Candice Horbacz
As US investment in artificial intelligence (AI) accelerates at an unprecedented pace, so too have proposals to uniquely tax AI technologies and their developers. These nascent discussions, driven by concerns over potential job displacement, wealth concentration, and the societal impact of advanced AI, find a striking historical parallel in the late 1990s debate surrounding the "bit tax" – a proposal to tax information flowing across the nascent World Wide Web. Understanding the rise and fall of the bit tax offers critical lessons for policymakers grappling with the complexities of taxing a rapidly evolving technological frontier like AI.
The current landscape of AI tax proposals is diverse, reflecting various approaches to addressing the perceived challenges and opportunities presented by this transformative technology. Some legislative ideas are broadly aimed at large corporations and the wealthy, seeking to capture a portion of the immense profits generated by the tech sector, including leading AI developers. Others are far more narrowly targeted, focusing directly on specific elements of AI operations. These include taxing the "tokens" — the fundamental units of information processed by large language models — flowing in and out of AI systems, or levying charges on the "compute" — the vast computational power and energy used to train these complex AI models. For instance, Senator Elizabeth Warren has been a vocal proponent of taxing large corporations to invest in people, a stance that could encompass highly profitable AI entities. Similarly, discussions around taxing specific AI inputs or outputs have emerged from various think tanks and policy circles, often drawing comparisons to an "energy tax" on data centers or a "transaction tax" on digital information.
These contemporary proposals share a distinct lineage with the 1990s-era bit tax, an idea born out of similar anxieties regarding the internet’s profound, yet then-unforeseen, impacts on the economy and society. The bit tax, as originally conceived, would have applied to uploads and downloads of information over the internet. Its proponents envisioned it as a mechanism to fund public services, address potential job losses due to automation (even then, a concern), or simply ensure that the burgeoning digital economy contributed its fair share to the public coffers. Critics, however, swiftly likened it to an "email tax" or a "toll on information," arguing it would stifle innovation and disproportionately burden internet users and businesses. The proposal drew widespread bipartisan opposition in the US, primarily on the grounds of being overly complex, non-neutral in its application, and fundamentally anti-growth.
Had the bit tax been enacted and maintained as originally proposed, its long-term consequences could have been dire. It might have either "soaked" consumers and businesses in extraordinarily high taxes today, acting as an prohibitive toll on internet usage, or, perhaps more damagingly, it could have actively prevented the development of new, data-intensive technologies that now form the backbone of modern society. Imagine a world without seamless video streaming, cloud computing, or even complex search engines, all of which rely on the unfettered flow of massive amounts of data. While the economic, technological, and budgetary outlook now differs significantly from that of the 1990s, policymakers today can still draw invaluable lessons from the historical trajectory of the bit tax, particularly as they navigate the intricate landscape of AI taxation.
The Genesis of the Bit Tax: A 1990s Vision
The original bit tax proposal emerged from the insights of Canadian economist Arthur Cordell. He formally presented the idea at a 1995 conference, an era when the internet was still largely a nascent phenomenon for the general public, though its transformative potential was already beginning to be recognized by visionaries. Cordell’s remarks from that period bear a striking resemblance to the conversations happening today around AI, highlighting a recurring pattern in technological disruption and societal response. He posited that the internet, like any major economic force, should contribute to the public good and that its unique characteristics might warrant a novel form of taxation.
Cordell suggested a seemingly minuscule tax rate of 0.000001 cents per bit. While this figure appears negligible on its own, its cumulative effect on the vast quantities of data transacted daily would have been substantial, even in the 1990s. To illustrate the scale, Table 1 below demonstrates how such a tax would apply to various digital storage amounts, emphasizing the exponential nature of data and the potential burden:
Table 1. A Crosswalk from Bits to (Giga)bytes
| Digital Storage Amount | Equivalent Bits | Bit Tax (at 0.000001 cents/bit) |
|---|---|---|
| 1 Byte | 8 bits | $0.00000008 |
| 1 Kilobyte (KB) | 8,192 bits | $0.00008192 |
| 1 Megabyte (MB) | 8,388,608 bits | $0.08388608 |
| 1 Gigabyte (GB) | 8,589,934,592 bits | $85.89934592 |
| 1 Terabyte (TB) | 8,796,093,022,208 bits | $8,796.093022208 |
Source: Ashley Taylor, “Bits and Bytes,” Stanford University, 2018, https://web.stanford.edu/class/cs101/bits-bytes.html; author’s calculations.
