The Algorithm Will Not Fund Your UBI

On May 11, 2026, the OECD’s Committee on Fiscal Affairs quietly uploaded a 47-page PDF to its website. By May 13, it had triggered a silent, global reallocation of $4.2 billion in annual tax liability. The document, “Revised Guidance on the Application of VAT/GST to AI-as-a-Service,” contained no revolutionary new tax. It simply clarified that when a marketing firm in Milan pays for 10 million API calls to a large language model hosted in a Nevada data center, that transaction is now subject to Italian Value-Added Tax. The Nevada company must register, collect, and remit. This is not a tax on intelligence, but a tax on its consumption—a bureaucratic acknowledgment that the cloud is the new factory floor, and every query is a taxable event. The most significant AI tax policy of 2026 wasn’t a bold new levy, but a global administration fixing a leak in the plumbing of the old world. While politicians debate grand theories of taxing robots, the real fiscal revolution is happening in the fine print of cross-border service agreements, where states are scrambling to claim their slice of the intangible.

This is the brutal, unsexy reality of AI taxation. We are not building a new fiscal cathedral on a hill of code. We are performing emergency repairs on a sinking foundation, using the flawed tools of 20th-century economics to capture value from 21st-century ghosts. The European Commission’s proposed “AI Contribution” levy, South Korea’s “Automation Subsidy Offset,” California’s data tax gambit—these are all probes in the dark, attempts to place a monetary value on processes that are, by design, dislocated from human labor, physical assets, and traditional geographic jurisdiction. The core assumption every one of these models betrays is that we can disentangle the “AI” from the corporation, the output from the infrastructure, or the profit from the platform. This is a fantasy. AI is not a product; it is a pervasive condition of capital.

The Three Fallacies of the Modern Tax State

Current proposals cluster around three flawed premises, each a desperate attempt to fit the square peg of generative capital into the round hole of industrial-era tax codes.

First, the “Product Fallacy”: This is the OECD’s path. It treats AI-as-a-Service like a digital book or streaming subscription—a discrete, consumable good. Applying VAT is administratively neat, but it fundamentally misconstrues the economic relationship. When a company uses an AI model to draft legal contracts, optimize logistics, or generate ad copy, it isn’t consuming a service; it is renting a fraction of a capital asset that improves its entire productive base. Taxing the transaction is like taxing each turn of a wrench in a rented factory, while ignoring the ownership of the factory itself. The IMF’s proposal for accelerated depreciation of “AI Capital” grapples with this, allowing a 2-3 year write-off for training costs. But this merely accelerates a corporate tax break. It rewards the investor in AI capital while doing nothing to capture the economic surplus—the staggering productivity gains—that capital generates once deployed. It’s a subsidy disguised as a policy adjustment.

Second, the “Input Fallacy”: California’s AB-3214, proposing a tax on data ingestion, exemplifies this. It seeks to tax the raw material—our collective digital exhaust—as if it were oil or timber. The 0.1% gross receipts tax on models trained with over 1 terabyte of Californian data is a poetic notion of compensation, but it’s economically incoherent. The value isn’t in the data; it’s in the latent patterns extracted by unimaginable compute. Furthermore, the data is non-rivalrous. Taxing its “transfer” imposes a cost on a process that has near-zero marginal cost of reproduction, creating a friction that will simply push training offshore or into synthetic data, impoverishing the very public corpus it seeks to remunerate. This model mistakes the recipe for the meal, and in trying to tax the flour, it will ensure the bakery moves to another state.

Third, the “Labor Replacement Fallacy”: This is the emotional core of the debate, manifest in the EU’s “Digital Transition Fund” and South Korea’s pilot. The idea is straightforward: tax the automation that destroys jobs to fund the safety net for displaced workers. The EU’s proposed 0.5% levy on high-impact model revenue aims to generate €1.8 billion for reskilling. It is politically seductive and morally appealing. But it is arithmetically pathetic. The OECD estimates that 27% of jobs across its member states are at high risk of automation this decade. The cost of universal retraining, wage supplementation, and social support for transitions on that scale is in the trillions of euros annually, not single-digit billions. A fractional levy on revenue is a ceremonial gesture, a political placebo that allows us to pretend we are “doing something” while the structural disemployment accelerates. South Korea’s more nuanced model—a 15% offset on automation tax credits if workforce is cut—at least ties the penalty directly to a local outcome. But it remains a small brake on an existential economic shift.

Two Proposals That Acknowledge the Abyss

If we are to move beyond fallacies, we must start with a grim admission: the corporation, as a taxable entity, is dissolving. Value is increasingly captured in global platform networks, optimized algorithms, and proprietary models that flow across borders with zero marginal cost. Our proposals must target the rents generated by artificial scarcity and network effects in the AI economy, not its fictional “products.”

