Paying for the AI Transition

Capturing AI Productivity to Fund UBI and UHI

AI and robotics could raise economic output while reducing the need for human labor. If that happens at scale, the US runs into a basic mismatch: much of today’s tax system is tied directly or indirectly to wages. Payroll withholding, wage income, and wage-driven consumer spending form a large part of the revenue “engine.” In a high-automation economy, that engine weakens unless the tax base shifts toward what still scales with output.

The AI transition as envisioned by some will cause massive upheaval to social, and economic systems across the world. Plenty of work must be done to plan for this potential future in developing frameworks to support such a transition. This article covers two things only: (1) the order-of-magnitude scale of UBI and UHI-style programs will require, and (2) the mechanisms government could use to capture enough AI-era productivity to fund them. It does not cover job churn, social stress, or geopolitical spillovers. The aim is to make the dollars concrete and map the major funding options. The intent of this article is not to predict the future, but provide guidance and a framework for what may be required and the scale of various options, and the magnitude of the challenges ahead.

For interactive scenario testing, see the Tax Project modeling tool to run your own simulations: https://taxproject.org/ai-transition-modeling/

“AI could wipe out half of all entry-level white-collar jobs – and spike unemployment to 10-20% in the next one to five years.”

Dario Amodei, CEO Anthropic

What “UBI” and “UHI” mean in this article

To keep the math understandable, costs are modeled per household (so family members are not counted multiple times).

UBI is treated as a “basic” cash floor, anchored to the real median household income. UHI is treated as a “high” cash floor, anchored to the 90th percentile household income (top 10% threshold). Those are not endorsements of any specific program design. They are benchmarks that help show how quickly the funding problem scales when the target shifts from “basic” to “high.”

Inputs used:

• US households: 132,737,146 (2024 ACS). [1]
• UBI level: $80,610 (2023 real median household income). [2]
• UHI level: $234,900 (2023 household income at the 90th percentile). [3]
• FY2024 federal receipts: $4.918 Trillion (for perspective). [4]
• Nominal GDP reference: about $29.8 Trillion (Q4 2024 annual rate). [5]

“AI will affect almost 40 percent of jobs around the world, replacing some and complementing others.”

“In advanced economies, about 60 percent of jobs may be impacted by AI.”

IMF [6]

Cost Overview: How big the dollars get, fast

The chart below (Figure 1) shows annual cost if UBI or UHI is paid to 25%, 50%, 75%, or 100% of US households. Coverage is included to show scale and sensitivity. It is not a policy recommendation.

The chart in figure 1 is used to illustrate a point: a broad UBI quickly becomes comparable in cost to the entire federal government; a “high” UHI level program becomes many times larger. Whatever a future program looks like, the US would be operating at a scale where the funding base and capture method must be designed to support an order of magnitude larger change, and put in place mechanisms to capture the AI transition gains.

What AI Productivity gains would be needed?

The term “Productivity gains” can sound abstract. In simple terms we can discuss the cost required in comparison to the share of the economy it represents.

Using the Gross Domestic Product (GDP) used to represent the Total Economic output of our country as a reference point of roughly $29.8T [5]:

• UBI (100% coverage) at $10.7T is roughly 36% of today’s GDP.
• UHI (100% coverage) at $31.2T is roughly 105% of today’s GDP.

So if Government wishes to implement UBI/UHI style programs the Economy must grow substantially, and the Government must create a mechanism to reliably capture these productivity gains in order to support the transfers required for these programs.

In other words, in a post AI transition economy – the AI productivity gains and value generated from AI would need to be of an enormous scale, and the collection mechanism would need to capture the value reliably in order for these programs to work.

A practical way to think about it is the “capture rate” for AI Productivity is what share of national output that ends up as public revenue. In FY2024, Federal revenue was $4.918T, which is roughly 16%-17% of GDP depending on the GDP measure used. [4][5] That gives a rough idea of what “normal” looks like today.

“It’ll be 10 times bigger than the Industrial Revolution – and maybe 10 times faster.”

Demis Hassabis, Google DeepMind

Now apply that to our UBI/UHI scenarios:

• If public revenue stayed around 17% of GDP, funding a $10.7T UBI would require an economy on the order of $60T+ per year (because 17% of $60T is about $10.2T). That is roughly 2x today’s GDP.
• Funding a $31.2T UHI at the same revenue share would imply an economy on the order of $180T+ per year, which is roughly 6x today’s GDP.

