TAI’s Potential Impact on the Tax Base
As TAI systems transition into the mainstream economy, they threaten to undermine a fundamental premise upon which modern government finance rests: the taxation of human labor.
For most of recent history, human labor has served as the backbone of the economy. Around the globe, labor receives around 50-70% of total national income (labor share of GDP) in most countries. As a result, labor-based income since the 20th century has been the principal source of revenue for the government in developed countries. With few rare exceptions, income taxes and social security contributions make up more than 50% of federal revenue in most developed countries. Corporate taxes, by comparison, account for a mere 12%¹.
If the assumptions hold true that transformative AI could eventually replace a majority of human labor with automated systems, we could see major upheavals in this status quo. We’ll discuss two key implications:
Transformative AI could lead to an unprecedented shift from labor to capital, which could drastically shift or reduce the overall tax base.
Much of the economic discussion around TAI has been around its potential impact on the labor share of income. Proponents of near-term TAI are concerned that a rapid shift of many cognitive labor roles to automated systems will lead to an unprecedented drop in the labor share, which would be captured instead by capital (via higher corporate profits and capital gains). This could have a massive impact on national tax bases, because of the relative effectiveness of capturing labor vs. capital income.
The average labor tax rate across the OECD is 34.9% – in other words, the percentage of the income of an average worker taxed via personal income, payroll, or employee and employer social security². In comparison, the corresponding average corporate tax is 20.2%³ – but that’s solely on profits, not on revenue. That means the difference is even greater than (34.9 - 20.2) = 14.7% implied by the overall tax rates.
Let’s break it down with a simplified hypothetical example.⁴ Consider a cognitive worker earning $100,000 a year in an average OECD country. Currently, they (along with their employer) would owe $34,900 in taxes. Now, imagine their role is perfectly replaced by a powerful AI system creating the same $100,000 in value. If that value was 100% profit, the corporation owning the AI system would pay $20,200 in taxes: already, a 42% decrease in tax revenue. But in practice, this could be even less, because labour income is usually taxed on a gross basis whereas capital income is usually taxed on a net basis.
Of course, this is just an illustrative and reductive example. AI systems will not replace human labor in a 1:1 fashion. Skyrocketing growth could counteract tax base challenges. Many other factors must be considered – we recommend looking at more detailed economic modeling on this topic.
Transformative AI could lead to a massive increase in the concentration of profits and economic growth among the largest AI-driven multinational enterprises.
One of the core implications of TAI is that by separating firms from their historical reliance on human labor, AI-driven corporations should be able to rapidly grow and outcompete their more human-driven counterparts on speed of execution, prices, and more. These corporations may also see massive economies of scale and first-mover advantages.
Waymo is a titular example: by no longer needing to pay humans, it will soon become cheaper, more available, and more reliable than other taxi services. By solving the operational pipelines of large-scale manufacturing, reliability, and building regulatory trust across thousands of jurisdictions, it may eventually become the leader of only a handful of automated corporations replacing tens of millions of human drivers.
One might expect that the majority of future economic growth across each sector could accrue to a similarly small class of “superstar firms”. There is some evidence we are already seeing a preliminary version of this growth concentration in today’s “AI bubble”.
The first tax challenge this exacerbates is that the largest digital multinational enterprises (MNEs) are still being taxed substantially below the corporate tax rate intended by most nations, because of poor global tax cooperation and tax loopholes that these corporations are well equipped to bypass.
About a decade ago, this problem was actually much worse. For a period of time, corporations like Apple were able to get away with paying essentially zero corporate taxes on billions in profits, via clever loopholes such as the “Double Irish”. Today, Apple and Google both are paying roughly 15.6% in effective corporate tax rates globally – below the 20.2% average headline rate, but significantly more than before.
