Senator Kelly’s Plan to “Make AI Work for Americans” Is Funded by Three Poorly Designed Taxes

Motivated by concerns that artificial intelligence (AI) adoption may lead to substantial and rapid changes in labor markets, Sen. Mark Kelly’s (D-AZ) proposed Make AI Work for Americans Act includes federal funding for several labor insurance and worker training programs. AI’s impact on the labor market remains debated, but some observers project that many workers will need to switch jobs.

To fund the additional spending, the proposal includes three new taxes: a digital advertising taxA tax is a mandatory payment or charge collected by local, state, and national governments from individuals or businesses to cover the costs of general government services, goods, and activities., a bit tax, and an excess profits tax.

Table 1. Sen. Kelly’s Proposed AI Taxes

Source: Senator Mark Kelly, “Kelly Introduces Bill to Make Sure Big Tech Pays Fair Share and Invests in American Workers,” Sep. 24, 2026, https://www.kelly.senate.gov/newsroom/press-releases/kelly-introduces-bill-to-make-sure-big-tech-pays-fair-share-and-invests-in-american-workers/; Tax Foundation analysis.

Sen. Kelly should be commended for attempting to pay for additional spending. With the federal government’s debt burden and interest on the debt at near-record levels and projected to increase further, policymakers should not propose new spending without identifying ways to pay for it. However, Sen. Kelly’s proposal may end up increasing deficits, since his proposal spends gross revenues but does not account for expected declines in other taxes, such as income and payroll taxes.

Each of the three taxes would take effect on January 1, 2027. We estimate that without any changes in behavior or economic growth in response to the tax, in the first year, the digital ad tax would raise $15.8 billion, the excess profits tax $6.5 billion, and the bit tax only $23.2 million.

Ten-year projections are challenging, since the growth rates of the affected tax bases are highly uncertain. Moreover, we have not modeled behavioral responses, including avoidance behavior, nor macroeconomic effects, and so our estimates overstate the revenue that would be raised. We estimate that over 10 years, the three taxes could raise a combined $380 billion, but this should be seen as a ceiling because it does not incorporate any feedback effects.

All three tax proposals are unsound. The digital ad tax is a gross receipts taxGross receipts taxes are applied to a company’s gross sales, without deductions for a firm’s business expenses, like compensation, costs of goods sold, and overhead costs. Unlike a sales tax, a gross receipts tax is assessed on businesses and applies to transactions at every stage of the production process, leading to tax pyramiding. that can be charged several times along the advertising chain; the bit tax may constitute a growing share of the cost of using AI and serve as a brake on technological development; and the excess profits tax may fall largely on one taxpayer.

If lawmakers want to pay for new spending, broad-based taxes—preferably on consumption—would do so at a far lower economic cost.

Historical Precedents

While Sen. Kelly may be the first lawmaker to package these three taxes together, none of them are new ideas. In fact, all have historical precedents at the federal, state, or international levels.

Digital ad taxes have been most popular at the state level. Maryland, Illinois, and Utah have all enacted digital ad taxes in recent years, though Maryland’s Tax Court struck down the state’s digital ad tax in August in a highly followed case. (The state of Maryland has appealed the Tax Court ruling.)

The bit tax has a longer history than digital ad taxes. The 1990s-era bit tax would amount to $80 per gigabyte (GB) today—arguably not the intention of its proponents, but an indicator that taxes that seem reasonable in one era can be completely impractical in the next.

Sen. Kelly’s tax amounts to $0.11 per GB. That may seem small today; chatting with an AI platform may only use one to five megabytes per hour, or 0.001 to 0.005 GB. In a decade or two, though, as AI platforms become more capable, the tax may become prohibitively large.

Excess profits taxes have an even longer and more well-worn history. Federal policymakers have often proposed an excess or “windfall” profits tax on oil and gas producers when energy prices rise. As Tax Foundation’s Alex Muresianu wrote in May 2026:

Several analyses of the 1980 Windfall Profits Tax have found it reduced domestic production and increased reliance on imports. A Congressional Research Service paper found that the tax reduced domestic oil production by between 1.2 and 8.0 percent, and increased reliance on foreign oil by between 3 percent and 13 percent between 1980 and 1988 (when the tax was eventually repealed).

The 1980s tax also brought in a fraction of the revenue projected, $80 billion actual compared to $393 billion projected. Sen. Kelly’s excess profits tax would have a narrow base—potentially, almost entirely one company—making it especially precarious.

The Ad, Bit, and Excess Profits Taxes Are Categorically Unsound

Sound tax policy is neutral, simple, transparent, and stable. Kelly’s proposed taxes are narrowly targeted, complicated to administer, and rest on tax bases that can change wildly due to growth, technological change, or tax avoidance.

The digital ad tax is a gross-based tax on a business input. The US has opposed similar digital service taxes abroad. These types of taxes distort production, can pyramid through supply chains, encourage vertical integration and other inefficient restructuring, and fall hardest on low-margin businesses.

Academics and policymakers have expressed concerns that AI is built on the uncompensated work of human creators—authors and artists, for example. A closely related economic concern is that AI chatbots will divert traffic and reduce monetization opportunities, leading to an underproduction of human-created content. Since many creators ultimately derive revenues from digital ads, a digital ad tax would be partially borne by these content creators. In short, a tax introduced in a bill to cushion the effect of AI on workers would ironically be partially borne by workers threatened by AI.

