Amid calls for regulation, many lawmakers want to tax artificial intelligence. But the playbook from which they’re working increasingly draws on some of the worst ideas from the early years of the internet. If we want to ensure that AI fuels future growth, we shouldn’t be pursuing tax policies that are stuck in the past.
There is no shortage of proposals coming out of Congress to mitigate AI’s purported threats. From Vermont Senator Bernie Sanders’s AI Sovereign Wealth Fund to Massachusetts Senator Elizabeth Warren’s call for an excise tax on data centers, the plans appear to express policymakers’ fears that, if AI is to become a dominant part of our economy, the tax code must have a say in shaping its future.
One proposal currently gaining steam is to place a new tax on tokens, the discrete units of data that large-language models like ChatGPT or Claude process when interpreting and generating text. Microsoft founder Bill Gates is on board—in fact, he’s been a backer of taxing automation for nearly ten years, long before the integration of AI into daily life. Other billionaires, including OpenAI CEO Sam Altman, are also supportive. The idea officially has legs in Congress, too: Texas Representative Greg Casar and California Representative Sara Jacobs have introduced legislation to create a “token tax.”
The idea of a token tax traces back to the early days of the internet in the 1990s. In 1995, Canadian economist Arthur Cordell warned that the web would displace a great deal of labor, doing things better, faster, and cheaper than human workers. To address these concerns, he proposed a “bit tax,” attaching a modest number to the design: a $0.00000001 fee, or one hundred-millionth of a dollar, per bit of storage. In practical terms, that would have amounted to about eight cents of tax on one megabyte of data.
Imagine what the world would look like if Cordell’s idea had come to fruition. Streaming one hour of music currently uses about 30 megabytes of data, meaning you would pay an extra $2.40 just to listen to your favorite album. An hour-long video conference at work uses nearly 500 megabytes of data, meaning it would cost your company $40 more just to host a meeting.
Similarly, token tax costs would rise with usage, which has done nothing but skyrocket. These fees would quickly add up for AI companies, whose entire design is reliant on data. But consumers could easily shoulder an even heavier burden.
Part of the reason our economy was able to benefit from the internet—and the reason we can stream music and host video conferences so readily today—is that ideas like the bit tax were never enacted. Throughout the 1990s, it drew widespread, bipartisan opposition precisely because political leaders feared it would diminish the promise of the internet. Congress and President Bill Clinton tried explicitly to prohibit states from enacting bit taxes in the Internet Tax Freedom Act of 1998 (ITFA). Clinton and other policymakers, including the bipartisan commission responsible for overseeing enactment of ITFA, said the bit tax was buried because it was complex, non-neutral, and posed a threat to economic growth.
The same principles of sound tax policy that kept the bit tax from seeing the light of day in the 1990s should steer today’s debate over AI token taxes. Concerns about how the internet would displace workers never came to pass—but imagine where our economy would be today without it. Rural Americans enjoy faster access to doctors through telehealth, millions of workers can raise their families anywhere they wish because of remote work, and multitrillion-dollar industries have arisen thanks to the free flow of information on the web.
Tax policy should be principled, not reactionary. It should support the same kinds of innovation that have turned the United States into a global leader in AI—and enable individuals and businesses to find ways to use it to accelerate economic growth. Policymakers should recognize the dangers that misguided and punitive taxes pose for the future of our new economy.
