transfer Pricing Guidelines issued by the OECD are explicit that legal ownership of an intangible asset does not confer the right to a 100 percent share of its returns: what counts is who does, supervises, and bears the risks for development, enhancement, maintenance, protection, and exploitation. If the foreign-owned affiliate holding the legal title to a platform, portfolio of patents, or software stack doesn’t control or perform those functions, that affiliate shouldn’t be guaranteed to get all the income flowing from those activities. The other side of this coin is equally significant for the facility argument: a data-center company that supplies power, space and operations but doesn’t own or develop the intangible shouldn’t automatically be where the model rent is located. All manner of revenue flows-royalties, license fees, cloud charges, cost-sharing arrangements, services agreements-can be used to redistribute income from the physical location, subject to arm’s-length and treaty limitations.

A blunt facility charge can be distortionary because, rather than falling on the parties best able to bear or pass on it, it targets a plausible global bottleneck. OECD evidence demonstrates that the most substantial AI and cloud companies have scale, scope, vertical integration, and multi-layer presence. Firms active across compute, infrastructure, models and applications are generally those best able to push workloads to others, repackage contracts, secure high-powered deals or bear higher facility costs for long periods. These advantages, which are unequally distributed within and across communities of local and global research users, universities, start-ups, and incumbent providers, will come into play in the event of a facility tax and may prevent it from falling solely on the incumbent rent recipient. Private providers, domestic research organizations, universities, and local start-up providers will be less able to pass that fee through to their own constituents. So targeting the rent through a facility tax may be costly even if the tax is invoked rhetorically.

The difference between operational and announced capacity aggravates the problem.^[34] Many public conversations about AI data-center booms often move from project announcements into assumptions about realized rent. However, the IEA Europe analysis explicitly distinguishes between pipeline capacity and installed capacity. It reports that the project pipeline in Europe implies a capacity of 130 percent of the current installed capacity, but that installed capacity by 2030 will increase by only about 70 percent in relation to 2024, due to delays and local restrictions (for example, on emissions or controls on end-user access due to grid congestion). In other words, levies or obligations related to announced capacity, memoranda of understanding, or speculative AI-hub announcements could be put in place long before the manufacturing activity is capital-intensive, operational, profitable, or yet to be built. For an already tenuous attempt to tax emerging AI rents, such a timing mismatch makes the proxy even less robust. On the whole, the takeaway is not that taxation of facilities is misguided-it is just a more limited tool than most supporters believe. It is effective at valuing local externalities and taxing the operations of facilities. It is ineffective at embodying the putative makeup of AI-generated rent when the chips, the models, the contracts, the data, the producers, the adopters, and the shareholders are frequently offsite. The farther away from local cost return a policy goes, and the closer it gets to claiming a share of AI's total economic surplus, the more likely it is to be invalid analytically and economically distortionary.

6. Europe’s Policy Trade-Off: Cost Recovery, Compute Expansion and the Risk of Taxing the Wrong Base

The European policy risk results directly from the above analysis.^[35] Europe is now pursuing two objectives. On one side, it is tightening up data centers' energy efficiency, use, and integration into systems. On the other hand, it is explicitly trying to grow its domestic cloud and compute capacity, as part of a new overall agenda for AI and technology-sovereignty. The Commission's AI Continent plan envisages tripling EU data-center capacity at least within five to seven years, and its proposal for a Cloud and AI Development Act aims at easing deployment and access to energy, land, water, and finance, towards building up strategic capacity for AI, cloud, and computing-intensive applications.^[36] Such a policy mix does not sound inconsistent at face value. It can only sound inconsistent if Europe is going to regard the data center equally as a limited factor of production that must be grown and as a fictional tax base convenient for extracting the entire rent of AI.

The pain is genuine because the pain of Europe's energy constraints is genuine. The IEA finds EU grid-connection waits that can range from two to ten years, and concludes that Europe will not realize the full volume signaled by its project pipeline under current constraints. Data centers are also expected to contribute 10 percent to EU electricity-demand growth by 2030 in current policy settings.^[37] In that setting, cost-reflective local levies are not a policy indulgence; they are a necessity both for economic management and public support. Host communities and network users cannot reasonably be expected to freely carry additional congestion, capacity shortages, water use, or system-upgrade costs. The mistake is elsewhere: in stacking on non-cost-reflective fiscal loads to imitate an AI rent tax, and simultaneously expecting a faster rollout of private digital infrastructure in a region already constrained in the squeeze between time to power.

These goals can, however, be partially reconciled. The first step is an honest assessment of the mission of each instrument. Cost-reflective connection fees, capacity-reservation charges, water charges, carbon tax, and environmental reporting covered local costs and externalities, whereas easier permitting, queue management, flexible interconnection design, and better planning addressed deployment bottlenecks. The IEA states that non-firm connections and queue management would provide a way to upgrade ready-to-build projects; and the Commission is developing so-called tripartite models between data center operators, other energy players, and public authorities precisely to make data centers an integrated part of the local energy system & heat networks. A Europe that makes data centers pay for the costs they impose and also reduces unnecessary permitting and grid frictions is not contradicting itself; a Europe that seeks to morph facility charges into a new, lower global AI profit surcharge does contradict itself.

The idea that certain frontier-model training might take place in the Middle East and other resource-plentiful regions, while inference occurs nearer to European endpoints, should thus be put to the test rather than simply accepted or dogmatically thrown out. Carnegie's cross-national competitiveness index places the UAE high in part because of project speed, and because prolonged delay dominates other cost factors; but it also illustrates how political, economic, and security shocks, or conflict-driven delay, can cause rapid oscillations in such rankings. The economics of locating some training in power-providing and fast-permitting economies and geographies is sound. Large training loads are capital-dependent, schedule-sensitive, and more easily relocated than many policymakers appreciate. But that alone is not necessarily enough of a cure to Europe's need to bring certain inference workloads loser to EU users, or to preserve some strategically necessary compute capacity, either within or beyond the EU's borders, for resilience, security, and public sector control; or to realize the EU digital-ecosystem objectives that cannot be achieved simply by externalizing certain capability-creating activities.

Both of the strongest potential counterarguments to the paper's thesis, therefore, warrant serious consideration. Data centers have the potential to produce significant investments, local construction expenditure, utility requirements, local government receipts, resilience gains, and ecosystem externalities; none of these arguments necessitates the assumption that the data center captures all of the rent accruing to AI, and none of them would be diminished if the Commission moved to regulate, tax, or subsidize facilities at the least-cost jurisdiction. The danger would come if one slides from those sensible host-jurisdiction arguments into the further conclusion that a tax on the host facility is therefore the best way of taxing AI.