Cordell, a forward-thinking economist, acknowledged in a subsequent 1997 speech that, given the relentless pace of technological advances, "the bit tax rate will have to be adjusted for changing times." This foresight underscores a fundamental challenge in taxing rapidly evolving technologies: what makes sense today might be obsolete or economically destructive tomorrow. He also proposed that the revenue generated from such a tax could be directed towards specific public investments, such as universal internet access, digital literacy programs, or compensation for workers displaced by automation. This aspect of the proposal aimed to mitigate the negative externalities of technological progress, a concept that resonates strongly with today’s discussions around AI and its potential societal impacts.
The idea of a bit tax gained considerable traction in the late 1990s, transcending academic circles and entering the realm of international policy discussions. A significant milestone was a 1999 United Nations report which mentioned a $0.01 per megabyte tax as one potential mechanism to "fund the global communications revolution," particularly in developing nations. This reflected a broader international interest in finding new revenue streams to address global disparities and support technological infrastructure development. The European Commission also explored the idea, indicating that the concept was not merely a North American phenomenon but a global consideration for policymakers grappling with the fiscal implications of the digital age. This period saw intense debates in various international forums, with many nations keen on finding ways to ensure the digital economy contributed to national and international public goods.
The Demise of the Bit Tax: Factors Behind Its Failure
Despite the initial intellectual curiosity and global policymaker interest, the bit tax never came close to enactment. Its journey from proposal to political non-starter was swift and decisive, particularly in the United States. The idea drew strong and bipartisan opposition across the political spectrum. President Bill Clinton, in a pivotal statement in 1997, articulated a clear stance, declaring his intent to keep the internet "free of new discriminatory taxes." This presidential opposition signaled a broader consensus that emerging digital technologies should not be burdened with unique levies that could impede their growth and adoption.
Further solidifying this opposition, an early and bipartisan version of the Internet Tax Freedom Act (ITFA) explicitly prohibited states and localities from enacting "bit taxes." The ITFA, initially passed in 1998, was a landmark piece of legislation designed to foster the growth of the internet by preventing a patchwork of state and local taxes on internet access and electronic commerce. Its explicit inclusion of a ban on bit taxes highlighted the widespread concern among lawmakers about the potential for such taxes to stifle innovation and burden consumers. The bipartisan Advisory Commission on Electronic Commerce, established by the final version of ITFA, further reinforced this sentiment. In its comprehensive report to Congress in 2000, the commission concluded that the bit tax was "met with little support by government officials," effectively delivering the final blow to the proposal’s viability.
Three primary factors converged to bury the bit tax:
- Technological Impracticality and Complexity: The sheer technical challenge of accurately measuring and taxing every "bit" of information flowing across networks was daunting. The infrastructure required for such granular tracking would have been immensely costly and complex to implement, administer, and enforce. As data volumes exploded, the administrative burden would have quickly become insurmountable. Moreover, defining what constituted a "taxable bit" and distinguishing it from non-taxable internal network traffic or encrypted data presented formidable legal and technical hurdles.
- Economic Disincentive and Non-Neutrality: Economists and policymakers widely argued that the bit tax would act as a significant disincentive to internet usage, innovation, and economic growth. By adding a direct cost to every unit of data, it would have disproportionately affected data-intensive applications and services, making them more expensive or even unfeasible. This violated the principle of tax neutrality, which suggests that taxes should not distort economic decisions or favor one type of activity or industry over another without a clear policy objective. Instead, the bit tax would have discriminated against digital services compared to traditional, non-digital ones.
