1. Governments face growing pressure to overhaul tax systems as artificial intelligence threatens to erode the revenue base built around human labor. Proposals range from taxing robots and AI services to levying charges on computing power and tech company profits, but policymakers remain divided, with critics warning that targeted taxes could deter investment and prove difficult to enforce. [para. 1][para. 2]
2. The deeper challenge is adapting existing tax systems if AI shifts income from wages toward corporate profits and capital gains. For China, where value-added tax (VAT) is the largest source of tax revenue, the debate raises questions about whether consumption taxes could play a greater role in financing public services and social security. [para. 3]
3. Bill Gates proposed taxing robots in 2017 and renewed the call in August, adding levies on AI tokens; OpenAI's April policy blueprint suggested exploring AI and robot taxes alongside higher corporate income and capital gains taxes. These proposals reflect fears that automation could reduce payroll tax and social security contributions while increasing demand for unemployment assistance and retraining, though the exact scale of AI's employment impact remains uncertain. [para. 4][para. 5]
4. Dedicated AI tax proposals fall into three categories: taxing AI use, taxing AI companies, and charging for environmental and social costs. Use-based taxes could target automated equipment, the number of jobs displaced, or the volume of AI services consumed—such as a token tax collected with help from cloud computing providers in auditing usage. [para. 6][para. 7]
5. Token taxes could leverage existing billing infrastructure and allow governments without major domestic AI companies to collect revenue from AI services consumed within their economies. However, token consumption does not necessarily reflect computing resources required, economic value generated, or workers displaced, and costs could be passed on to customers through higher prices. Other proposals target computing resources directly, including FLOPs used in AI calculations, excess profits of AI companies, or electricity consumed by data centers. [para. 8][para. 9]
6. Each approach faces obstacles. Defining which technologies qualify as AI, measuring their productivity contribution, and identifying how many jobs they replace would require complex monitoring of corporate operations, and technology-specific taxes could become outdated as models evolve. Wang Dehua of the Chinese Academy of Social Sciences argues that auditing algorithms and devices replacing human labor would be costly and could encourage companies to adopt less efficient technologies to avoid taxation. [para. 10][para. 11]
7. Some researchers instead favor adapting existing systems to capture a greater share of automation-generated wealth, concerned that labor income could account for a declining share of economic output as productivity gains accrue to capital owners. OECD data show personal income tax averaged 23.7% of member tax revenue in 2023, second only to social security contributions at 25.5%. [para. 12][para. 13]
8. Zhang Fan of Peking University argues tax reform should reduce the burden on work while increasing taxation of economic rents arising from platform monopolies and network effects—through higher excess-profit and capital-gains taxes, measures OpenAI also proposed. But taxing capital presents challenges: governments must balance redistribution against investment needs, while cross-border capital and intangible assets complicate profit attribution and enforcement. [para. 14][para. 15][para. 16]
9. Another approach shifts some burden from labor income toward consumption. A February 2026 NBER working paper by Anton Korinek and Lee Lockwood argues consumption taxes could help finance government spending and redistribution as labor's share of income falls, though taxing capital may become more important if advanced AI systems absorb economic resources without generating corresponding human consumption. [para. 17]
10. VAT is already central to China's tax system and, unlike payroll taxes, does not distinguish between output generated by human workers and that produced using machines. Wang argues it could therefore capture some productivity gains from automation, provided output and consumption continue to grow. Zhang likewise suggests VAT and property taxes should account for a larger share of the future tax mix, with VAT potentially playing a greater role in financing social security. [para. 18][para. 19]
11. The challenge extends beyond tax collection to social security systems, which rely heavily on employment-linked contributions and remain vulnerable if AI reduces jobs or pushes workers into underemployment. Zhang calls for broader coverage including medical insurance, pensions, disability support and training, with unemployment protection extending to those whose hours or earnings are reduced by automation. Researchers have also proposed replacing part of social security contributions with a dedicated VAT-based levy, easing the burden on employers and employees while broadening the funding base. [para. 20][para. 21]
AI generated, for reference only