1. Introduction
In January 2026, Anton Korinek and Lee Lockwood published “The Future of Tax Policy: A Public Finance Framework for the Age of AI” through the Brookings Institution. The paper diagnoses a real problem: AI-driven automation will erode employment-linked taxation, which accounts for roughly three quarters of US federal revenue and over 42% of UK tax receipts. On the diagnosis, they are correct. On the prescription, they are wrong in ways that matter.
Their central recommendation is to shift from labour taxation to consumption taxation while protecting AI capital assets from all taxation in the near term. They explicitly compare compute taxes to “taxing steel during the industrial revolution” and recommend sovereign wealth funds and voluntary corporate windfall clauses as fiscal insurance.
The Sovereign Energy and Bandwidth Excise (SEBE) framework offers a direct alternative. This document examines where Korinek and Lockwood’s reasoning fails and why infrastructure-based taxation (energy and bandwidth) succeeds where consumption taxation cannot.
Source: Korinek, A. and Lockwood, L.M. (2026). “The Future of Tax Policy: A Public Finance Framework for the Age of AI.” Brookings Institution, 8 January 2026. https://www.brookings.edu/articles/future-tax-policy-a-public-finance-framework-for-the-age-of-ai/
2. The Consumption Tax Fallacy
Korinek and Lockwood propose consumption taxation as the primary replacement for eroding labour income tax. This is the central pillar of their framework. It does not stand up.
2.1 The circular dependency
Consumption tax revenue requires consumers with purchasing power. If automation displaces workers (which the paper acknowledges is happening), those workers lose income. Unemployed people consume less. Consumption tax receipts fall. The revenue base erodes in lockstep with the problem it is supposed to solve.
The authors acknowledge this indirectly when they note that “government revenues from payroll taxes as a fraction of GDP will decline just as needs for retraining programs and transition support increase.” They then propose consumption taxes to “bridge this gap.” But the gap exists because people have less money to spend. Taxing their reduced spending does not bridge anything. It widens the gap.
2.2 Consumption tax is regressive
This is not a contested point. It is established public finance. Consumption taxes take a larger share of income from low earners than from high earners because low earners spend a higher proportion of their income. The authors acknowledge this (“consumption taxes are often criticized as regressive”) and propose offsetting it through progressive design (taxing the difference between income and net savings at progressive rates).
This offset requires knowing everyone’s income and savings. It requires the same administrative apparatus as income taxation. It solves the regressivity problem by rebuilding the income tax system inside the consumption tax system. At which point the question becomes: why not just fix the income tax system? Or better, build a new one that does not depend on employment at all.
2.3 The demand destruction problem
The paper does not model what happens to aggregate demand when automation displaces enough workers. A consumption tax on a shrinking consumer base is a fiscal death spiral: automation reduces employment, employment loss reduces consumption, reduced consumption reduces tax revenue, reduced revenue reduces the state’s capacity to support displaced workers, reduced support further depresses consumption. Each turn of the cycle makes the next turn worse.
SEBE breaks this cycle because it taxes the production infrastructure, not the downstream spending. Revenue grows with automation regardless of what happens to consumer demand. More automation means more energy consumption at the macro level, which means more SEE revenue, which means more capacity to redistribute, which means more consumer spending, which means more conventional tax revenue. The feedback loop runs in the right direction.
3. The “Don’t Tax Capital” Error
The paper’s second major recommendation is to avoid taxing AI capital assets. They argue that robot taxes and compute taxes discourage investment and innovation. They compare compute taxes to “taxing steel during the industrial revolution: a self-defeating policy.”
This conflates three different things.
3.1 SEBE is not a capital tax
A capital tax penalises ownership. A robot tax penalises the decision to deploy automation. SEE does neither. SEE taxes energy consumption at the point of load. The asset is irrelevant. What matters is how much electricity it draws.
A company that replaces 100 workers with 1 efficient server pays less SEE than a company running a legacy data centre at twice the wattage for the same output. The incentive is to invest in better, more efficient capital. That is the opposite of discouraging investment.
The steel analogy is wrong. Taxing steel penalises the asset. SEE taxes the coal that fires the furnace. The incentive is to build a better furnace, not to avoid building one.
3.2 The efficiency incentive
Korinek and Lockwood worry that taxing compute infrastructure will “slow the productivity growth needed to fund public priorities.” SEE does the opposite. By taxing energy consumption (not compute operations, not hardware ownership, not FLOPS), SEE creates a direct financial incentive to:
- Build more energy-efficient data centres
- Invest in renewable generation (lower marginal cost per kWh)
- Optimise workloads to reduce energy waste
- Locate facilities where energy is cheapest and greenest
Every one of these is a productivity improvement. SEE aligns the tax incentive with the efficiency incentive. A capital tax or robot tax would not.