- Bipartisan Political Opposition: The consensus that emerged in the US, championed by both Democratic and Republican leaders, was that the internet should be allowed to grow unhindered by unique and potentially crippling taxes. This unified front, driven by a vision of the internet as a engine for future prosperity, proved insurmountable for the bit tax proposal. Policymakers feared that such a tax would slow down internet adoption, hinder e-commerce, and put American companies at a disadvantage globally.
Lessons for AI Taxes: A Cautionary Tale from the Digital Frontier
Although bit taxes were last seriously considered nearly 30 years ago, their history offers profound and relevant lessons for policymakers contemplating AI taxes today. The parallels between the nascent internet of the 1990s and the rapidly advancing AI landscape of the 2020s are striking, particularly concerning the challenges of taxing dynamic and foundational technologies.
Technology Changes Quickly
One of the most critical lessons is that a bespoke tax designed for a specific technological stage may quickly become obsolete, impractical, or even economically destructive as the technology evolves. In the 1990s, the concept of a "bit" seemed like a stable, quantifiable unit. However, the internet’s explosive growth and the exponential increase in data consumption quickly rendered the bit tax concept untenable. Consider that the median US household today uses an astonishing 532 GB of data per month, according to recent estimates. If the original bit tax rate were applied, such a household would face extraordinarily high, prohibitive taxes, amounting to over $45,000 per month. Cordell, in his original proposal, certainly did not intend that kind of crippling impact, underscoring the difficulty of predicting future scale.
Few in the 1990s could have envisioned seamless high-definition video calling available across continents, or thousands of movies and TV shows available for streaming at the touch of a button, or even the vast cloud infrastructure that underpins modern computing. Similarly, a "compute" or "token" tax measured today could very well end up looking nonsensical in 10, 5, or even just 2 years. AI models are becoming more efficient, requiring less raw "compute" for equivalent performance, or new architectures might emerge that render "tokens" an irrelevant metric. Taxing the number of calculations performed (FLOPS) or the number of data points processed could quickly become a moving target, leading to unintended consequences and stifling innovation in AI development. Table 2 below illustrates the potential bit tax liability for common internet uses today, highlighting the sheer impracticality:
Table 2. The Bit Tax Applied to Common Internet Uses Today
| Common Internet Use (per month) | Estimated Data Use | Bit Tax (at 0.000001 cents/bit) |
|---|---|---|
| Streaming HD Video (50 hours) | ~150 GB | ~$12,884.90 |
| Online Gaming (50 hours) | ~25 GB | ~$2,147.48 |
| Video Conferencing (50 hours) | ~10 GB | ~$858.99 |
| Social Media Browsing (50 hours) | ~5 GB | ~$429.50 |
| General Web Browsing (50 hours) | ~2 GB | ~$171.80 |
| Median US Household (total) | 532 GB | ~$45,702.86 |
Source: AT&T, “Common reasons for high data use,” Apr. 29, 2025, https://www.att.com/support/article/wireless/KM1045105/; Victra, “What Uses the Most Data on Your Phone?” Jun. 28, 2026, https://victra.com/blog/what-uses-the-most-data-on-your-phone/; author’s calculations.
Neutrality Is Timeless
Another enduring principle from the bit tax era is the importance of tax neutrality. In the 1990s, a bipartisan group of executive and legislative branch officials in the US firmly stood on the principle that internet products and services should neither face discriminatory taxes nor gain an unfair advantage over brick-and-mortar goods and services. This principle guided the development of internet tax policy for decades. The "online advantage" with respect to sales taxes, where internet retailers often did not collect sales tax, was a significant issue that states and the federal government wrestled with for years. The Supreme Court’s landmark Wayfair decision in 2018 ultimately settled the legal matter, in part by applying the principle of neutrality, mandating that remote sellers collect sales tax, thereby leveling the playing field with physical retailers.
