What Could an AI Tax Look Like
Without question, we will eventually need an AI tax but how do we implement it without discouraging innovation and growth?
How to tax AI, or rather the beneficiaries of AI, has become a topic of discussion. The assumption being that AI will lead to sweeping job losses, rather than job creation. I’ve compared the scale of this disruption to something like streaming video, at least with the current tech stack, instead of comparing it to a society-altering paradigm shift like the Industrial Revolution. Even if AI does bring on the Fourth Industrial Revolution, there is historical precedent that it would create more jobs than it would destroy. The Industrial Revolution was literal automation, just of the mechanical variety, and it unlocked massive amounts of productivity and growth until the United States decided to outsource this profit center in favor of profit-taking.
Even so, if automation does continue to advance and grow, it’s worth looking at how to offset the societal impacts of AI adoption and disruption, should it lead to permanent job losses.
Why Tax AI?
The reasons to tax AI beneficiaries are many. In the United States, our entire income tax system is based on humans earning and contributing at the state and federal level to provide basic civic services. Depending on your political leanings, you may agree or disagree with this arrangement, whether the government should be doing more or whether the government should be taxing income at all.
At present, taxes fund important services like Social Security, which is under threat after unchecked spending in Washington and a general disinterest in reconciling the budget. Other services include Medicare, Medicaid, the Department of Education, and the list goes on.
The central issue is that AI isn’t taxable for income. It’s not defined as a person with a Social Security number, it’s not necessarily a sentient being, and it’s really defined more as a service right now. How long things will remain that way is unknown. It’s a moral, ethical, and philosophical conundrum that our current political environment is not equipped to resolve, not in the least.
So while AI remains in this gray area, any job losses attributed to it are just gone with no recourse. There’s no ledger keeping track of which jobs were lost to AI and who was affected. There’s no compensation or ramifications for eliminating jobs in the name of AI, which is perhaps why we saw companies act when they had the chance.
AI Washing
2025 was a difficult year for Artificial Intelligence’s reputation, as company after company laid off droves of employees in the name of AI, at least that was what captured headlines. Earlier in 2026, reports were coming out about companies not seeing any discernible lift in productivity from AI. This echoed a 2025 MIT report about 95% of companies not being able to see any gains from AI integration into the workforce. So if AI wasn’t actually productive in 2025, leading into 2026, why was it being blamed for the layoffs at these companies?
Many of the companies that announced layoffs were already in precarious positions as businesses. Most of them were announcing these layoffs after a down quarter or multiple down quarters, so the reprieve of a layoff was likely a welcome one for CEOs looking for a way to cut heads for underperformance while framing it as a forward-thinking strategy.
Some of the companies, once Wall Street darlings, had become seriously embattled.
Amazon
Over-hired during the pandemic.
Margins under pressure due to tariffs.
Expensive AI infrastructure buildout.
Ongoing corporate restructuring.
Meta
Spent tens of billions on Reality Labs.
The company’s “Year of Efficiency” predates the AI narrative.
Multiple restructuring cycles before AI became the explanation.
Block/Square
Prior to announcing an AI restructuring, Block/Square had three consecutive earnings misses.
Experienced general growth concerns across businesses.
Investor pressure.
The company had taken on efforts to improve and expand margins.
Block had a unique amount of bitcoin exposure during a down cycle, bitcoin had crashed from 94K to 66.8K.
Salesforce
Activist investor pressure.
Public commitment to improving margins.
Cost-cutting campaigns already underway.
In all of these cases, the companies were already under financial pressure. The pattern was one of contraction, not expansion, so the AI growth narrative isn’t a clean fit. That’s not to say growth and efficiency can’t coincide, but the timing of these announcements aligns more with cost-cutting rather than expansion.
Although Wall Street loves layoffs, they come with baggage for the CEO who presides over them. In most cases, they’re viewed as a sign of failure because they’re often a symptom of organizational failure to execute on some level. However, reorganizing for AI gave underperforming companies the perfect cover to have their cake and eat it too. Struggling companies were able to cut headcount while also looking like savvy business leaders.
