Shifting a million AI remote workers to a tax haven
AGI economy series - Nr. 11
Tax avoidance means using legal structures, regulations, and loopholes to reduce the amount of taxes owed. Tax avoidance is prevalent among large digital companies. Digital businesses can often pay an effective tax rate of less than half that of traditional businesses. A key reason for this is that the key assets of digital businesses, like intellectual property, data, and intangible services, can be more easily shifted to subsidiaries in low-tax countries for tax purposes than physical goods.
In some scenarios, a few AGI firms might capture much of the future wealth from advanced AI. If the top tech companies end up commanding a “country of geniuses in a datacenter” that might make the future tax avoidance challenge more difficult. As discussed in the post on the Windfall Trust, corporate philanthropy may be part of the solution but it cannot replace taxes.
In this post, we will look at the Base Erosion and Profit Shifting (BEPS) project by the group of advanced market economies OECD. The goal of the project is to close gaps in international tax rules to ensure companies pay taxes where they have real economic activities. I think that the resulting treaty of this process could be of significant relevance to AGI. Yet, neither the tax lawyers nor the AI governance people seem aware of this. So, let’s try to bridge this gap.1
1. Why does corporate tax governance matter for AGI?
a) The relative importance of corporate taxes could increase
As discussed before the labor share of income may be significantly lower in the future, and this would impact payroll tax and income tax.2 From a US tax perspective, this means the two largest pillars of US tax revenues are weakened, increasing the relative importance of corporate income tax. However, if we actually look at a revenue chart of the US government we can see that corporate income tax as a share of government revenue is near a historic low.
b) Without coordination there is a race to the bottom in corporate taxes
In recent decades US corporate income taxes have faced downward pressures. On the one hand, there is a positive feedback loop between corporate lobbying and corporate income tax reductions. On the other hand, multinational enterprises have found ways to artificially shift their profits to low or no-tax locations, even though they have little or no economic activity there. This creates a "race to the bottom," where especially small countries with a small taxbase competitively lowered their corporate tax rates to attract foreign businesses, resulting in very low effective tax rates for multinational enterprises globally.
c) Cross-border business-to-consumer services could increase
Traditionally, companies are taxed where they have a physical presence (like offices or factories). However, many digital companies can offer Internet-based services without any physical presence there. This has led to concerns that these companies aren’t paying their fair share of taxes in the countries where their customers are.
Specifically, it is hard to determine where the service is consumed and which country should collect value added tax (VAT). Tax fraud is common in this category because sellers and buyers can easily hide or misrepresent their locations to reduce VAT, and it's challenging for tax authorities to monitor all these small, intangible transactions. Today, we mostly live in a service economy, there is no information loss to transmit data from a datacenter over large distances, there are big local differences in electricity costs for datacenters, and most services are latency-insensitive compared to transport at the speed of light (see also “local vs. global market”)
Hence, if AGI can perform a lot of service jobs, such as business consulting, personal training, education etc., as a ‘drop-in remote worker’ the importance of cross-border B2C services and cross-border B2B services is likely to increase (see also “switch from in-house capacity to an outsourced service?”). This is also relevant for ensuring that developing countries are not completely left behind in an AGI take-off.
d) AGIs could gain legal personhood through corporations
This is more speculative, but as I have argued before:
future AI agents can make their own money.
through money, future AI agents can control some types of corporations.
through ownership of a corporation, future AI agents can gain legal personhood.
Hence, AGI agents could theoretically own themselves and other AI-systems; own corporate bank accounts, stocks of other countries, patents, datasets, AI hardware, electricity infrastructure, buildings etc.; buy land, production equipment, AI hardware, electricity infrastructure, etc.; create as many copies of themselves as they have hardware access to; legally hire other firms and humans for different roles and tasks; earn money through investments and businesses; sue and protect its rights in courts.
A robot tax that taxes systems owned by a company is not very practical. However, AGI does not have to be the ‘economic equivalent of slave labor’. If an AGI owns a company, it is not just fed enough electricity to run, it also earns the economic surplus generated and in that case it should pay taxes. However, since it lacks a natural personhood there is no existing framework for it to pay income taxes. In contrast, it seems likely that such AGIs will have a legal personhood as a corporation. Hence, they can pay corporate taxes.
2. The BEPS project
Base erosion means reducing taxable income within a high-tax country by inflating deductions or expenses. Profit shifting means moving actual profits from one country to another, typically into a low-tax country or tax haven.
In July 2021, more than 130 countries reached a historic agreement on a 15% global minimum tax rate and a commitment to reallocate some taxing rights from multinational enterprises to countries where they have significant consumer bases. This will be implemented through both a multilateral convention as well as domestic policy.3
a) Residual taxing rights for consumer markets (“Pillar One”)
The new rules allow consumer markets to tax a small portion of the profits of large multinational enterprises without requiring a physical presence. Only multinational enterprises with more than 20 billion EUR revenue and a profit margin above 10% are affected. This applies to ca. 100 multinational enterprises.4 From these, consumer markets can collect corporate income taxes on 25% of “residual profits”, defined as profit exceeding 10% of revenue from multinational enterprises. If residual profits are not attributable to any jurisdiction, a part of it could be directed into a global pool or a special allocation mechanism, from which developing countries could also receive a share.