The second tax challenge this exacerbates is that the distribution of these taxes on digital MNEs is often largely skewed towards a small subset of countries. For example, Google pays over 80% of its corporate taxes in the US. Some may consider this fair, as Google is a US corporation. In practice, it means that Google contributes vanishingly little in corporate tax revenue to most other countries.⁵
What this means is that highly automated, multinational “superstar firms” would cause even greater havoc on a global taxation regime already struggling to properly account for today’s digital MNEs. AI-driven powerhouses like OpenAI may similarly contribute only 15% in corporate profit taxes, largely to the US government at the exclusion of other countries. This could be catastrophic in scenarios where such “superstar firms” capture an ever-increasing proportion of global economic growth, while only needing thousands of employees in a few AI-leading countries.
In summary – the vast majority of countries should be concerned about the potential of falling tax revenues in a world with transformative AI. A shift of economic value from human labor to capital could reduce global tax revenues. A concentration of profits among the largest highly-automated MNEs would further reduce corporate tax revenues, as well as concentrating taxable profits in only a few AI-leading countries. It’s a nuanced topic – but we think it is clear that national tax experts should be considering the possibility of transformative AI economic scenarios in their long-term tax modeling.
One Solution: A Progressive Corporate Tax
There are a wide variety of plausible taxation responses to the challenges described above. In this essay, we’ll propose just one for thought. A well-enforced progressive corporate profit tax would directly counteract tax base erosion from both a shift from labor to capital, and a concentration of profits among the largest MNEs.
Today, most national corporate taxes are flat: nearly all companies within a jurisdiction pay roughly the same percentage of their profits, no matter their size.⁶ A progressive corporate profit tax simply means that the largest and most profitable firms should pay a higher rate of taxes on their profits. If OpenAI or a similar firm earns $100 billion in profits, it could presumably pay 30+% in profit taxes, as compared to a mom-and-pop business paying the default 20%.
Such a progressive corporate tax would directly counteract the two primary tax base challenges described above: a shift from labor to capital, and a concentration of profits among the largest MNEs.
It could do this via the following two effects:
It could increase the overall rate of taxation on profits derived from capital rather than labor, potentially restabilizing tax bases worldwide and improving tax incidence.
It could increase the proportion of taxes paid by the largest and most profitable corporations, while not impacting smaller or less profitable firms. The vast majority of firms and workers would be unaffected by a well-designed progressive corporate tax.
In short, such a policy is conceptually well designed for exactly the economic scenario we are describing above due to TAI.
This proposal aligns with the strategy of capturing economic rents, which has broad support among economists. It is theoretically most efficient to tax supernormal profits, or profits beyond those useful to continue operations or reinvest in further growth. In TAI scenarios, superstar firms would likely be capturing massive amounts of economic rents, as they generate profits well beyond normal returns that accrue to a smaller and smaller subset of people.
Philosophically, this also has solid underpinnings. Most would agree that the largest beneficiaries from AI should contribute proportionally more to the societies that enabled their success. The public investments, collective human knowledge, and regulatory environments that power AI development should receive fair compensation.
Drawbacks of a Progressive Corporate Tax
What would be the conceptual drawbacks of a progressive corporate tax, as compared to the status quo (flat, low corporate taxes) in most jurisdictions today?
There are a variety of valid responses:
Flat corporate taxes incentivize less profit-shifting. Because corporations and capital are highly mobile, many countries are concerned that complex corporate tax laws could dissuade local investment, or that corporations would simply bypass them by shifting profits offshore.
We’ll discuss solutions to this in the following chapter.
Economists have historically viewed low corporate tax rates and higher personal income taxes to be less distortionary. Corporate taxes are often maligned for creating "double taxation" – profits are taxed once at the corporate level, then again when distributed to shareholders. Excessive distortions could create “deadweight losses”.
Despite these criticisms, nearly all economists are aligned that some nonzero level of corporate taxation is efficient and necessary, for reasons beyond the scope of this piece.
Progressive corporate taxation could encourage tax fragmentation. Corporations might be incentivized to split themselves into smaller legal entities to avoid taxation. We expect this to be a minor concern, as this strategy should primarily target the largest MNEs, which have high public profiles, competing interests, and would certainly face precedent-setting litigation.