This is also where the conceptual difference between facility taxation and market-jurisdiction taxation becomes essential. The 2026 briefing document on cross-border services taxation produced by the European Parliament states that digitalization means that services can be supplied from afar at near-zero additional costs, and describes one policy lever as trying to reverse the balance of taxation taxing rights – nations where the services are supplied to global markets instead of where the services come from.^[38] But whether or not one favors any digital-levy measure, this particular debate involves a different kind of tax entity, a different tax object entirely from a local charge on a data-centre facility. One involves global customers and globally sourced services. The other involves a locatable infrastructure asset. Mixing the two leads to bad policy in compelling a local utility-and-environment measure to do the job that a global profit allocation inquiry should.

The simplest and most justifiable European stance is therefore a tiered one. Taxes and charges on AI facility use should be narrowly correlated with local costs, congestion externalities, and environmental impact, and levied universally among all workloads, inclusive of AI, proportionately to their demands. Any policy to expand AI compute capacity should be targeted at shortening power-up times, integrating more additions to national grids, siting capacity where there is genuinely existing headroom, and protecting critical inference and public-interest AI computing within the Union and allied safe havens. Where broader AI profits and rents are taxed, the taxation of broader AI profits and rents should be directed at taxing those profits and rents rather than relying on the mistaken assumption that the full equation of taxable surpluses resides wherever a server rack is sited, as long as it happens to be getting the power from a server rack. Europe's mistake would not be if data centers were made to pay their way locally. It would be treating that local payment as a substitute for taxing AI value where it is actually earned.

7. Conclusion - Tax the Facility for Local Costs, Not the Entire AI Value Chain

The case for taxing data centers starts from a solid intuition and would end up in a narrower conclusion than many of its proponents seem to desire. Data centers are fixed, resource-use-heavy sites that impose tangible burdens on land, grids, cooling, water, and local public services; the home jurisdiction can thus fairly seek to recover local expenditures, impose local externality prices, and tax the profits that are truly generated there by data center operation. It does not contain an unearned claim to the full AI rent in the form of the full economic rent, simply on the grounds of having a few servers stacked within its geographical jurisdiction. AI's value is spread across chips, clouds, model-builders, software system-stacks, proprietary data, user firms, and final customers: each with different geographies, jurisdictions, and legal considerations. A server hall is a single essential point on an AI value chain, but by no means is it the natural geographical location of that value.

The strategic implication is unambiguous. By using facility taxation as an approximation of AI rent taxation, Europe (or another state) may be collecting too much tax on a relatively immobile taxable input, and too little on more mobile, legally complicated forms of ability-to-pay and value creation. The right way forward is institutional separation. Local facility-specific charges to reflect relative resource costs, taxed in a tightly calibrated manner at the facility level, and a separate global-policy debate to attribute other profits, intangible assets, and jurisdictional rights. The priority is not the abolition of data center taxation, but the restoration of its proper function.

This article was prepared as an independent research contribution following the conference Inequalities in Longevity, held at Fondazione Giorgio Cini in Venice on 3–4 July 2026. It develops themes presented and discussed at the conference but is separate from the official conference proceedings.

Unless expressly stated otherwise, this publication has not been commissioned or endorsed by Fondazione Giorgio Cini. The analysis, interpretations, and conclusions are those of the author(s) and do not necessarily reflect the official positions of Fondazione Giorgio Cini, Swiss Institute of Artificial Intelligence (SIAI), or their respective affiliates.

References

[1, 2, 3, 7, 12, 15, 16, 32] OECD (2025) Competition in Artificial Intelligence Infrastructure. OECD Roundtables on Competition Policy Papers, No. 330. Paris: OECD Publishing.

[4, 5, 20] OECD (2017) Model Tax Convention on Income and on Capital: Condensed Version 2017. Paris: OECD Publishing.

[6] OECD (2022) OECD Transfer Pricing Guidelines for Multinational Enterprises and Tax Administrations 2022. Paris: OECD Publishing.

[8] Sastry, G., Heim, L., Belfield, H., Anderljung, M. et al. (2024) ‘Computing Power and the Governance of Artificial Intelligence’. arXiv, 2402.08797.

[9] Pilz, K. and Heim, L. (2023) ‘Compute at Scale: A Broad Investigation into the Data Center Industry’. arXiv, 2311.02651.

[10, 27, 28, 30] European Commission (2026) Energy Performance of Data Centres. Brussels: Directorate-General for Energy.

[11, 22, 23, 31, 34, 37] International Energy Agency (2026) Key Questions on Energy and AI. Paris: IEA.

[13] Stanford Institute for Human-Centered Artificial Intelligence (2026) Artificial Intelligence Index Report 2026. Stanford, CA: Stanford University.

[14] International Monetary Fund (2024) Broadening the Gains from Generative AI: The Role of Fiscal Policies. Washington, DC: International Monetary Fund.

[17] International Energy Agency (2026) ‘Data Centre Electricity Use Surged in 2025, Even with Tightening Bottlenecks Driving a Scramble for Solutions’. Press release, 16 April.

[18] OpenAI (2026) Europe Terms of Use. Updated 16 January 2026.

[18] Google Cloud (2026) Google Contracting Entity. Google Cloud.

[19] Google (n.d.) Hamina, Finland — Google Data Center Location. Google Data Centers.

[21] OECD (2019) Taxing Energy Use 2019: Using Taxes for Climate Action. Paris: OECD Publishing.

[24] European Union Agency for the Cooperation of Energy Regulators (2025) Market Monitoring Report: Electricity Wholesale Markets. Ljubljana: ACER.

[25] Central Statistics Office Ireland (2026) Data Centres Metered Electricity Consumption 2025. Cork: Central Statistics Office.

[26] European Parliament and Council (2023) ‘Directive (EU) 2023/1791 of 13 September 2023 on Energy Efficiency and Amending Regulation (EU) 2023/955’. Official Journal of the European Union, L 231, pp. 1–111.

[26] European Commission (2024) ‘Commission Delegated Regulation (EU) 2024/1364 of 14 March 2024 on the First Phase of the Establishment of a Common Union Rating Scheme for Data Centres’. Official Journal of the European Union, L 2024/1364.

[29] European Commission (2026) ‘Rating Scheme for Data Centres in the EU — Commission Launches Call for Feedback’. Directorate-General for Energy, 27 March.

[33] Phillips-Robins, A., Tawil, T. and Winter-Levy, S. (2026) The Compute Coalition: How to Build the Future of AI in the Free World. Washington, DC: Carnegie Endowment for International Peace.

[35, 36] European Commission (2025) The AI Continent Action Plan. Brussels: Directorate-General for Communications Networks, Content and Technology.

[36] European Commission (2026) Cloud and AI Development Act. Brussels: Directorate-General for Communications Networks, Content and Technology.

[38] Amaro, F. and Picciotto, S. (2026) Possible EU Own Resource Based on a Digital Levy: Cross-Border Services Trade, Digital Transformation and Tax Implications. Brussels: European Parliament.