Just as these jobs were lost to a black hole of uncertainty, so was the blame as it was thrown into the AI pit, something we’ve now seen the effects of as the vast majority of adults have turned on the technology. All this is to say that there isn’t really any evidence yet that AI has actually replaced any jobs, though its body count is climbing as its reputation plummets.
The Current Proposals
So now that we’ve established why we might need an AI tax and whether we have an acute AI job replacement problem, let’s look at the proposals being bandied about by tech and political leaders. These proposals range from improved measurement and tracking, workforce education initiatives, and universal basic income programs to more aggressive ideas such as transferring ownership stakes in AI companies, imposing wealth taxes, or levying excise taxes on massive data center build outs.
You will notice that conservatives are notably absent from this discussion. That is because, thus far, they have not put forth any major tax proposals specifically targeting AI. In their stead, I will reference British author Matt Ridley, whose book How Innovation Works provides a useful framework for understanding the tension between technological progress and government intervention.
Dario Amodei, CEO of Anthropic
Reading through Amodei’s proposals, they’re not surprisingly very well reasoned and well thought out. The guy thinks about AI all day and all night. It’s funny how policies are either hyper-reactive, as a result of something that just happened, or as a result of some extrapolation into the distant future based on any number of circumstances. Amodei’s policy proposals are the latter.
Measuring and Tracking, We touched on the hazard of this earlier. All the jobs that were lost in 2025 are just lost in a wash with any other data; they aren’t attributed to AI in any specific way. Maybe this was intentional, maybe not. Amodei stresses the importance of tracking this data in order to accurately impose any sort of legislation, penalties, or benefits.
Pro-Employment Benefits, He’s more vague about this one. It’s the carrot-versus-the-stick approach to get employers to retain talent and not give in to their instinct to cut costs and fully automate. The forms he suggests are grants and training programs, but this would likely look like tax credits or some other well-known tool to incentivize businesses to retain talent over machines.
Long-Term Economic Support, This is essentially universal basic income, which, depending on your political leanings, should be implemented today, even without the threat of AI, or is a reason to be burned at the stake. He shares some different mechanisms for this, but the reason why this would be introduced would be for a Depression-style disaster caused by AI job loss.
Bernie Sanders, U.S. Senator
Senator Sanders, unsurprisingly, is recommending some version of federalizing AI companies via a stock transfer. Once a company reaches $200 million in annual sales, this triggers the tax, which automatically transfers 50% of the company’s stock to a sovereign wealth fund. From there, the fund is managed by a committee appointed by the President. He argues that every American could see an annual $1,000 dividend from the fund, which isn’t nothing, but it’s also not going to make up for mass unemployment.
There are a few issues here:
Discouraging VC Investment, Startups have long been reliant on VC money to get started. Once this policy goes into effect, investing in an AI startup is somewhat pointless for VCs looking for another home run. If the company will just be taken away by the government at a certain point, or if there’s a hard cap on growth, there is no real point in investing in AI startups anymore.
Creating Market Volatility, There’s the issue of when a company reaches $199,999,999 in annual revenue and the stock just craters as literally everyone divests from it, not wanting to own what will ultimately be a different company with a different board now that the government owns so many shares. There’s also the possibility that the extra shares come from massive dilution.
A Powerful but Unqualified Board, The people managing the fund will likely be sitting on the board of a number of AI companies now. I would like to believe that the President would be able to select qualified people to sit on these boards and make smart decisions about these companies, but Washington’s track record with tech, cybersecurity, and now AI doesn’t inspire confidence.
Elizabeth Warren, U.S. Senator
I like Elizabeth Warren. I always think her policies are very thoughtful. I know people give her a hard time for some stuff that happened 40 years ago, but she’s proven to be a thoughtful legislator. Senator Warren’s proposal goes more after the picks and shovels, not the AI companies themselves, which is actually pretty clever.