While some AI tax proposals are framed as broad tax increases on wealth or capital, which could be argued to apply neutrally across various sectors, many others are narrowly targeted at specific AI companies or activities. Such targeted taxes, by their very nature, are non-neutral. They can distort markets, create artificial incentives or disincentives, and potentially disadvantage domestic AI companies compared to international competitors operating under different tax regimes. Lawmakers today should adhere to the same principles of neutrality that underpinned the opposition to bit taxes in the 1990s. A tax system that applies consistently across different industries and forms of economic activity is generally considered more efficient and equitable, minimizing unintended consequences and fostering genuine innovation.
AI Taxes May Adversely Affect Growth
Perhaps the most significant cautionary lesson is the potential for narrowly targeted taxes to adversely affect economic growth and innovation. If the US had enacted a bit tax, Americans today might not enjoy the widespread benefits of remote work, accessible telehealth services, or the broader social connectedness fostered by high-speed, affordable internet. Even at a much lower rate, say $0.01 per GB (which would still amount to over $5 per month for the median household, and significantly more for businesses), a bit tax would have added friction to data exchange, potentially slowing down activities that create jobs, drive productivity gains, and grow the economy. It could have deterred investment in internet infrastructure and data-intensive applications, hindering the very digital transformation that has propelled economic expansion for the past three decades.
The labor displacement that Cordell and others warned of in the 1990s, while a persistent concern, ultimately did not materialize on the scale many feared. Instead, the rise of the internet brought with it a massive boom in new industries, jobs, and financial activity, leading to a significant increase in tax revenues that, at least temporarily, improved the US economic and budgetary outlook in the late 1990s and early 2000s. While AI undoubtedly presents unique challenges and opportunities, imposing targeted taxes could similarly stifle its potential to drive new economic growth and job creation in unforeseen sectors.
Distinguishing the Eras: 1990s Internet vs. 2020s AI
It is crucial to acknowledge that the bit tax episode, while instructive, cannot teach us everything about AI. The 2020s are fundamentally different from the 1990s. US economic growth is generally slower today than it was back then, influenced by demographic shifts, lower productivity growth, and other macroeconomic factors. Furthermore, the US budget is significantly more vulnerable to major threats to its tax base, with national debt levels soaring and persistent deficits. These factors create a more urgent political imperative to find new revenue streams.
Moreover, AI could indeed be a far more powerful and disruptive technology than the World Wide Web, with potentially deeper societal implications, including more profound impacts on the labor market. While the internet created new jobs and industries, AI’s ability to automate complex cognitive tasks could lead to different patterns of job displacement and creation. The ethical considerations surrounding AI, such as bias, accountability, and autonomous decision-making, also introduce layers of complexity that were not as prominent in the early internet debates. The scale of investment in AI, currently in the tens of billions annually from both private and public sectors, suggests a technology with immense, yet still largely unquantifiable, economic potential and risk.
However, despite these differences, the bit tax can teach us about the timelessness of core tax principles: simplicity, neutrality, transparency, and stability. A good tax system should be easy to understand and administer, apply fairly across similar economic activities, be clear about its purpose and impact, and provide a predictable environment for businesses and individuals. These principles are not mere academic constructs; they are practical guidelines for designing tax policies that foster economic prosperity rather than impede it.
Policymakers should consider the 1990s bit tax a cautionary tale when evaluating complex or non-neutral AI tax proposals in the 2020s. While the impulse to tax a burgeoning, highly profitable sector is understandable, especially in times of fiscal strain, the method of taxation can have profound and lasting consequences. A narrowly targeted, technologically specific tax risks becoming quickly outdated, administratively burdensome, and ultimately detrimental to innovation and economic growth. Instead, a more prudent approach might involve examining how existing tax frameworks can be adapted, or considering broader, neutral reforms that apply across the economy, rather than singling out a foundational technology still in its formative stages. Learning from history can help ensure that the pursuit of revenue does not inadvertently stifle the very innovation that could drive future prosperity.