Example: Big Tech Inc. has a revenue of 100 billion EUR and a profit of 20 billion EUR. The share of the profits that can be taxed by consumer markets corresponds to 2.5 billion EUR.5 This tax substrate is allocated between consumer markets based on their size. If Big Tech Inc. makes 40 % of its global sales from selling to EU citizens, EU countries could tax 1 billion EUR of profits.6 Assuming a 20% corporate income tax rate, EU countries would receive 200 million EUR in taxes.
Pillar One7 is set to be implemented through the Multilateral Convention to Implement Amount A of Pillar One. This is the treaty that we are interested in. It is also known as the “Multilateral Convention” or MLC. As of now, the MLC has not yet been signed. Negotiations on the convention are still ongoing. Once signed, it will need to be ratified by the participating countries before it enters into force.
b) A global minimum corporate tax rate of 15% (“Pillar Two”)
If the effective tax rate of a multinational enterprise in a jurisdiction is below 15%, the new rules allow other jurisdictions where the enterprise operates to impose a top-up tax to bring the effective rate up to 15%. So, even though not every country can be forced to introduce a 15% tax rate, multinational companies that are registered in tax havens but still operate in countries other than tax havens will be forced to pay a 15% tax rate.
Unlike Pillar One, the global minimum tax rate is being implemented through domestic legislation in each participating country rather than a multilateral treaty. The OECD has provided model rules, and countries are incorporating these into their own tax laws. Some countries have already started this process, and the global minimum tax is expected to start applying soon.
3. Winners and losers
The two pillars are a package deal. Big consumer markets, like the European Union, advocate for Pillar One to get a minimum share of the digital economy and would otherwise move forward with unilateral digital service taxes on big tech. The US government is more cautious about others taxing profits from tech companies and some in the US argue that it should not acknowledge the right of markets to tax residual profits. On the other hand, the amount that has been negotiated is a very small piece of the pie. If that is locked-in and remains as the digital economy goes this would in fact be very favourable to the US. The US government is also one of the big winners from a global minimum tax. Low-tax jurisdictions, such as Ireland and Caribbean countries, are less enthusiastic about Pillar Two as it erodes their competitive advantage.
However, the impact is not just a distributional shift between states. Indeed, if that were the case it would not have been able to get so many countries on board. Rather this is a distributional shift between the Westphalian system of state sovereignty over territory and a technopolar system dominated by multinational corporations. According to the OECD, the expected net impact of residual taxing rights for consumer markets is about $20 billion annually in additional government revenue. The global minimum tax is set to raise global tax revenues by $220 billion annually. So, this is a fundamental adaptation of the Westphalian system that may be necessary to ensure its survival in the 21st century.
The alternative to a global agreement is a jungle. In such a scenario, strong countries are highly likely to protect their interests in other ways. For example, countries will want to charge their value added tax for cross-border B2C services. To ramp up pressure for a Pillar One agreement, countries like France, the UK, India, and Canada have introduced a patchwork of unilateral digital service taxes which would also apply to cross-border AI services. Nor would I be surprised if some countries would eventually introduce datacenter localization requirements for AI services. In contrast developing countries have smaller markets and less administrative capacity. If a global tech firm deems the market too small to affect its bottom line, it may choose to not comply or could leverage its resources to lobby for exemptions or delays.
Similarly, without Pillar Two we might see the future emergence of AI tax havens.8 Big countries like the US can apply some arm-twisting to tax havens, as it’s already the case today. However, the net impact of ‘tax haven whack-a-mole’ will still be a more limited ability of most countries to raise taxes compared to a scenario with international coordination.
Thanks to
, Julius, Steve Newman & for valuable feedback on a draft of this essay. All opinions and mistakes are mine. Sadly, this is already my last post produced as part of the Roots of Progress blog-building fellowship. It was an honor & pleasure & of course I will keep blogging 🫡For example, the authors of the Windfall Clause (2020) rejected the idea of taxes arguing: “Given current realities, we do not anticipate that legally enforceable taxation and global distribution of AI windfall is politically feasible, whereas beginning a conversation around a voluntary commitment such as the Windfall Clause may be more so.” The BEPS project has been ongoing since 2013.
Individual income taxes include capital income (dividends, interest, capital gains). However, this is taxed at significantly lower rates than labor income (wages, salaries).
Strictly speaking this is BEPS 2.0. The initial BEPS 1.0 project was a 15-step action plan published in 2015 and the subsequent Multilateral Convention to Implement Tax Treaty Related Measures to Prevent Base Erosion and Profit Shifting. This treaty is also known as the “Multilateral Instrument” or MLI. It modifies bilateral tax treaties to implement BEPS-related measures. However, I’ve moved it to the footnotes because it’s not the main treaty that we’re focusing on, but it sounds the same. Tax lawyers are even worse than OpenAI at naming things.
Natural resources such as oil, gas, minerals as well as financial services are excluded.
25% * (20 billion EUR - 10 billion EUR)
40% * 2.5 billion EUR
Just to be extra-clear: these two pillars are in no way related to the three pillars of the pension system, discussed in a previous post.
This is highly speculative today. However, in the long run, we might see tax havens that don’t just serve human-led AI companies but that specifically aim to attract capital owned by AGIs. AIs with legal personhood could buy a charter city where they can largely dictate the laws. These laws would include no or very little taxes and AGI businesses that are legally registered there can offer cross-border B2C services. Such tax havens would arguably accelerate the reduction of the share of the global economy that is owned by humans / that can be taxed by states for human welfare.