All of these criticisms certainly hold value and we acknowledge them. But – they may have reduced relative importance in a world where an ever-increasing proportion of economic growth flows directly to AI capital and bypasses labor.
Finally, it’s crucial to note that this strategy is not a panacea for the tax base challenges surfaced by TAI. A healthy tax base is composed of layers of thoughtful, well designed mechanisms, most of which will need to be rethought for an AI-driven economy. Beyond updating existing mechanisms, even more radical ideas may need to be considered, such as:
Higher consumption taxes (Korinek, Lockwood)
Equity stakes in AI corporations (Yelizarova)
Business wealth taxes (Gamage)
Increasing land-value taxes (Convergence Analysis)
How to Implement a Progressive Corporate Tax
A practical implementation of a global progressive corporate tax would need to build on OECD BEPS 2.0. We believe that BEPS 2.0 could eventually “set the floor” for taxing the most profitable digital MNEs, while allowing individual countries to set their national corporate tax progressivity for less profitable corporations according to domestic priorities.
Summarizing BEPS 2.0
Unfortunately, corporate taxation for the largest MNEs cannot be contained within a single jurisdiction – it is reliant on global geopolitical dynamics. The mobility of legal corporations, competition for taxing rights between countries, and the global nature of most digital enterprises has led to a “race to the bottom” that has characterized the past three decades of global corporate taxation.⁷
In practice, a progressive corporate income tax enacted unilaterally would likely encourage the largest corporations to offshore their profits to jurisdictions with lower corporate taxation. Today, legal defenses against this form of offshoring are improving, but still quite porous. Academic studies estimate that around 40% of global corporate profits are still being shifted offshore⁸. It’s estimated that this costs governments between 100 and 240 billion dollars annually⁹.
Notably, the OECD has already made substantial progress towards the most comprehensive global tax base reform in history: the BEPS 2.0 proposal, which comprises two pillars.¹⁰ In our opinion, BEPS 2.0 is the most thoughtful, effective, and feasible solution for fair global corporate taxation governance today. This perspective is supported by an overwhelming coalition known as the BEPS Inclusive Framework: over 140 countries have previously committed to this reform, including every major economy, the US, and China. However, large portions of BEPS are currently stalled primarily due to issues raised by the US.¹¹
BEPS is a deeply complex topic, so we’ll summarize just the relevant portions for our proposal.
Pillar One targets profit-shifting.
Traditionally, corporate taxes have been tied to the location of a company’s headquarters or physical presence. With the advent of digital corporations, this has increasingly become obsolete: a company like Google could capture billions in ad revenue from France, but as long as it didn’t have a significant physical presence, it could pay France zero in corporate taxes.
For the largest MNEs, Pillar One mandates that a proportion of their residual profits (e.g. “economic rents”, or profit margins > 10%¹²) must be distributed to the country in which it receives revenue. For example, if we had a firm with €30bn revenue and €6bn profit (20% margin):
Residual profit: €30bn × (20% - 10%) = €3bn
Pillar One mandates that a certain rate of the residual profit is reallocated to the market jurisdictions. Currently that rate is at 25%. In our example, the so-called Amount A would be:
Amount A: 25% × €3bn = €750m
If 10% of the revenue came from France, France is entitled to tax 10% of Amount A (i.e. €750m × 10% = €75m).¹³ In essence, countries would be permitted to tax a portion of an MNE’s residual profits based on where the MNE’s customers are based, instead of solely where its profits are booked.
Pillar Two targets tax competition and the race-to-the-bottom.
Pillar Two seeks to ensure that the largest MNEs pay a total effective corporate income tax rate of at least 15%, independent from where they operate. It achieves this through a set of coordinated rules¹⁴ that reassign taxing rights to other jurisdictions when a country chooses not to exercise them, guaranteeing a global 15% minimum tax rate if successfully implemented.
Building on BEPS 2.0
It’s exceedingly clear to us that any plausible global corporate tax proposal must leverages the work previously done by BEPS 2.0. In this piece, we’ll describe an implementation strategy where a future BEPS project (e.g. BEPS 3.0) acts as the top rungs of a global corporate tax ladder.