3.3 The sovereignty argument they miss
The paper does not discuss compute sovereignty at all. SEBE’s Digital Customs Duty (DCD) makes offshore compute more expensive than domestic compute (2x the domestic SEE-equivalent rate). This incentivises building UK data centres, reducing dependence on US hyperscalers (AWS, Azure, GCP). Korinek and Lockwood, writing from a US perspective, naturally do not consider that other nations might want to tax cross-border compute precisely to build domestic capacity. For any non-US nation, this is a strategic priority the paper ignores entirely.
4. The Windfall Clause Fantasy
The paper proposes “windfall clauses, voluntary commitments by AI companies to share gains broadly if they achieve transformative breakthroughs” as fiscal insurance alongside sovereign wealth funds.
Voluntary commitments are not tax policy. No government funds public services from corporate goodwill. OpenAI restructured from non-profit to capped-profit to for-profit in under five years. Anthropic’s voluntary commitments exist at the pleasure of its board. Neither entity is bound by statute to share anything.
Sovereign wealth funds are a serious instrument, but they require a revenue source. Norway’s Government Pension Fund Global is capitalised by petroleum taxation, which is a tax on energy production. The analogy supports SEBE, not the Brookings framework. You need a revenue mechanism to capitalise the fund. SEBE provides one. Windfall clauses do not.
5. What They Get Right
The paper correctly identifies that:
- Labour taxation is structurally doomed as automation scales
- The fiscal crisis will arrive before political consensus on what to do about it
- International coordination is needed to prevent tax arbitrage
- Timing matters: getting the transition wrong is costly
On all four points, SEBE agrees. The disagreement is entirely about the mechanism. Korinek and Lockwood propose taxing displaced workers on their reduced consumption while protecting the infrastructure that displaced them. SEBE proposes taxing the infrastructure directly, at the physical layer where measurement is objective, evasion is difficult and revenue scales automatically with the problem.
6. The Category They Cannot See
The Brookings framework offers a binary: tax consumption or tax capital. This is a false choice. SEBE taxes neither. It taxes the physical inputs to production: energy (kWh) and cross-border data (TB). These are not capital assets (owning a server is not taxed). They are not consumption (buying an AI service is not taxed). They are the operational footprint of automated production, measurable at physical infrastructure (power meters, internet exchange points) with existing technology.
The paper cannot evaluate SEBE because its framework has no category for it. Infrastructure-based taxation sits outside the consumption-versus-capital binary that organises the entire analysis. This is the central intellectual failure: the taxonomy is incomplete, so the best option is invisible.
7. Revenue Comparison
Korinek and Lockwood do not provide revenue estimates for their consumption tax proposals. SEBE does.
| Mechanism | 2030 Revenue | 2040 Revenue | Scales with automation? |
|---|---|---|---|
| SEBE (SEE + DCD) | £34-46B | £93B | Yes (energy + bandwidth growth) |
| Consumption tax on AI services | Unknown | Unknown | Only if consumers can afford to buy them |
| Windfall clauses | £0 (voluntary) | £0 (voluntary) | No |
| Sovereign wealth fund | Requires revenue source | Requires revenue source | Depends on capitalisation |
SEBE generates £34-46 billion at launch in 2030 (2026 prices), growing to £93 billion by 2040 and £159 billion by 2045 as automation scales. All rates are CPI-indexed. All projections are in 2026 real prices. The full revenue model is published with source data and methodology (SEBE Revenue Model).
The consumption tax alternative has no published revenue estimate, no rate structure, no implementation timeline and no growth model. It is a direction of travel, not a proposal.
8. Conclusion
Korinek and Lockwood have written a careful paper that arrives at the wrong answer. Their diagnosis is sound: labour taxation cannot survive automation. Their prescription (consumption tax plus capital protection plus voluntary corporate generosity) fails on three counts:
- Consumption tax is self-defeating. It depends on consumer spending that automation erodes.
- Capital protection is a category error. SEBE does not tax capital. It taxes energy consumption, which rewards efficiency and investment.
- Voluntary mechanisms are not fiscal policy. Windfall clauses and sovereign wealth funds without revenue sources are gestures, not instruments.
SEBE provides what the Brookings paper cannot: a specified mechanism with published rates, revenue estimates, evasion countermeasures and an implementation pathway. It taxes the physical infrastructure of automated production at the point where measurement is objective and avoidance is difficult. Revenue scales automatically with automation. The harder the problem gets, the more revenue SEBE generates.
The Brookings paper asks the right question. SEBE is the answer they could not find because their framework had no room for it.
Copyright 2026 Jason Huxley. Licensed under CC-BY 4.0.