A 1990s proposal to tax internet data flows at 0.000001 cents per bit—or roughly $84 per gigabyte—offers lessons for lawmakers now considering special taxes on artificial intelligence, according to an analysis published by the Tax Foundation. The report examines how the "bit tax," which drew bipartisan opposition and never became law, mirrors current proposals to tax AI companies, the tokens used by AI models, or the computing power needed to train them. Author Andrew Lautz argues that the principles that killed the bit tax—simplicity, neutrality, and growth concerns—remain relevant today.
Canadian economist Arthur Cordell introduced the bit tax concept at a 1995 conference, warning that the internet could "replace people in a great number of functions" and threatening Canada's tax base through labor displacement. He suggested levying the tax on interactive digital communications like emails and calls, collected by telecom carriers based on local averages rather than individual use. By 1999, a UN report mentioned a $0.01 per megabyte rate as one funding mechanism for the "global communications revolution," and the European Commission explored the idea as well. President Clinton declared in 1997 that he wanted to keep the internet "free of new discriminatory taxes," and the bipartisan Internet Tax Freedom Act explicitly barred states from imposing bit taxes. The bipartisan Advisory Commission on Electronic Commerce reported to Congress in 2000 that the bit tax proposal "met with little support by government officials."
Applied to current internet usage, Cordell's original rate would create absurd burdens: the median American household, which uses 532 GB monthly, would face extraordinarily high levies. Even a single hour of Netflix streaming at standard definition—consuming 1 GB—would trigger an $84 tax under the original formula. An hour of high-definition video would cost $252, while uploading 100 photos to social media would run $84. The report notes three factors buried the bit tax: complexity in measuring and enforcing cross-border data flows, non-neutrality by taxing digital interactions differently than in-person ones, and fears it would throttle innovation and economic expansion tied to the internet.
The bit tax failed because policymakers recognized that a specialized levy aimed at emerging technology could either crush consumers with costs or prevent data-intensive innovations from developing at all. According to Lautz, a bespoke tax that seems reasonable in one era "may not make sense in the next"—a compute or token tax calibrated today "could end up looking nonsensical in 10, 5, or even 2 years." The labor displacement Cordell feared never materialized; instead, the internet sparked a financial and tax revenue boom that temporarily strengthened the U.S. economic and budgetary position in the late 1990s and early 2000s. Had the bit tax passed, Americans might not enjoy remote work, telehealth, or the connectivity of high-speed internet today.
The report concludes that while the 2020s differ from the 1990s—economic growth is slower, the budget faces bigger threats, and AI may prove more disruptive than the web—the core tax principles of simplicity, neutrality, transparency, and stability remain timeless. Lawmakers evaluating narrow or non-neutral AI tax proposals should treat the bit tax as a cautionary tale, one that warns against complexity and discrimination in favor of broad-based, neutral taxation. The episode can't predict AI's full impact, but it can pump the brakes on targeted levies that won't stand the test of time.