Bill Gates revived his robot‑tax idea in an essay posted to Gates Notes in August 2026 and added a novel token‑usage levy on generative‑AI models. The proposal, reported by Les Numériques, calls for a tax on machines calibrated to the payroll taxes and social contributions a displaced worker would have paid, plus a charge per data token processed by AI systems.
Two‑part tax proposal
The French tech news site quotes Gates verbatim: “Gates avance deux mécanismes. Le premier est une taxe sur les robots, calée sur les impôts et cotisations qu'aurait générés le salarié remplacé. Le second, plus inédit, est une taxe sur les jetons, ces fameux tokens, ces unités de données que traitent les modèles d'IA à chaque requête.”Les Numériques The robot tax would be set at the level of the taxes and contributions that the replaced employee would have generated. The token tax would charge companies for each data token processed by generative‑AI models.
Fiscal mechanics and sector context
Gates frames the robot tax as a way to neutralise the fiscal advantage of buying a machine instead of hiring a person. As the article explains, “Une entreprise a le choix. Pour accomplir une tâche, elle peut embaucher une personne, et payer alors des charges sociales et des cotisations. Ou acheter une machine, un logiciel, une IA, et déduire aussitôt cette dépense de ses impôts.”Les Numériques The second part – a token tax – targets the computational work of large language models, measured in the number of tokens they generate per request.
Both mechanisms aim to address a “déficit fiscal” created when automation reduces payroll tax revenue. By tying the levy to the same fiscal base that a human employee would have contributed, the robot tax seeks to level the playing field for firms that automate tasks.
Implications for large tech firms
Microsoft, the world’s second‑largest software company, provides a useful scale reference. Its most recent Form 10‑K filing (filed 29 July 2026) shows a net income of USD 133.749 billion for the fiscal year ended 30 June 2026, total assets of USD 758.376 billion, shareholders’ equity of USD 442.387 billion, and 7.427 billion shares outstanding.SEC filing The company’s revenue in 2011 (the only earlier figure supplied) was USD 36.148 billion for the quarter ending 31 December 2010.SEC filing
| Metric | Value (USD) | Period end |
|---|---|---|
| Net income (FY 2026) | 133.749 billion | 30 Jun 2026 |
| Total assets (FY 2026) | 758.376 billion | 30 Jun 2026 |
| Shareholders’ equity (FY 2026) | 442.387 billion | 30 Jun 2026 |
| Shares outstanding (FY 2026) | 7.427 billion shares | 30 Jun 2026 |
| Revenue (Q4 2010) | 36.148 billion | 31 Dec 2010 |
| Source: Microsoft Form 10‑K (2026) and Form 10‑Q (2011), SEC EDGAR. | ||
With earnings of over $130 billion, a robot‑tax calibrated to payroll contributions could represent a material cost if a substantial share of Microsoft’s cloud‑based AI services were re‑engineered to replace human labour. The token tax, measured per data token, would directly affect the pricing of Azure OpenAI services, which bill customers by the number of tokens processed.
Other large‑cap U.S. tech firms – Alphabet, Amazon, Meta – face similar exposure. Their AI‑driven products already generate billions of tokens daily; a per‑token levy would add a variable cost component that scales with usage, potentially reshaping pricing models and profit margins.
Open questions and next steps
The proposal remains an essay, not legislation. Gates does not specify the exact tax rate, the administrative mechanism for counting tokens, or how the tax would be collected across jurisdictions. The August 2026 essay is the only source confirming the two‑part design; no government body has yet signaled adoption.
Key uncertainties include:
- How regulators would define a “robot” for tax purposes – hardware, software, or a combination.
- Whether token counting would rely on existing usage logs from cloud providers or require a new reporting standard.
- How the tax would interact with existing R&D incentives and depreciation schedules.
Stakeholders are likely to test the idea in policy circles. The European Union’s 2017 rejection of a robot tax shows the difficulty of achieving consensus, but Gates’ stature may revive the debate, especially as AI‑driven layoffs accelerate worldwide.
For investors, the proposal adds a new variable to the cost structure of firms that monetize AI at scale. Analysts will need to model potential tax liabilities under different adoption scenarios, factoring in the proportion of revenue derived from token‑intensive services.
Until a concrete legislative framework emerges, the robot‑and‑token tax remains a thought experiment. Its merit will be judged on whether it can balance fiscal equity with the need to preserve incentives for automation and AI innovation.