Those are not forecasts. They are scale checks. They show that sustaining large transfers requires either:

• Much higher Tax Rates
• A much larger economy (Meaning AI is going to have to create ALOT of value and productivity gains)
• Lower or Narrower benefits than the “all-on” benchmarks used here.

This is why funding design matters as much as growth assumptions. The question becomes: what can the US reliably tax when labor income is no longer the central revenue base and jobs and income are increasingly lost to automation.

Funding: what can are the captures mechanisms in an AI-heavy economy?

Here we look at possible mechanisms that could be put into place in a framework that would capture the AI Productivity and fund the UBI/UHI style programs. In a post AI transition steady state, the most stable funding tends to come from bases that are broad, measurable, and hard to evade. “AI-specific” taxes can contribute, but most are better as supplements than as the backbone.

AI Productivity Capture Mechanisms

Which Option?

The Tax Project does not comment on policy recommendations. However, a stable post-transition funding system is unlikely to rest on a single “magic tax.” It is more likely to resemble a stack. Broad consumption-based funding (Option C) is structurally strong because it scales with the economy regardless of labor share. Rent-like taxes (Option F) are also structurally strong because they target bases that cannot be moved offshore easily (land and spectrum are the clearest examples). Corporate redesign and capital income taxes (Options E and D) matter because AI-era gains may concentrate in profits and asset returns, but these are also the most contested and gameable bases, which is why they tend to work best as pillars alongside broader bases.

The AI specific ideas—charging by “labor equivalents (LE) replaced” (Option A) or metering compute/tokens (Option B) – may read well politically, but they struggle as a primary funding backbone. LE charges tend to collapse under definition and when corporations game the system, and usage taxes are proxies that can be arbitraged. This is not to say that corporations are doing anything wrong, they will naturally optimize for their circumstance. These mechanisms may still contribute at the margin, especially if applied upstream where metering is clean, but the heavy lifting usually falls to broad, durable bases and, over time, to ownership structures that turn productivity into dividends (Option G). Whatever is chosen, this will be a tricky path. There are always second order effects, and any tax can potentially harm productivity and the competitiveness of those being taxed, particularly in a global economy.

The Road Ahead: Plan now

If AI and robotics displace labor at scale, as many have predicted, the US cannot assume that today’s labor heavy tax base will keep funding the government it currently has, let alone new multi-trillion-dollar benefit programs in the form of UBI or UHI. The scale shown above implies a hard reality: sustaining UBI or UHI at meaningful coverage requires some combination of (1) a much larger economy, (2) a higher and more durable share of GDP collected as public revenue, and (3) redesigned tax bases that still work when wages are no longer the main revenue collection mechanism.

That redesign cannot be improvised in a crisis. It requires advance planning: defining viable tax bases, building administrative systems that can operate at scale, and establishing rules stable enough to survive political cycles. Even if the transition is slower than expected, the time to build a credible framework is now – these systems and the required legislature will take a good deal of time.

References

[1] US Census Bureau, 2024 ACS 1-year estimate: total households (United States) = 132,737,146. (data.census.gov)
[2] US Census Bureau, Income in the United States: 2023 (P60-282): real median household income = $80,610. (census.gov)
[3] US Census Bureau, P60-282 Table A-3: household income at the 90th percentile = $234,900 (2023). (www2.census.gov)
[4] US Treasury, FY2024 total federal receipts = $4,918.1B. (fiscal.treasury.gov)
[5] FRED/BEA nominal GDP series (Q4 2024 annual rate ~ $29.8T). (fred.stlouisfed.org)
[6] Georgieva, Kristalina. “AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity.” International Monetary Fund Blog, January 14, 2024. https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity
[7] World Economic Forum. “Future of Jobs Report 2025: 78 Million New Job Opportunities by 2030, but Urgent Upskilling Needed to Prepare Workforces.” World Economic Forum (Press Release), January 2025. https://www.weforum.org/press/2025/01/future-of-jobs-report-2025-78-million-new-job-opportunities-by-2030-but-urgent-upskilling-needed-to-prepare-workforces/