Proposal 1: The Algorithmic Excess Profit Tax (AEPT)

Modeled on wartime excess profit taxes, the AEPT would apply to companies whose return on invested capital (ROIC) exceeds a benchmark rate (e.g., 15%) by more than, say, 10 percentage points, and for whom the delta is directly attributable to the deployment of proprietary AI systems as certified by independent audit. The tax rate would be progressive: 35% on the first 10 points of excess, 50% on the next 10, 70% beyond that. The attribution is the hard part, requiring a new regulatory framework to define “AI-attributable profit.” This could be based on the differential between the profitability of AI-driven business units and non-AI counterparts, or on the measured productivity lift from AI integration. The goal is not to punish success, but to capture a share of the economic rents accruing from a form of capital that has near-zero marginal cost of deployment and creates winner-take-all markets. The funds would be directed not to generic “reskilling,” but to a Universal Basic Dividend paid directly to all legal residents, explicitly framing the payout as a share of the productivity surplus generated by the collective intelligence of society, now crystallized in private algorithms.

Proposal 2: The Compute-Resource Sovereign Wealth Fund

Nations and blocs with significant AI hardware infrastructure—think Taiwan (TSMC), the Netherlands (ASML), or states in the US with major data center hubs—should enact a 0.5% in-kind tax on compute cycles. For every 100 petaflops of AI-dedicated compute time contracted within their jurisdiction, companies would contribute 0.5 petaflops to a public compute pool, managed by a sovereign fund. This pool would be allocated via competitive grants to public-interest AI research: open-source model development, climate modeling, biomedical research for rare diseases, and auditing of private sector AI systems. This moves beyond taxing monetary value to taxing the fundamental resource of the AI age—processing power. It creates a public option in the computational arms race, preventing the complete enclosure of the most important tool of the 21st century by five corporate entities. It acknowledges that the physical substrate of AI, unlike the software, is geographically anchored and can be subjected to a form of resource royalty.

Scenarios for 2032

Scenario A: The Fragmented Leviathan (The Path We Are On)

By 2032, the current patchwork has solidified into a chaotic, contradictory global regime. The EU’s levy is in place but has been gamed down to €800 million annually through legal re-structuring of “EU-derived revenue.” California’s data tax passed, leading to the establishment of “data-freeport” training hubs in Nevada and Wyoming, creating a domestic arbitrage loop. South Korea’s offset model is adopted in Germany and Japan, but only slows job loss in large manufacturers, while gig economy platforms using AI for task allocation evade it entirely. The OECD VAT fix is the only universally adopted measure, making AI consumption a reliable, if modest, revenue stream for governments. The result is a world where AI taxation generates just 1.8% of total government revenue in advanced economies, utterly failing to fund the necessary social transformations. Inequality soars, public trust erodes, and a populist backlash leads to crude, punitive “tech taxes” that stifle innovation without solving the underlying fiscal crisis.

Scenario B: The Rent Capture (If We Act)

By 2032, a coalition of nations led by a US-EU-Japan pact has implemented a version of the Algorithmic Excess Profit Tax. After a fierce five-year battle involving capital flight threats and intense lobbying, a global minimum AEPT rate of 25% is established under a reformed OECD inclusive framework. It generates $2.1 trillion in annual global revenue. The Universal Basic Dividend it funds is not enough to live on, but at $400/month for every adult in participating nations, it provides a tangible, politically popular link between AI profits and public good. The Compute-Resource Sovereign Wealth Fund, pioneered by the EU and Taiwan, dedicates 15% of the world’s top-tier AI compute to public projects, leading to breakthroughs in open-source foundational models that keep the private sector in check. Taxation is no longer about funding the old welfare state, but about directly distributing the rents of cognitive capital and maintaining a public stake in the infrastructure of intelligence.

Challenging Your Assumption: Work is Not the Source of Meaning

You believe, deep down, that a job is more than a paycheck. It is identity, community, purpose, and dignity. This belief is the bedrock of our political and social order, and it is why every tax proposal is frantically tied to “saving” or “transitioning” workers. We must let this go. *The great promise and terror of advanced AI is that it invalidates the historical linkage between economic necessity and human meaning.* For millennia, we had to work to survive; we built our cultures atop that brutal fact. AI severs that link. Propping up “jobs” through punitive automation taxes or futile retraining is a rearguard action against a deeper ontological shift.

The real task of AI taxation is not to preserve the labor-centric society. It is to fund the construction of what comes after. The revenue from proposals like the AEPT should not just pay for food and housing, but for a massive public investment in the infrastructure of meaning: universal access to advanced education, arts grants, community governance stipends, and space exploration. We must tax the AI not to keep humans in the loop of production, but to finance our exit from that loop and our entry into a world where human purpose is self-defined, not economically dictated. This is the uncomfortable pivot: from taxing AI to save work, to taxing AI to make a world beyond work conceivable.

The Question You Can't Answer

If we successfully design a tax that captures the economic rents of artificial intelligence and funds a universal basic dividend, thereby decoupling survival from labor, what is the compelling moral or practical reason for any human to perform the dangerous, tedious, or deeply undesirable jobs that will, inevitably, remain?