When the rules break: The tax base is dying, but Africa cannot afford to mourn it

On the paradox of automation, and why African governments must design new revenue instruments before the old ones collapse.

Here are some numbers that should keep every finance minister on the continent awake at night. In the US, roughly three-quarters of all federal tax revenue derives from labour income. In most African economies, where the tax-to-GDP ratio averages 16.5% against an OECD average of 34.1%, the fiscal base is thinner still. Now consider that the IMF estimates that 60% of jobs in advanced economies are exposed to artificial intelligence, with about half of those facing genuine automation risk. The ILO also reports that nearly one in four jobs are at risk of being transformed by generative AI. Most crucially, the global share of national income going to workers has been falling: from 53.0% in 2014 to 52.4% in 2024, with Africa and the Americas experiencing the steepest declines.

The 20th-century state was funded by the sweat of the brow. What happens when the brow is no longer sweating?

We at Africa Practice have called this the terminal fiscal paradox. Automation is eroding the labour-based tax revenues that fund the state, at precisely the moment when Africa’s demographic trajectory is demanding that the state spend more. The research supports this framing. For instance, Brookings researchers argue that the main burden of taxation will have to find new bases beyond labour income, as AI reduces demand for human workers. They propose a phased approach: broadening consumption taxes in the near term, then developing instruments like token taxes on consumer-facing AI services as AI systems become more autonomous.

However, they also argue, persuasively, that crude robot taxes and compute taxes would be counterproductive. That taxing the ownership of robotic equipment or levying charges on computational resources would directly penalise productive capital investment. They compare it to taxing steel during the Industrial Revolution. For an advanced economy with strong domestic AI capacity, this logic possibly holds. For Africa, I am less convinced.

The distinction is that in a jurisdiction where the AI infrastructure, the models and the platforms are domestically owned, I can concede that taxing their inputs might risk slowing the innovation that generates prosperity. Africa is not in that position however, because the models are, by and large, built abroad; the compute infrastructure is overwhelmingly foreign-owned, and the data that feeds the algorithms is extracted from African populations, agricultural systems and health networks, processed offshore and sold back as finished services at prices set elsewhere. The extraction trap will simply migrate from physical minerals to digital value chains.

So the question for African policymakers is arguably different from the one Brookings is answering. It is not: how do we avoid penalising domestic AI investment? It is: how do we capture value from AI-driven productivity gains that flow from our data, our labour markets and our sector-specific applications, when the bulk of productive apparatus sits outside our borders?

Several instruments are emerging in the global debate, each with trade-offs. One writer proposes an AI-hours tax: a levy on computational time usage, analogous to how labour hours are taxed via payroll contributions. One of his most interesting design features is the augmentation/substitution distinction: firms where humans remain the primary decision-makers, using AI as a tool, would pay lower rates than firms where AI executes full workflows autonomously. This aligns the tax incentive with the policy goal of keeping humans economically relevant. The problem however, is jurisdictional mobility, because cloud infrastructure can be relocated far more easily than a factory.

OpenAI’s April 2026 policy proposals take a different approach, in that, rather than taxing AI directly, they suggest (amongst other things) that governments take equity stakes in AI infrastructure, with returns distributed to citizens through public wealth funds. This is closer to the sovereign wealth fund model that several resource-rich African states already understand, as it reframes the relationship from taxation to ownership. The question is, however, whether African governments have the negotiating leverage to secure equity in foreign-owned AI infrastructure deployed on their soil.

Some may say that digital service taxes offer a more immediately practical path for the future, and several African countries have already moved in this direction. Kenya transitioned from a 1.5% Digital Service Tax (DST) to a 3% Significant Economic Presence (SEP) tax, collecting KSh 2.3 billion from 454 foreign digital providers by August 2025. For proponents of this approach, this may be proof of concept that African countries can enforce digital taxation on foreign providers, even without physical presence.

My honest futures view is that no single instrument will or should replace labour-based taxation. Africa will need a portfolio approach. It is up to our governments, whether this approach includes some form of broadened consumption taxes (such as VAT with AI-assisted administration); SEP taxes on foreign digital providers; sector-specific data extraction levies tied to the commercial use of population-level African data for AI model training; negotiated equity positions in AI infrastructure being built across the continent by hyperscalers; or graduated automation levies that distinguish between augmentation and substitution, incentivising firms to keep humans in the loop. Fiscal architecture, however, is not an end in itself. If these instruments succeed in capturing value from machine-driven productivity, the distribution question cannot be deferred. Several of them, particularly the augmentation/substitution distinction, only make design sense if there is a stated policy commitment to keeping humans in the economic loop; what they fund, and for whom, needs to be answered with the same rigour applied to the instruments themselves.

Ultimately, no one instrument suffices, but together, they may form a fiscal architecture for the dawning age of AI, and crucially; the window for designing these instruments is narrower than most governments appreciate. 

However, a caveat is warranted, particularly as other researchers argue that the AI productivity boom has not yet materialised at the macroeconomic level. For instance, an NBER study found 90% of firms report no measurable productivity improvement from AI. The Solow Paradox, thought to have been resolved, is repeating, but there is a visible historical pattern: the IT productivity payoff took a decade to appear, and when it did, it transformed everything. The time to build fiscal instruments is during the lag, not after the transformation is complete.

Africa’s finance ministries need to stop asking whether the old tax base will hold. In our view, it will not. But the more useful observation is this: the countries that industrialised first are all, to varying degrees, trapped by pension obligations, by entrenched fiscal frameworks, and by political commitments to labour arrangements that are becoming economically redundant. That means Africa’s fiscal architecture for the economy that is arriving can be constructed from first principles. The question is what replaces it, and whether the continent designs that replacement on its own terms or inherits one written elsewhere.