Abstract

Artificial intelligence does not mechanically erode the tax base; the fiscal impact depends upon how the productivity gains are shared and where the income generated is taxed. This paper studies AI as a productivity shock, an income-distribution shock and a fiscal-transmission shock. It separates task automation from job displacement, firm productivity from macro employment effects and ordinary business returns from AI rents. The evidence suggests that generative AI can create productivity gains in selected tasks, particularly for lower-skilled workers,but has not yet produced broad employment displacement. Nevertheless, it is uncertain whether output resulting from such gains will always offset labor substitution in sectors with weak demand growth, in concentrated markets or where complementary investments are lacking. Accordingly, the fiscal impact will rest on whether such gains are ultimately paid to workers, consumers, domestic firms, investors, or foreign AI suppliers. Countries that rely heavily on labor taxes and social contributions, import substantial AI services or tax capital income weakly may be most affected. The most pressing policy dilemma is not in taxing AI as a technology, but in safeguarding the domestic tax base as digitalization shifts income from broad, immediately taxed payroll remuneration to more concentrated, mobile, or deferred forms of income.
 

1. Introduction - Does AI Replace Tasks or Complement Workers?

AI can achieve superhuman performance on selected tasks, but it cannot create a superhuman worker. For example, modern classification, summarising, translating, editing, compiling, searching, document comparison, pattern recognition and generation technologies can perform certain activities at a speed that is significantly faster than unassisted humans. Many processes that took days may now take hours. But the largest demonstrations of task acceleration can hardly be taken as evidence that the overall productivity of a professional, such as a lawyer, has been multiplied ten times. The agent’s job may again involve problem framing, fact checking, forecasting, communication, coordination and responsibility for error. AI may accelerate one step of an otherwise unchanged procedure or a new requirement for review or oversight may be introduced.

The strongest experimental and workplace evidence points to material but bounded productivity improvements. In a randomized trial of professional writing tasks, providing generative AI access cut average task duration by 40 percent and boosted average assessed quality by 18 percent.^[1] A study of customer-support agents shows that an AI assistant resulted in about 14 percent more issues resolved per hour on average, with substantially larger improvements among relatively inexperienced and initially lower-performing workers.^[2] In a randomized trial of management consultants, AI usage led to more task completions, hastened completion speed and better answers when assignments hovered within the model's capability frontier, but to poorer answers on some tasks designed to be at least partly outside of that frontier.^[3] The empirical case thus tilts towards selective amplification rather than a universal productivity multiple across intellectual work.

This distinction is important because the work task is often conflated with the entire job. A model could generate a draft version that automates but does not displace the task of a worker who defines the goal, checks the output and assumes professional responsibility. A worker can complete a routine job analysis faster but needs more time to do unusual analyses or advise customers. A company can output less labor content to accomplish one internal procedure, while increasing sales, quality, or new services. An economy can eliminate some tasks and still create new jobs in implementation, verification, data stewardship and customer service.The International Labor Organization's refined global index of occupational exposure underscores this gap between predicted exposure and employment displacement. It estimates that approximately 25 percent of global jobs are exposed to some degree of generative AI, with only 3.3 percent experiencing the highest levels of exposure.^[4] Exposure level is highest in high-income economies (34 percent) but still low in the lowest-income countries at 11 percent^[5] due to the preponderance of clerical, professional and digitized service work in the global North. Clerical jobs are the most exposed, but the International Labor Organization finds this unlikely to lead to complete job replacement, as most occupations blend activities subject to augmentation or automation with other activities requiring human input nonetheless. Measures of exposure are thus indicators of the technical feasibility of automation should it be adopted, not the probability of unemployment, nor a timeline for displacement.^[6]

Macroeconomic estimates are closer to the more moderate of the task-level results. Acemoglu's task-based study, for which he approximates the effects of presently available AI applications, reports them as being potentially substantial but, in aggregate terms, modest over ten years, with total-factor-productivity improvements below the more revolutionary estimates typically ascribed to generative AI.^[7] Early labor-market evidence has also failed to justify a prediction of employment losses across the economy. Administrative evidence from Denmark, tracking exposed occupations and thousands of workplaces, found no statistically significant net effect on earnings or recorded hours in the first two years of widespread chatbot adoption.^[8] That survey nonetheless documents several new tasks associated with implementation, oversight, integration, ethics and compliance. An alternative analysis of occupational task exposure indicates that increased mean AI exposure in affected occupations can decrease labor demand in the occupations themselves, whereas only mixing AI concentrations over a subset of a job’s tasks allows workers to reallocate effort, while enough productivity growth at adopting firms to counteract its direct substitution is somewhat, yet not fully, offset by scale effects, yielding much more modest total employment changes than one would expect based solely on simulated task exposure effects.^[9]

AI should therefore be interpreted as comprising three shocks at once. It is a productivity shock, in the sense that it may lessen the human effort and time necessary to undertake some particular types of activity. It is an income-distribution shock, in that the generated surplus will flow to one or another of labor, consumers, adopting firms, technology suppliers, investors, or scarce complementary employees. It is a fiscal-transmission shock, in so far as such recipients will be taxed through various mechanisms, at various effective rates, at different times and in different jurisdictions. Personal income tax, social contributions and payroll taxes are collected relatively quickly, whereas retained profits and unrealized capital gains may find their way into the domestic fiscal system more distantly. Hence, the central case should be put to the test and not taken for granted. The increased fiscal erosion will be more likely if substitution effects outweigh augmentation and scale expansion, if output increases are too small, if wage pass-through is weak, if rents are highly concentrated and if a large share of the technology payment or the proceeds of technology ownership arise outside the country that is home to the workers and customers. Gross domestic product is not equivalent to domestic fiscal capture.

2. Do Firms Expand Output Enough to Offset Labour Displacement?

Given a level of output, successful automation will require less labor for a given output. This accounting relation is commonly cited as a prediction of job destruction, though it describes only the initial phase of adjustment. Cheaper unit costs can be used to lower prices, increase quality and delivery speed, raise margins, or free up resources for new products and if demand grows enough, a firm may still be able to produce more goods with the same number of jobs, or raise employment while still reducing the labor hours needed per unit. The ability of expanded scale to offset displacements will depend on the elasticity of demand, competition, availability of capital and organizational capacity and on how much of the production process AI is capable of improving.

The relevant level of analysis is the task. An automated task can be defined as a task for which an AI-enabled workflow can produce a sufficiently acceptable output with little or no additional human input. An advantaged task involves the use of AI in a way that extends the speed, scope, or quality of work performed by a worker. An affected task is one in which there is some amount of partial automation, but which produces additional work in areas such as verification, communication, documentation, or other transaction costs. Jobs are constituted with some mixture of such task categories.^[10] Firms use compositions of job task bundles that vary substantially. An accounting clerk sending data to or from a standardized system, such as VAT returns, might be reduced to ordinary substitution; by contrast, an accountant using the system to analyze complex transactions might receive means of augmentation.