The structure of the bit tax is similar to the 1990s versions, which were proposed as “dollars per unit of data” taxes. If adopted, these would have raised little revenue at first, but would have quickly become an astronomical tax burden as internet traffic doubled annually and costs collapsed. If AI use takes off in a way similar to the internet—and AI costs per unit collapse—the proposed bit tax would quickly turn into a substantial burden on the adoption of this technology.

Changing the bit tax to an ad valorem tax, such as in the token tax proposal by Rep. Greg Casar (D-TX), would largely fix the problem of an escalating burden. However, it would still distort production away from the use of AI models and toward less efficient input choices. The bit tax would also add significant tax complexity, especially in policing technical avoidance strategies, including service offshoring.

The excess profits tax applies a high rate on a narrow base. The tax would likely be paid by few firms—our analysis suggests that in 2027, almost all of the tax would be paid by one firm. Singling out certain firms for additional tax based on arbitrary financial conditions and industry participation is non-neutral tax policy. The tax could also be avoided through firm restructuring or financial engineering, ultimately causing economic distortions while raising much less revenue than our estimates. Finally, insofar as it is paid, the tax captures the payoff to successful innovation at a high rate while the costs deducted earlier at a lower rate; that intertemporally asymmetric treatment would deter innovation.

Big Picture

While Sen. Kelly differs from other lawmakers in attempting to pay for new spending under his Make AI Work for Americans Act, his proposed AI taxes miss the mark. The digital ad tax is inefficient and complex, potentially targeting revenue that has little or nothing to do with AI. The bit tax may not stand the test of time, especially as consumer- and business-facing technologies grow and data usage increases. And the excess profits tax suffers from a very narrow base, meaning actions of just one or a few companies could render the tax moot.

Lawmakers seeking to pay for new spending—or reduce unsustainable federal deficits and debt—would do better to broaden the income tax baseThe tax base is the total amount of income, property, assets, consumption, transactions, or other economic activity subject to taxation by a tax authority. A narrow tax base is non-neutral and inefficient. A broad tax base reduces tax administration costs and allows more revenue to be raised at lower rates., reform the corporate income taxA corporate income tax (CIT) is levied by federal and state governments on business profits. Many companies are not subject to the CIT because they are taxed as pass-through businesses, with income reportable under the individual income tax., or tax consumption instead of income. Several of these options and more are featured in Tax Foundation’s Options for Reforming America’s Tax Code 3.0.

Appendix: Methodology for Estimates

We estimate the revenue effects of the three proposed taxes without capturing any behavioral feedback. We multiply the projected tax base by the proposed tax rate. Because the estimates do not account for the responses of taxpayers—including tax avoidance—they should be treated as illustrative.

Our revenue estimate for the digital ad tax starts with the Interactive Advertising Bureau’s (IAB) revenue report, which puts 2025 US internet ad revenue at $294.6 billion. The vast majority of ad revenues (92 percent) came from the top 25 companies, which serves as our base for the digital ad tax; five percent of that is about $13.6 billion. We checked this by producing a separate estimate from the SEC filings of public companies, which yields a similar estimate of about $14 billion. Both approaches likely understate the tax base (due to considerations of gross vs. net ad revenue, and ad revenue missing from our SEC filing search), so we start with the $14 billion estimate and increase it by 20 percent. Grown at the IAB’s average annual growth rate for 2021 to 2025 (about 12 percent), revenue is about $21.1 billion in 2027, or $15.8 billion after the 25 percent income and payroll taxA payroll tax is a tax paid on the wages and salaries of employees to finance social insurance programs like Social Security, Medicare, and unemployment insurance. Payroll taxes are social insurance taxes that comprise 24.8 percent of combined federal, state, and local government revenue, the second largest source of that combined tax revenue. offset, and about $278 billion over 10 years after the offset.

Our revenue estimate for the bit tax begins with reported AI image generation from OpenAI and Google, which total about 225 billion images per year. We assume images are the main source of the tax in the near term; switching to other mediums, such as video or streaming video games, is baked into assumed growth rates. At about 2.5 MB per image, that is about $62 million of tax before the exemption. We assume that all consumers would fall below the 10 GB exemption amount and be exempt, but that businesses would often have usage above the exemption amount and be taxed. About a quarter of Google’s token usage is through its API. We take this as a proxy for the business usage share, so we assume a quarter of usage would fall above the 10 GB per-user exemption. We assume usage doubles annually, as internet traffic did for about a decade, yielding about $30.9 million in 2027, or about $23.2 million after the 25 percent income and payroll tax offset. Continuing to assume an annual doubling, the tax would raise $24 billion over 10 years after the offset.

We assess that the excess profits tax affects only a handful of companies. Using SEC filings, we estimated taxable incomeTaxable income is the amount of income subject to tax, after deductions and exemptions. Taxable income differs from—and is less than—gross income.   for publicly traded companies and identified those above the 40 percent of receipts threshold. Private AI labs appear to be loss-making, though a few private firms may owe tax we do not capture. We estimate the tax would collect about $7.6 billion in 2027 (almost all from one company) under a narrow—and our preferred—reading of the bill, and about $11 billion under a broad one. We apply a 13.6 percent offset for this tax, rather than the 25 percent offset for the other two taxes. This 13.6 percent offset is the ratio of corporate tax receipts 2027-2036 over corporate taxable income from CBO’s February 2026 projections. This offset is justified under the assumption that the excess profits tax is deductible against the corporate income tax and therefore reduces corporate income tax receipts. Applying this offset reduces 2027 revenue to $6.5 billion. Grown at four percent annually, in line with nominal GDP, it raises about $79 billion over 10 years, after the offset.

Share this article