Our proposal would set the rules and rates for multinational firms at the very top – those with the highest profits and global reach. Each country could then fill in the lower rungs of a progressive tax according to their domestic priorities.
In this approach, the BEPS Inclusive Framework would continue to set the minimum taxation requirements for the most profitable and largest digital MNEs globally, as it already does with Pillar One and Two. However, as economic scenarios shift the Overton window, it would gradually evolve to raise taxation rates on the most profitable global firms.
Eventually, the tax rates on these top performers would surpass the corporate tax rates of individual countries. In practice, this would result in a progressive corporate tax. Smaller, less profitable companies would be taxed according to a local (national level) set of regulations. The larger, profitable global firms would be taxed more, according to rules set globally by the BEPS Inclusive Framework.
Over time, there may be some incentives for nations to create further progressivity in their corporate tax rates in order to align with BEPS and avoid any “cliff-edges”. They might update their tax code so intermediate-sized corporations pay higher rates than their current flat corporate tax rates, but lower rates than the “top rung” as set by the BEPS Inclusive Framework.
Crucially, each nation would retain full autonomy over how to structure tax rates for corporations that fall below the BEPS threshold. This would preserve national sovereignty over the vast majority of domestic businesses and protect local firms, while still forcing the largest multinational corporations to be held accountable on a global scale.
Implementing this strategy would require changes to BEPS that are straightforward in concept but politically challenging to enact:
Policy Modification (Pillar 1): Higher share of residual profit for market jurisdictions
Under the current Pillar One design, only 25% of residual profits is reallocated to market jurisdictions. This is a reasonable and modest baseline – but it may not be enough to respond to TAI scenarios.
We suggest that in a future where economic growth is concentrated to a small set of capital owners globally, a greater share of economic rents may need to be distributed based on proportion of consumption.
We propose that policymakers aligned with BEPS could gradually increase the share of residual profit allocated to market jurisdictions from the most profitable companies. Increasing this rate from 25% to 30%, 40%, or even 50% would better counteract the economic power concentration we expect from TAI and ensure the economic gains from AI are distributed widely.
Policy Modification (Pillar 2): Higher global minimum tax rate
Pillar Two is a landmark in global tax cooperation, but its 15% minimum rate reflects political compromise more than economic logic. It sits well below the OECD average of about 20.2% and beneath the levels in most major economies. As a result, it acts as a global floor – serving as the minimum viable taxation rate for global firms.
In this progressive taxation strategy, we propose inverting this logic on its head. Pillar Two could serve not just as a global floor, but to raise the top rung of the corporate tax ladder – serving as an anchor point for the optimal level of taxation on the largest and most profitable global corporations. Raising this “anchor point” above the default rate of most countries would not only strengthen their revenue bases¹⁵, but also eventually make BEPS a powerful enabler of more progressive corporate taxation at the domestic level.
Addressing Feasibility Concerns
Every tax policy expert who has spent time working with BEPS will have deep skepticism about the plausibility of this proposal, and rightfully so.
In the short term, this iteration of BEPS 2.0 is already stalling out due to non-participation from the Trump administration.
On a high level, competing nationalistic incentives make it extraordinarily difficult to secure global cooperation, especially when negotiations over taxing rights often seem zero-sum in nature.
Proposing higher values on rates that are already the result of political compromises can seem divorced from reality.
To this valid pushback, we’ll offer two thoughts:
If a significant recession appears driven by AI in the next decade, or an economic crisis proportional to the suggested impact of TAI materializes, it is entirely possible that the economic consensus could shift towards more radical proposals.
The vision of a strong, well-enforced global tax system that fairly captures the economic rents from massively profitable AI corporations may seem beyond the pale today. But – we believe society can get there, and that the conditions may emerge for us to collectively design the right future.
Let us navigate this upcoming transition with thoughtfully designed policies, strong global coordination, and a commitment to economic prosperity for all.