Firm-level European evidence supports the possibility of expansion. Akin to a finding reported in the European Investment Bank’s 2026 assessment of about 13,000 firms in the European Union and the United States, where AI adoption in European firms is associated with stronger productivity performance without establishing a uniform employment effect^[11] and the strongest benefits seem to accrue to medium-sized and large firms that can combine AI with investments in software, information and skills and organizational capacity.^[12] While not being able to establish that all adopting firms expand and that all occupations are protected in all adopting firms, these results can hardly be interpreted as contradicting the idea that the direct labor effect of AI can be offset at the level of the firm when the investments in complementary factor inputs and the output growth are sufficiently large.

A firm may grow, even if industry employment falls. It can happen if more productive early adopters take market share from other poor performers, raising their employment but reducing sectoral labor demand; or if the cost savings give rise to firm-specific new demand by boosting output, thus raising employment in the whole sector. Hampole et al. assist in distinguishing these processes. Their task-based proxies indicate that occupations in which the average exposure to AI/machine-learning abilities is higher suffer lower relative labor demand.^[13] However, this effect is mitigated when only some of an occupation's task package is exposed, as workers can compensate by intensifying other, less exposed activities. At the firm level, demand for labor related to increased productivity tends to compensate for much of the reduction in demand caused by the exposure of existing tasks.^[14] The overall impact thus remains relatively weak vis-à-vis the occupational substitution effect. This does not imply that all former employees will be retained, since expansion might draw more workers into other occupations, plants, or skills. It demonstrates that automation in tasks does not equate with total employment.

The scale response is sector-specific. In software, consulting, marketing and some professional services, lower production costs can enable doing more projects because the incremental cost of serving one more customer is low and there might be a significant latent demand. In the private and public healthcare, education and administration sectors, it won’t necessarily generate commercial sales but will accelerate throughput or improve quality. The fiscal benefit may appear by decreasing waiting time or allowing the same staff to serve more users. On the other hand, relatively fixed demand seems to apply to long-established back-office functions. Consequently, double the number of reconciliations, reports, or activities would not be realized just because AI enables them to be cheaper and faster. How the increased saving translates into expansion depends on the level of contestability. The firm, in a competitive market, is willing to cut prices or increase quality to the extent that it keeps its market share, which would increase demand and share some of this gain with consumers. If the market power is high, the savings are reflected in a bigger margin, thus curbing the scale effect. Thus, employment effects may be stronger in contestable markets and weaker in concentrated ones.

The market structure influences the share of static surplus, which goes to the innovator, customers, or the upstream technology supplier. The pattern of adjustment makes interpretation less straightforward. In the short run, firms frequently implement AI within current procedures. Employment contracts, uncertainty regarding precision, legal standards and organizational inertia constrain direct substitution. Employees may utilize the time savings to clear workloads, generate more exhaustive outputs, or participate in more roles. In the long run, firms can develop procedures, alter occupational structures and trim hiring in exposed positions. The lack of sizable employment impacts in the initial years of chatbot installation argues against a swift implosion, but does not prove it to be settled in the long term.

Entry-level jobs are especially critical. Young employees do many codifiable tasks from which they may build experience and tacit knowledge. AI can be a complement to inexperienced workers, guiding their work immediately and therefore at a lower cost to firms than on-the-job training, enabling its use to a greater range of labor market entrants.^[15] It can also cut back on the number of apprentices needed; experienced workers get their routine work done faster. The fiscal implications are even more far-reaching: weaker entry routes diminish lifetime income, revenue taxes and the future stock of skilled workers.

Creation of new tasks is the key to the long-term saturation of automation as a force. Adoption of artificial intelligence (AI) gives rise to work in the areas of workflow design, model interpretation and evaluation, data governance, security, compliance, customer interaction and specialized applications.^[16] It may also render feasible services that were previously unprofitable, broadening appeal for human interaction and cognition. However, it will not necessarily draw forth labor in equivalent quantity, geography, remuneration and output access. A handful of top-tier specialists can be present alongside disappearing labor in routine intellectual work. The crucial issue is whether the new and expanded activities generate adequate labor demand to sustain the overall wage pool as a share of output. Thus, the available evidence tends to accept a conditional proposition. Output expansion will offset displacement where demand is elastic, competitive pressure transmits cost savings onto consumers, complementary stimulating investment exists and workers are able to move into less exposed or newly created tasks. Replacement will dominate rather than complement where demand is saturated, production is standardized, output cannot expand and adoption is mainly focused on cost reduction. There is no support for either universal complementarity or universal replacement.

3. Are Productivity Gains Passed Through as Wages, Retained as Profits, or Transferred to Technology Suppliers?

A productivity gain creates an economic surplus without indicating the ultimate beneficiary. Where an AI-enabled employee takes four hours for work that previously required eight, the saving may turn into a larger payroll, a lower customer price, a larger corporate margin, extra output, a technology-service payment, or a higher shareholding valuation. Its distribution hinges on labor bargaining, the degree of competition, the ownership structure, buyer power and the supply of complementarities. For the state, the ultimate distribution matters as much as the gross increase since each form of income hits the tax machinery in distinct ways. Workers gain where AI increases their marginal productivity and labor-market institutions convert those gains into income.

Customer-support evidence suggests that AI can transmit expertise and improve the output of inexperienced workers, perhaps expanding access to productive employment.^[17] The EIB firm research finds higher wages among AI adopters, although some of the association can be explained by traits of adopting firms and their personnel. By contrast, Danish administrative data have no measurable average earnings increase in the initial two years of AI adoption, with no measurable change in recorded hours.^[18] Overall, the evidence suggests wage pass-through is feasible but neither instantaneous nor assured. Bargaining power decides whether productivity results in pay. Workers receive a larger share where complementary skills are scarce, job opportunities are high, collective bargaining extends to innovation and it's still performance-related pay. Firms win a bigger slice where workers can't monitor improvement in productivity, job opportunities are limited, or AI frees up labor. The very same technology can thus both tighten performance pay gaps and widen internal income gaps of occupational groups. The worst-performing workers may get more productive, but at the same time, engineers, domain experts and managers who are able to put AI to work may take the lion's share of wage premiums.