The Excise Tax, This is the heart of her proposal, and it really shows an understanding of the issues and what would impact the industry, raise capital, but not discourage innovation. This is really a tax on the massive, and massively unpopular, data center buildout. The beauty of this is that by taxing the data centers, she points out that you’re also taxing the companies investing heavily in AI. She adds that the tax would scale with the size of the data center, which is a nice touch as well.
Taxing Investments in Technology, This is a nuance that has gone unnoticed for a long time, but this loophole got blown open recently, which is why data centers are popping up everywhere. Senator Warren notes, “Right now, companies pay payroll taxes for their workers but get tax breaks for investing in technology,” so the more expensive the buildout, the bigger the tax break.
Wealth Tax, Past versions of this tax from Senator Warren were something like 2% on people making $50,000,000 annually, and it capped at 3% for billionaires.
Matt Ridley, Author of How Innovation Works
If you sneered during the last two proposals, you may like what Matt Ridley would say on the issues of taxing companies and the effects it has on innovation. I’ve only read one of his books, How Innovation Works, but from that I can apply some of his notions to this problem. Nuclear power is still a controversial topic even in 2026. In Ridley’s book, he notes how it was slowed and essentially killed off following the disasters at Chernobyl and Three Mile Island.
In How Innovation Works, Ridley argues that bricolage (tinkering and combining existing ideas) is paramount to the success and proliferation of innovation. Once the government begins regulating and interfering, the technology reaches a point of stasis, no longer able to evolve and grow freely as it once did. He uses nuclear power as an example, citing how the technology hadn’t been able to advance for 40 years. It was only when fission began to round the corner that nuclear technology started to seem safe again, but fission is a fundamentally different approach to nuclear power than what was used at Chernobyl or Three Mile Island.
“Because it is a monopoly, government brings inefficiency and stagnation to most things it runs; government agencies pursue the inflation of their budgets rather than the service of their customers; pressure groups form an unholy alliance with agencies to extract more money from taxpayers for their members.”
How Ridley’s notions apply to AI and taxation/regulation, Using these statements as a guide, I would hazard a guess that Mr. Ridley would be against taxing and regulating AI at this stage, just a hunch. His main reasons would be that a catastrophe actually hasn’t happened yet, and even if it had, it’s not worth losing the ability to keep advancing the innovation in favor of whatever benefits the taxation or regulation might bring.
My Humble Proposal
Taxes aren’t just a way to raise money. While they do serve to fund many of the services we have, we have somewhat lost the plot on why we have taxes in the first place. Taxes are intended to encourage or discourage behavior in a society. A prime example is how we’ve essentially legislated away smoking tobacco, only to have younger generations revitalize it via vaping.
I can’t ignore how impactful indoor smoking bans were in reducing smoking, but the knockout punch was the taxes on cigarettes. A pack of cigarettes in my hometown of Minneapolis is $15.00 a pack. Sure, inflation is partly to blame, but these prices have been climbing for years as we try to tax away tobacco use.
I say all this because when we impose taxes on something nascent like AI, we have to be careful that we don’t stifle innovation and that we are encouraging or discouraging the desired behavior from businesses.
Tax the Replacement, Not the Adoption
Much of the angst around AI is about job replacement, yet many of the policies are focused on company ownership, data centers, or the complete destruction of the labor market. These are all treating the symptoms, not the underlying issue we want to address. I’d measure the two things that matter here: headcount and revenue.
For established companies, we would use the 2024 tax numbers to set the baseline for future years. New companies would start fresh, since they don’t have prior records to measure against, and generally speaking, the size of new companies is typically smaller, making the measurement less important. The reason for using 2024 is that it provides a reference point before widespread AI adoption.
The goal is not to stymie AI adoption or penalize a company for being good at AI. If a company can run 100 agents to great effect, good for them. If you want to spend $500,000,000 on tokens, be my guest. The thing we’re watching here is headcount relative to revenue.
