An AI token tax is on the table: what should you do?

The idea of an AI token tax is gaining traction in policy discussions. As AI, and specifically AI agents, begin to perform more productive work traditionally done by humans, governments are exploring ways to replace lost income tax revenue. This post breaks down the arguments for and against taxing AI usage and what it could mean for your business's cost structure, competitive positioning, and innovation strategy.

In a recent episode, the AI Daily Brief explored the escalating debate over implementing a tax on AI model usage, specifically on the 'tokens' that serve as the basic unit of computation for Large Language Models (LLMs).

This isn't a fringe idea. It's being discussed by prominent figures, from tech executives like Anthropic's CEO Dario Amodei to entrepreneurs like Mark Cuban and politicians like US Senator Elizabeth Warren. The core premise is straightforward: if human labour is taxed, and AI agents start performing a significant amount of that labour, the tax base could erode. Taxing the AI's 'work'—measured in tokens—is proposed as a potential solution.

The arguments for and against

The discussion highlights a clear tension between fiscal responsibility and technological progress. Proponents argue that an AI tax could help fund social safety nets, such as retraining programs or a Universal Basic Income (UBI), for workers displaced by automation. It's seen as a way to ensure the productivity gains from AI are shared more broadly across society.

The counterarguments are just as compelling, focusing on the practical and economic drawbacks of such a tax.

Essentially, the debate pits the need to manage the societal transition against the risk of throttling the very technology that promises significant productivity breakthroughs.

For business owners and operators, this is more than an academic debate. The introduction of an AI token tax would have direct, material consequences on the adoption and application of agentic AI.

A new line item on your P&L

A token tax would represent a new, direct, and variable cost on AI-driven workflows. Currently, businesses calculate the ROI of an AI agent based on:

• API costs (e.g., cost per million tokens from OpenAI or Anthropic).
• Infrastructure costs (if self-hosting).
• The productivity gain or cost saving achieved.

A token tax inserts a new variable into this equation. A small tax might be easily absorbed, but a more substantial one could render many borderline use cases uneconomical. This would force a re-evaluation of every workflow you're considering for automation, from customer service bots to automated code generation.

The drive for token efficiency

A tax would create a powerful new incentive to optimise for token efficiency. Today, the focus is often on getting the best possible output. Tomorrow, it might be on getting an acceptable output for the fewest possible tokens. This has implications for several areas:

• Model Selection: You might choose a smaller, less powerful model (Claude 3 Haiku instead of Opus, for example) for tasks where top-tier reasoning isn't essential, purely to minimise tax exposure.
• Prompt Engineering: The discipline of crafting concise, efficient prompts would become even more critical, moving from a best practice to a core cost-control measure.
• Workflow Design: Businesses might favour simpler, single-shot AI tasks over more complex, multi-step agentic workflows that consume significantly more tokens.

Competitive and strategic risks

How a tax is implemented could create significant market distortions. If Australia introduced a tax but other nations did not, local businesses could be at a competitive disadvantage when competing globally. It could also create a divide between:

• Large Enterprises vs. SMEs: Larger organisations may have the resources to absorb the tax or invest in complex token-optimisation strategies, while smaller businesses find the cost prohibitive.
• Proprietary vs. Open-Source: If the tax only applies to major API providers, it could inadvertently push companies towards self-hosting open-source models, creating a different set of technical and security challenges.

The biggest risk is that a tax on usage punishes the experimentation necessary to unlock the true potential of agentic AI. The most valuable applications of AI may not be obvious, and discovering them requires trial and error—which means burning through a lot of tokens.

While an AI token tax is not imminent in Australia, the global conversation is serious enough that proactive businesses should start preparing. Here are four practical steps to take now.

1. Start monitoring the policy debate. Pay attention to discussions from government bodies, industry groups, and major AI labs. Understanding the direction of regulatory thinking will help you anticipate changes rather than just react to them. This is no longer a purely technical domain; it has become a matter of public policy.

2. Audit and benchmark your AI usage. You cannot manage what you do not measure. Begin tracking your organisation's token consumption. Identify which business units, applications, and workflows are the most token-intensive. This data will form the baseline for any future cost modelling and optimisation efforts.

3. Model the financial impact. Once you have a usage baseline, run simple scenarios. What would a 1%, 5%, or 10% tax on your current token consumption look like in dollar terms? How would it affect the ROI of your key AI initiatives? This analysis will be invaluable for making informed investment decisions and for internal advocacy.

4. Instil a culture of token efficiency. Regardless of whether a tax is ever implemented, optimising token usage is a smart financial decision. Encourage your teams to:

   • Use the smallest, cheapest model that is effective for a given task.
   • Develop strong prompt engineering skills to get the desired output with less iteration.
   • Implement caching for common queries to avoid redundant API calls.
   • Evaluate workflows to see if tasks can be achieved with simpler, non-AI automation.

Treating AI usage as a resource to be managed carefully will not only prepare your organisation for potential regulatory costs but will also improve the financial performance of your AI strategy today.