Consumers realize benefits through lower prices, quicker delivery of services, a wider product range, or better service quality. Consumer returns can therefore lead to an increase in real living standards despite static nominal income. Consumer surplus, however, is not directly taxed.^[19] It enters into revenue when lower prices release income for other taxable spending, or if improved services lead to a rise in the volume of taxable transactions. The fiscal capture from a large consumer return can then be weak. Adopting firms keep the gain if competition is sparse, or prices are sticky, or AI is embraced, which makes internal processes more effective, whose gains are not readily observed by customers. Retained gains might show up as higher operating margins, cash flow, or investment in further intangibles. Direct taxation on the domestic taxable profit may grow if the gain is implemented in an approximate proportion within the economy. However, the overall productivity advantage and once-off domestic taxable gain are not ipso facto the same. Businesses need to pay for the supply of models, software, cloud services and consulting and integration, as well as complementary capital. These costs imply that the size of the surplus left to the adopter will be conditional on the expenditures made by the adopter.

Technology suppliers form their own claims. There are many forms of purchase. AI is often bought on a subscription. It is bought on a usage-based contract or a license through cloud services, enterprise software and licensing. A local provider may enjoy having a higher revenue per employee, at the same time as it gives away most of the added value to an upstream supplier. If this supplier is foreign, the domestically paid sum reduces domestic taxable profit and generates supplier income abroad.^[20] The country can still benefit in terms of wages, consumer surplus and, as long as downstream competitiveness exists, the figures used to measure national productivity can exaggerate the part of the surplus that can exist in the corporate tax structure. Supplier capture is affected by market structure. Convergence of models and complementary services in a commoditized market could hold down prices, shifting more of the surplus to adopters and consumers. While economies of scale, proprietary data, switching costs and distribution premiums tend to reinforce market power in favor of technology suppliers, early adopters might pay a premium for the limited services available in a nascent market and, as the phenomenon diffuses, advantages accruing to technology suppliers might diminish.

Shareholders are paid the residual value of expected future profits. Gains from AI may be capitalized into equity prices prior to accounting income or dividends. The taxation of this value will be very different from payroll. Wages are normally taxed on an ongoing basis by withholding and social contributions. Capital gains can be taxed on a realization basis, at a lower rate, in a different jurisdiction, or not at all in some institutional or tax-privileged arrangements.^[21] A productivity improvement can therefore increase private wealth without delivering an equivalent current flow of revenues domestically. AI-complementary workers in short supply form an intermediate category. Their compensation is legally labor income, although part may reflect scarcity rents. The key fiscal inference is that governments do not tax productivity as an abstract concept. They tax the wages, profits, dividends, realized gains and consumption through which the surplus is allocated.

4. How Might AI Shift the Labour Share?

AI share is not an officially established 'national-accounts' concept; it's a useful heuristic. It refers to that part of value added^[22] that flows to AI-related capital, intellectual property, technology suppliers and economic rents after the extraction of labor compensation. Labor share is defined as employee compensation (sometimes with the addition of an adjustment for the labor share of self‑employment), as a share of value added. The research issue is whether and, if so, where and how AI influences this share. Five mechanisms alter the direction. Augmentation creates more output per worker and maintains or increases labor share if the benefit occurs alongside proportional wage increases. Substitution decreases the need for labor, exerting downward demand pull on the share. Scale expansion may restore labor demand if the lower costs induce enough additional output. The creation of new tasks supports labor income, though the creation of activities where the worker maintains the comparative advantage. Concentration of rent enables either the investor or the upstream supplier to convert its creation of productivity into profit. It turns on the relative power of the mechanisms, not the technical exposure.

There is as yet no reliable empirical measure indicating that generative AI has brought down the economy-wide labor share by a quantifiable observed amount. Adoption is recent and measurement remains inconsistent and labor shares are subject to sectoral composition and business cycle effects, housing income, self-employment and transnational accounts. The global share of labor income had already begun heading down even before the current generative systems became widespread in their use. The ILO estimates that the global labor-income share fell from 52.9 percent in 2019 to 52.3 percent in 2022 and remained broadly unchanged through 2024,^[23] but this longer-term trend cannot be traced to generative AI; it merely shows that labor entered the AI era without an irrefutable claim on productivity. Model-based analysis provides a cautious indicator of what scale we might expect: Acemoglu's upper-bound scenario assumes a small increase in the share of capital and a fall in labor's share, assuming that the impact of AI on output exceeds the growth of wages in the economy.^[24] The resulting shift is comparable to small fractions of a percentage point rather than the sharp decline as postulated at times in the public discussion.

Labor-share pressures are likely to be more visible in particular sectors than in the aggregate. Administrative support jobs, hidden in well-often concentrated in protected sectors, have generative-AI exposure and may be more easily measured and improved than other work tasks, as they involve repetitive tasks, which are relatively cheap to automate.^[25] For example, simple computing tasks like data entry, calendar planning, routine filing, accounting and copying may be affected proportionally. As the demand for this underlying service is likely to be fixed in each task, the source of the decline in employment in administrative jobs is likely to be primarily reduced pay bills, rather than increased output. They are one of the clearest early signs of a decline in labor share. Finance, insurance, accounting and legal services are somewhat more mixed. Text, data and pattern-recognition tasks generate far more exposure, yet regulation, fiduciary duties, model risk and client confidence preserve important human roles. Jobs may stay relatively stable while importance is added and profit-per-partnership or shareholder increases faster than costs. Labor share thus can fall without mass layoffs. Within a business, junior roles may shrink while senior staff and rare technical professionals benefit from extra income.

Public services need a different measure. If a tax office, hospital, school, or local authority handles many more cases with the same number of staff, there's a chance that measured labor share will hardly fall, because output in the non-market is often based on input prices.^[26] Productivity improvement might take the form of fewer queues, better quality or efficiency, not higher 'market' profit. Standards of national labor-share figures could underestimate salient fluctuations in productivity in the public sector. The other factor is firm heterogeneity. As shown by the EIB evidence, productivity growth is heavily biased towards medium and large data, capital, software and skilled- worker intensive adopters.^[27] Such productivity growth is potentially accompanied by market share shifts away from less productive small companies and a resulting reduction in aggregate labor share because the less productive small companies are more payroll-intensive than they are value-added-intensive. On the other hand, new, less costly AI tools may remove at least some competition barriers for small companies, the self-employed and independent professionals, thus avoiding income concentration.

The falling labor share does not necessarily mean real wages are falling. Wages may rise with a smaller share of a faster-growing total. Consumers may gain via lower prices. The fiscal worry is that governments tax wages/employment contributions sooner and more smoothly than retained earnings, unrealized capital gains, or border-crossing supply of income-that is, a small shift in factor shares may lead to a much larger shift in revenue composition, destination and timing. The current evidence belies any assertion of a sharp collapse in the labor share, as well as the presumption that distributional impacts are minor. Nor can any aggregate shift caused by generative AI be established, but significant reallocations at the firm and sector levels seem feasible and could be underway already. The key variables are payrolls as a share of value added; wage growth relative to labor productivity; entry-level employment in new firms; domestic operating profits; payments for imported technology services; and the generation of new labor-intensive jobs. Exposure indices simply cannot determine the extent to which the labor share will decline.