Headcount goes up and revenue goes up? Great.
Headcount goes down and revenue goes down? Unfortunate, but acceptable.
Headcount goes down while revenue goes up? Now you’ve got a problem.
Why the Ledge, Not a Smooth Curve?
A gradual curve has no forcing function. There has to be a decision point between adopting more AI and hiring more people. With a tiered system that includes a steep ledge, there are critical points where a company would have to think hard about paying for more compute or bringing on more employees. Bringing on more employees has the added benefit of increasing the threshold for AI usage.
Here’s a rough breakout:
Under 20 Employees
Exempt.
We want experimentation, and this is where new ideas and exciting things happen.
50+ Employees
As long as workforce reduction remains under 10%, exempt.
A 10% year-over-year reduction can reasonably be viewed as normal attrition.
This could potentially be gamed through repeated annual reductions, though the effect diminishes as the workforce shrinks.
10%+ Headcount Reduction from 2024 Baseline or Year-over-Year
This is where RPE (Revenue Per Employee) comes into play.
How RPE Is Factored In
RPE flat or declining: Exempt
RPE up modestly (1%–9%): 10% tax
RPE up significantly (10%–19%): 30% tax
RPE up sharply (20%+): 50% tax
It’s a Simple Solution to a Complex Problem
I was racking my brain trying to figure out how to track compute, tokens, and whether we would need agent IDs or some other form of employee identification. All of those things could be gamed and would require new tools and legislation. What I like about this solution is that it uses tools already at our disposal to provide answers today.
One additional benefit of this mechanism is that it elegantly encourages the behavior we actually want. If an employer wants to add more AI, there are points where they must choose between AI and a human worker. At the same time, there are benefits to hiring humans because doing so also expands their AI allowance. The RPE ratio also allows policymakers to adjust thresholds over time to fit the needs of the economy and labor market.
Take a company with 1,000 employees in 2024 that later lays off 600 workers. It doesn’t necessarily need to hire back all 600 in the exact same positions, though I would require that new hires be full-time American citizens. I don’t say this for ethnocentric reasons but more because tech companies are outright exploiting H-1Bs which is working out badly for both the exploited workers and American workers. The company could reduce its RPE by hiring back 200 or 400 employees in different parts of the business while still adopting AI.
The intent is not to create more overhead and bureaucracy for AI providers or the companies using AI. Nor is it to discourage AI adoption, especially if it’s helping companies grow. What we want to penalize is one thing: removing human workers and replacing them with AI workers solely to increase profits.
Is it a perfect solution? No. I’m not a legislator, and there is likely an attorney somewhere who has already turned this policy concept into Swiss cheese. What I’m aiming for is a framework that helps us adopt AI without knee-capping ourselves. Looking at Senator Warren’s policies, I could see them working well in concert with something like this.
Nothing Is Certain Except Death and Taxes
There will be an AI tax. The question is: what form will it take?
Not to be dramatic, but Benjamin Franklin’s famous quote, “Nothing is certain except death and taxes,” could prove ironically accurate for the AI sector if the wrong approach is chosen. The longer we wait, the more likely this policy area becomes defined by knee-jerk reactions rather than long-term thinking.
Already, many of the current solutions being proposed seem punitive in nature instead of seeking to avoid the unintended consequences of an AI-driven economy. Part of that is due to the self-inflicted public relations disaster the AI sector has created for itself. The proprietors of frontier models are in full pitch mode as they attempt to justify their valuations, which means projecting far into the future where their models achieve uninhibited success.
Forecasting mass job losses only stimulates a defensive response from the public rather than a willingness to accept and adapt to the technology. My suspicion is that, after the IPOs, some of the hype will die down as people gain a clearer understanding of what is actually behind the frontier models, how these businesses operate, and what can reasonably be projected.
Until then, nothing is certain except death and taxes.