5. Are the Resulting Wages and Profits Taxable Where Affected Workers and Customers Reside?

The most direct channel is personal income tax. If AI raises wages and employment, then taxable labor income increases. If firms cut hours, cut hiring, or automate, there are larger cutbacks in the base. Since top personal income tax schedules are progressive, distributional effects matter. Substantial redistributions to a small number of highly skilled specialists might generate large revenue, but may be offset by large layoffs among middle-income employees when allowances, thresholds and behavioral responses are taken into account. Employee and employer social contributions are especially vulnerable as they are formally linked directly to pay and used to pay for pensions, health care, unemployment insurance and other benefits.^[28] The bottom line for contribution bases can be eroded even if aggregate national income rises if compensation is shifted from wages to profits. Contribution ceilings may aggravate the problem by providing tax shelter above a cap for highly paid workers without generating any further revenue. A broad-based increase in middle-income wages is monetarily different from a similar total increase concentrated among a handful. Payroll taxes are subject to the same logic. They are mostly tied to the place where the worker works and collected on an ongoing basis, so that they are, in effect, more stable and less mobile.

Corporate profits and capital income are more concentrated, more volatile and sometimes more globally mobile.^[29] A government may indeed hold on to total revenue during an initial AI-driven expansion while substituting a large labor base for a narrow profit base and its distinctive cyclicality. When the lost payroll adopters keep the productivity gain as a domestically taxable profit, their corporate-income-tax receipts can offset the losses in payroll. Reduced labor costs, enhanced output and improved margins all raise the corporate obligations. But this offset is not automatic; taxable profit depends on the ways a firm can manage its deductions, losses, financing, depreciation, investment incentives and where it claims income. A few dominant corporations may dominate corporate receipts, a source of increased volatility and political liability. It also needs to be separated from payments to suppliers of the technology. Even a domestic firm can have had higher productivity in the home market but paid huge sums for foreign models, cloud services, software, integration, or IP, which eat into the profit that remains at home. The adopting country can continue to gain from wages, lower prices and higher competitiveness, but gross productivity growth will overstate the corporate income tax that can be levied by the government when so much has been exported to foreign suppliers.

Dividends and capital gains also offer another revenue stream. Increasing profits can be shared directly as dividends, or capital gains can increase to the extent investors anticipate future profits. Capture through capital gains is a function of the investor's residence and ownership, realization rules, exemptions and timing. Capital gains can be unrealized for years, can be expatriated, or can be held through pension funds and other tax-privileged vehicles and the requisite future tax receivable embedded in an equity valuation cannot fund today's unemployment benefits, retraining, or pensions in exactly the same manner as a monthly payroll withholding. The consumption tax only indirectly takes some of the gain. Rising real wages, dividends, or corporate profits may enable more household expenditures. Falling prices can raise real buying power, freeing income for other taxed purchases. However, surpluses are not themselves taxed. If AI lowers a taxed service price, with a fixed quantity, value-added tax revenue may fall. Spending the savings on a different good may restore overall VAT returns. This effect is driven by observed nominal expenditure, not uplift in consumer welfare. Public expenditure can increase during adjustment. Those displaced from exposed occupations may need unemployment insurance, income security, retraining, job-search assistance and mobility programs. Aggregate employment stability does not remove this cost, since disruption may be absorbed by individual occupations, geographic areas, or ages. An important national effect can have a profound local adjustment. Ineffective retraining, which produces costs without improvement in employment, should also be taken into account.

AI can also boost the public balance sheet through tax administration and public-service efficiency. Against a background of expanding data collection, tax authorities are increasingly harnessing AI and big data-driven systems to improve risk management, limit tax evasion and avoidance, select cases and provide services to taxpayers. OECD evidence indicates that a large majority of surveyed tax administrations by 2023 were either already exploiting or in the process of deploying AI.^[30] Better targeting can drive compliance, increasing revenue and cutting collection costs; automation of routine administration can release officials to concentrate on more complex cases. These improvements will depend on quality data, supervisory human oversight, cybersecurity, legal protection and sophisticated procedures for challenging automated judgments. These benefits should be viewed as a potential fiscal offset, rather than a windfall. The three hypothetical 100 productivity-gain cases do show the importance of incidence. They are stylized accounting cases, not empirical estimates or forecasts. Assume that there is also a 20 percent tax on the additional wages, 10percent on the incremental domestic profit and the consumer-benefit component produces an illustrative VAT effect of 10 percent. The VAT assumption does not reflect the direct taxability of consumer surplus; it is just a simplified increase in taxable expenditure with regard to the benefit.

In Case A, €60 appears as wages, €30 as domestic profit and €10 as consumer benefit. Wage tax yields €12, profit tax €3 and VAT €1, producing total domestic revenue of €16. This is the widest base, widest in the sense that almost all of the gain ends up as domestic labor income. In Case B, €20 goes to wages, €60 to domestic profit and €20 to consumer benefit. The effect of wage taxes is €4, profit taxes €6, VAT effect €2, thus €12 in all. Corporate revenue is higher than in Case A, but not enough to offset the weaker labor base, assuming these tax rates. In the simple example of Case C, €20 goes to wages, €20 to domestic profit, €40 to payments to foreign AI suppliers and €20 to consumer benefit. If in this case we suppose no direct domestic tax collection on the foreign income outside the model, the wage is taxed at €4, the profit is taxed at €2 and the VAT effect is €2, totaling €8 in domestic revenue. The same €100 gross productivity advantage produces only one-half the revenue of Case A in this case because a relatively much greater share appears outside of the model's assumed domestic base.

These scenarios do not prove that AI will cut government revenues by 50 percent. They do not allow for progressive rates, social contributions, deductions, investment, supplier payroll, withholding taxes, 'trade' effects, behavioral responses and deferred taxation of capital gains. They accept the full €100 as available surplus, although real adoption implies complementary spending. Their purpose is narrower: calculating fiscal projections on the basis of GDP or productivity alone can be highly misleading when the distributional and 'jurisdictional' aspects of the gain are neglected. Current tax arrangements are therefore relevant to this composition problem. Personal-income taxes and social-security contributions combined constitute a significantly larger proportion of total revenues than do corporate-income taxes throughout the OECD.^[31] Within the EU, taxes on labor (including social contributions) constitute roughly half of total tax revenue.^[32] A continuing transfer of revenues from payrolls toward profits or capital gains or to foreign providers may harm the bases upon which social insurance and other public services are paid, even if GDP continues to grow.

Tax treatments might also shape the adoption path. In their model of a sub-optimally high automation rate, Acemoglu, Manera and Restrepo show that the US tax system taxed labor more heavily than capital over the course of the last century, thus incentivizing automation above the socially optimal level.^[33] However, it must be noted that this study deals with automation as a whole, rather than production using generative AI specifically and reaches its conclusion on the basis of a set of assumptions made in the model. Nevertheless, it sends a clear message: tax policies should not unintentionally incentivize labor substitution in an artificially attractive light relative to augmentation, organizational change, or skills acquisition.

Measurement remains a central fiscal institution. Departments of finance, statistical offices, tax authorities and social-insurance agencies require linked data on firm-level adoption, payrolls, occupations, wages, sales, profits, imported services and the taxes they pay. Circumstances where firms have furloughed, increased wages, or transferred the proceeds abroad cannot be identified from aggregate exposure indices alone.^[34] Within sector monitoring must compare observed developments with what technical potential permitted while monitoring flows into entry-level jobs, contribution density, a firm's labor share and intermediate supplier expenditure. Transition policies should be based on actual displacement, not on conjectural estimates of national employment figures. Unemployment insurance and income support are required where displacement takes place and retraining should be related to credible labor demand and assessed by subsequent employment and earnings. Short-term wage insurance may be more appropriate for a few mid-career workers than long classroom courses. Policies should be amplified where verified displacement increases and changed if labor demand resurges. Thus, AI may boost corporate revenue at the expense of wages, increase tax compliance at the expense of adjustment costs and improve consumer well-being at the expense of rent-sharing. The tax problem is not merely a single change in the base but one in the composition, timing, localization and jurisdiction of revenue.

6. Which Fiscal Structures Are Most Exposed?

The highest fiscal risk occurs where a number of vulnerabilities combine. High dependence on labor taxes alone is inadequate because employment and real wages could increase if augmentation succeeds. High occupational vulnerability alone will also be inadequate, because exposed jobs can be converted into other types of occupations. The greatest danger exists if all six of the following factors reinforce one another: high dependence on labor taxes; available substitutions of cognitively based employment; weak capital-income tax payments; low domestic ownership of capital stock; payments to foreign producers of technology; and rising expenditure pressure. Employers are hit directly by social-contribution systems as they fund them through payroll. The share of social-security contributions in total tax revenue in 2023 was over 40 percent in Czechia, Slovenia and Slovakia.^[35] While these economies might not be the most exposed to generative AI in employment terms, especially where manufacturing is still relevant, the impact in fiscal terms is clear. A shift in the value-added composition from payrolls to profits or imported services undermines the funding base of pensions and social-insurance systems directly.

An analogous situation exists among countries exhibiting high tax wedges on formal employment. Belgium, Germany, France and Austria all register some of the highest totals of personal and social contributions borne by the typical worker of the OECD.^[36] Again, a high tax wedge on employment does not necessarily point to a net employment effect of AI; it simply emphasizes that every euro taken away from the wage bill requires the outflow of much profit and capital income. If these are taxed less than current income forms, the tax effect of diversion increases. A similar sort of risk is represented by countries whose reliance on personal-income taxes, as opposed to social contributions, is especially pronounced. Denmark is just such an example, with significant levels of government revenues coming from the former while contributing less to the latter. Its highly digitized labor market entails significant technical exposure, but high levels of general consumption taxation, an efficient administrative bureaucracy and effective labor-market institutions are compensating factors. This sheds light on just why exposure should not be ranked within only a single tax category.

The highest amount of occupational exposure is likely to exist in service-intensive high-income economies. According to the ILO, 34 percent of employment in high-income states has at least some generative-AI exposure, relative to 11 percent in low-income states. In these economies, there is a fairly large prevalence of clerical, financial, professional, technical and administrative jobs, many of which have relatively high wages, meaning these economies could garner the most benefit from productivity but could also see the most significant movement away from broad labor compensation if rent concentration and substitution take hold.

Another distinct vulnerability arises if domestic ownership is weak. While a country can quickly utilize the AI to boost productivity and consumer benefits , payments for technology will be made to foreign owners, which will shrink the domestic operating surplus and needed capital income base and enhance "ownership income" in the homeland less than what would be reflected in the productivity figures. Firms, cloud providers, software and other intellectual properties have mostly foreign owners, though the business and capital stocks could grow less than the productivity data would indicate. The IMF scenario estimates help distinguish economies that produce and own important AI assets from those that only import these services, while acknowledging the scenario's nature.^[37] Ownership should not be conflated with the physical location of the infrastructure. A state can host infrastructure and not be the owner of the most sophisticated models, software, or intangible assets and a domestic firm can own artificial intelligence assets while contracting for infrastructure services outside. In terms of fiscal incidence, the key variables are the recipient of the income, the contractual form of the transaction and the jurisdiction of the taxation of profit or capital income. An infrastructure nexus so detailed belongs to a different tax question.

In countries with only weak effective tax on capital and profits, exposure increases even with AI rents remaining within the economy. Changing from wages to retained profits, dividend income, or capital gain tax will lessen revenue where that source is either exempt, deferred or weakly enforced; and statutory corporate tax rates will be just half of the factor. Tax expenditures and old-style profits-for-years-with-no-tax arrangements, with institutional ownership and realization-based capital gains provisions, alter collection. Highly profitable economies can be susceptible in this way and multinational nodes can generate significant revenues through large, highly concentrated sets of firms, thus allowing profit taxes to offset a feeble labor force. However, reliance on such a limited and heavily internationalized pool is itself subject to fluctuations in the decisions of multinational enterprises. Ireland and Luxembourg should not be treated simply as weak-profit-tax jurisdictions. Their potential exposure to risk might stem from concentration and mobility rather than weakness.

All revenue sources are made more susceptible by aging and debt. In countries with growing pension, health care and long-term-care costs, a decline in contribution income in itself or an erratic tax portfolio will have harsher revenue consequences than the same or smaller downturn in payrolls for a more youthful, less indebted economy. The European Commission's aging projections are built-in scenarios based on what-if demographic and policy assumptions, but they clearly demonstrate the cost squeeze European countries face in future decades.^[38] A small permanent weakening of payrolls in such a macroeconomic environment could be much more harmful than a temporary large disturbance to a younger population in a more lightly indebted country.

Cross-country fiscal evaluation, consequently, must involve many indicators. Countries should assess their reliance on labor taxation and social contributions, the skill composition of the formal sector, the true take-up and labor market responses, the effective taxation of profits and capital income, payments to foreign technology providers, domestic ownership of productive assets and the medium-term budgetary and debt constraints. There is no one measure of exposure. The existence of national stress tests can incorporate alternative distributions of incidence without assuming the ability to predict technological advances: in an all-wage scenario, an augmentation model may take full wage pass-through; in a high-corporate-profit model, a declining payroll and rising domestic corporate income model may be applied; in an imported-rent model, a significant share of suppliers receiving international payments can be incorporated. Each should apply the appropriate personal tax schedule, contribution system, effective corporate rates, consumption taxes and benefit expenditure schedule and be cast as conditional ranges that should be refined iteratively as administrative data improves.

Policy should match the diagnosed vulnerability. Payroll-dependent countries should monitor wage pass-through, contribution density and entry-level employment; weak capital income taxation requires broader effective capture, rather than a special AI levy; AI-service importers require better data on technology payments and a greater share of home value added. Heavily indebted and aging relies on contingency planning for adjustment costs and revenue volatility. Tax administrations can benefit from AI for ensuring compliance, while not compromising due process, transparency and human accountability. Pretax fiscal incidence is also affected by the provision of labor-market institutions. Institutions such as training, portable social insurance, efficient job matching, collective bargaining and competition can provide augmenting and scale-expanding effects. They do not just serve as reinforcing improvements after the shrinking revenue base has occurred. Instead, they influence whether higher productivity is revealed in increased wages, job creation, consumers’ gains or concentrated rents. The most durable fiscal cushion will not result from heavier taxation of a declining wage bill, but from a healthy economy where rising productivity is reflected in and taxed from widely spread, domestically generated income. Countries most exposed are those in which the gains will be incorporated into forms that the domestic fiscal system captures weakly, slowly, or not at all, while public obligations remain tied to employment, aging and social insurance.

7. Conclusion - When Does AI Expand Rather Than Erode the Tax Base?

Artificial intelligence does not mechanically erode the tax base. It shifts the tasks that produce income and can enhance productivity, wages, corporate profits, aggregate consumption and fiscal administration. Preliminary evidence suggests significant productivity gains in selected tasks and significant workplace exposure, but not a general job-shedding in the economy. The strategic risk is a divergence of productive capacity relative to domestic fiscal capture. An economy may gain productivity while shifting a broad, immediately taxed payroll base toward concentrated profits, deferred capital gains and revenue from foreign suppliers. High wage pass-through, competitive diffusion, new-task creation and growth of domestic enterprise can produce the opposite outcome as well. Governments must avoid complacency on the one hand and destructive technology-specific taxes on the other and should base intervention on measured rather than speculative displacement. Fiscal resilience will depend less on headline productivity improvement than on the extent to which the income generated by that productivity is distributed broadly and locally and ultimately able to be taxed.

This article was prepared as an independent research contribution following the conference Inequalities in Longevity, held at Fondazione Giorgio Cini in Venice on 3–4 July 2026. It develops themes presented and discussed at the conference but is separate from the official conference proceedings.

Unless expressly stated otherwise, this publication has not been commissioned or endorsed by Fondazione Giorgio Cini. The analysis, interpretations, and conclusions are those of the author(s) and do not necessarily reflect the official positions of Fondazione Giorgio Cini, Swiss Institute of Artificial Intelligence (SIAI), or their respective affiliates.

References

[1] Noy, S. and Zhang, W. (2023) ‘Experimental evidence on the productivity effects of generative artificial intelligence’, Science, 381(6654), pp. 187–192.

[2, 15, 17] Brynjolfsson, E., Li, D. and Raymond, L.R. (2023) Generative AI at Work. NBER Working Paper No. 31161. Cambridge, MA: National Bureau of Economic Research.

[3] Dell’Acqua, F., McFowland III, E., Mollick, E.R., Lifshitz-Assaf, H., Kellogg, K.C., Rajendran, S., Krayer, L., Candelon, F. and Lakhani, K.R. (2023) Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge-Worker Productivity and Quality. Harvard Business School Working Paper No. 24-013. Boston, MA: Harvard Business School.

[4, 5, 6, 25] Gmyrek, P., Berg, J., Kamiński, K., Konopczyński, F., Ładna, A., Náfrádi, B., Rosłaniec, K. and Troszyński, M. (2025) Generative AI and Jobs: A Refined Global Index of Occupational Exposure. ILO Working Paper 140. Geneva: International Labour Organization.

[7, 24] Acemoglu, D. (2025) ‘The simple macroeconomics of AI’, Economic Policy, 40(121), pp. 13–58.

[8, 18] Humlum, A. and Vestergaard, E. (2025) Large Language Models, Small Labor Market Effects. NBER Working Paper No. 33777. Cambridge, MA: National Bureau of Economic Research.

[9, 13, 14, 34] Hampole, M., Papanikolaou, D., Schmidt, L.D.W. and Seegmiller, B. (2025) Artificial Intelligence and the Labor Market. NBER Working Paper No. 33509. Cambridge, MA: National Bureau of Economic Research.

[10] Autor, D.H. (2015) ‘Why are there still so many jobs? The history and future of workplace automation’, Journal of Economic Perspectives, 29(3), pp. 3–30.

[11, 12, 27] European Investment Bank (2026) EIB Group Investment Survey 2025/2026. Luxembourg: European Investment Bank.

[16] Acemoglu, D. and Restrepo, P. (2019) ‘Artificial intelligence, automation and work’, in Agrawal, A., Gans, J. and Goldfarb, A. (eds) The Economics of Artificial Intelligence: An Agenda. Chicago: University of Chicago Press, pp. 197–236.

[19] Brynjolfsson, E., Collis, A. and Eggers, F. (2019) ‘Using massive online choice experiments to measure changes in well-being’, Proceedings of the National Academy of Sciences, 116(15), pp. 7250–7255.

[20, 21, 28, 29, 37] International Monetary Fund (2024) Broadening the Gains from Generative AI: The Role of Fiscal Policies. Staff Discussion Note 2024/002. Washington, DC: International Monetary Fund.

[22] ILOSTAT (2024) Labour Income Share as a Percent of GDP: ILO Modelled Estimates and Methodological Description. Geneva: International Labour Organization.

[23] International Labour Organization (2024) World Employment and Social Outlook: September 2024 Update. Geneva: International Labour Organization.

[26] European Commission (2013) European System of Accounts: ESA 2010. Luxembourg: Publications Office of the European Union.

[30] Organisation for Economic Co-operation and Development (2024) Tax Administration 2024: Comparative Information on OECD and Other Advanced and Emerging Economies. Paris: OECD Publishing.

[31, 35] Organisation for Economic Co-operation and Development (2025) Revenue Statistics 2025. Paris: OECD Publishing.

[32] European Commission (2026) Data on Taxation Trends. Brussels: Directorate-General for Taxation and Customs Union.

[33] Acemoglu, D., Manera, A. and Restrepo, P. (2020) ‘Does the US tax code favor automation?’, Brookings Papers on Economic Activity, 2020(1), pp. 231–300.

[36] Organisation for Economic Co-operation and Development (2026) Taxing Wages 2026. Paris: OECD Publishing.

[38] European Commission (2024) 2024 Ageing Report: Economic and Budgetary Projections for the EU Member States, 2022–2070. European Economy Institutional Paper 279. Brussels: Directorate-General for Economic and Financial Affairs.