OpenAI Calls for AI Taxes to Protect Social Safety Nets： Industry Implications a

Why are AI giants proactively calling for taxes, representing the most paradoxical strategic shift in tech history?

Direct answer: This is not pure altruism but a deep risk-avoidance and agenda-setting strategy. Leaders like OpenAI recognize that inaction as AI exacerbates unemployment and inequality will ultimately lead to devastating regulatory backlash, social unrest, and consumer boycotts. Proposing a ‘constructive taxation framework’ allows them to seize the moral high ground and discourse power in policy debates, steering regulation toward directions with relatively manageable impacts on their business models, such as taxing ‘automated labor’ or ‘capital gains,’ rather than directly restricting model development or applications.

Historically, technological revolutions have always been accompanied by creative destruction, but the speed and breadth of AI’s disruption are unprecedented. According to estimates by the McKinsey Global Institute, by 2030, up to 800 million jobs globally (approximately 20% of the global workforce) could be affected by automation. The key issue is that traditional social safety nets (such as social insurance, unemployment benefits) primarily rely on taxing ’labor income.’ When labor is largely replaced by algorithms and robots, the tax base will erode like sand in an hourglass.

The implicit sense of crisis in OpenAI’s report lies in the fact that this erosion is nonlinear. Initially, it may affect positions like customer service and data entry, but with the development of multimodal AI and embodied AI, the impact will extend to manufacturing, logistics, and even some professional services. Once a tax revenue gap forms, it will be difficult to fill. Therefore, rather than passively waiting for governments to impose potentially innovation-stifling punitive taxes (such as high corporate income taxes or special taxes on AI models), it is better to proactively propose a scheme linking taxation to ‘incentive measures.’

Behind this strategic shift is a growing consensus within the AI industry on the ‘survival environment.’ Anthropic proposed similar tools as early as October 2025 in its economic policy research. Academics, such as economists at the University of Virginia, have also explored the possibility of taxing ‘compute’ or ‘hardware.’ This forms an interesting alliance: leading AI companies, some economists, and policymakers concerned about fiscal sustainability. Their common goal is to design a mechanism that captures AI’s ‘windfall gains’ while avoiding killing the goose that lays the golden eggs.

The table below compares different potential AI taxation targets and their pros and cons:

mindmap
  root(OpenAI's Call for AI Taxes:<br>Multiple Strategic Intentions)
    (Risk Management)
      (Prevent Devastating Regulatory Backlash)
      (Shape Favorable Public Opinion)
      (Ensure Long-term Operational Stability)
    (Agenda Setting)
      (Steer Tax Base Discussion Away from<br>Core Algorithm Restrictions)
      (Frame the Debate Around<br>"Fiscal Sustainability")
      (Align with Policymakers and<br>Some Economists)
    (Industry Coordination)
      (Set Regulatory Expectations for<br>the Entire AI Industry)
      (Avoid Individual Company Actions<br>Leading to Overall Stigmatization)
      (Collectively Define What Constitutes<br>"Responsible AI Development")
    (Business Model Defense)
      (Design Potential Tax Liabilities as<br>Predictable Operational Costs)
      (Gain Legitimacy for Technology Applications<br>Through Taxation)
      (Secure a Legitimate Position for AI Companies<br>in the Social Contract)

Can AI taxes truly save social safety nets, or are they just a new form of ‘greenwashing’ by tech giants?

Direct answer: It depends on the specific design, enforcement, and use of funds. If AI taxes are merely symbolic or funds are misappropriated, it is undoubtedly a public relations stunt. However, if a transparent, efficient, and earmarked mechanism can be established, linking tax revenue directly to workforce transition programs, social safety net replenishment, and public wealth funds, it has the potential to be a key tool in mitigating technological shocks and maintaining social cohesion. The real test lies in whether tech giants are willing to accept substantive tax rates and whether governments can effectively tax their complex global operations.

A key but easily overlooked proposal in OpenAI’s report is the ‘Public Wealth Fund.’ This is an extremely ambitious concept aimed directly at addressing the wealth concentration exacerbated by AI. The logic is: since AI’s productivity gains are highly concentrated among a few companies and capital owners, a portion should be extracted at the source to establish an investment pool belonging to the entire population. This resembles models like the Alaska Permanent Fund or Norway’s Government Pension Fund Global (GPFG), but with funding sourced from AI profits.

According to research by the International Monetary Fund (IMF), over the past two decades, the share of global labor income in GDP has shown a declining trend, while the share of capital returns has risen. AI is expected to accelerate this trend. Assuming that over the next decade, AI brings an additional 1-1.5 percentage points to global GDP growth annually, but 70% of the gains accrue to capital and high-skilled labor. Then, a modest AI tax (e.g., a 5-10% levy on identified AI-driven windfall profits) could generate hundreds of billions of dollars in global tax revenue annually. If this funding is injected into a public wealth fund, its investment returns distributed per capita could substantially subsidize the livelihoods of impacted groups or provide funding for universal basic income (UBI) experiments.

However, the devil is in the details. First, how is ‘automated labor’ quantified? Is it calculated based on reduced full-time equivalent (FTE) headcount, or as a proportion of AI software licensing fees? The former is difficult to audit, while the latter may not reflect the true extent of replacement. Second, profit shifting by multinational corporations is a persistent challenge. AI services are particularly intangible, making it easier to recognize income in low-tax jurisdictions through intellectual property licensing fees. This requires global coordination, and current progress in international tax reform (e.g., the global minimum corporate tax) is slow.

A more fundamental question is: Does this distract from more urgent issues? For example, immediately strengthening unemployment insurance, massively expanding free vocational retraining programs, and reforming the education system to cultivate core competencies for the AI era. Taxation is a fundraising tool, but how the money is spent and how systems are reformed are the core of saving safety nets. The discussion on AI taxes should not grant tech companies a moral exemption of ‘we have fulfilled social responsibility through taxation,’ but should compel them to take more direct responsibility in technology design and workforce transition planning.

timeline
    title Key Challenges Timeline from AI Tax Proposal to Potential Implementation
    section Political and Legislative Challenges
        2026-2027 : Conceptual Debate Phase<br>Governments establish task forces for study<br>Tech lobbying and labor groups vie for influence
        2027-2029 : Pilot and Legislation Phase<br>Individual countries or regions (e.g., EU)<br>introduce limited trial taxes
        2030- : International Coordination Phase<br>Organizations like OECD attempt to set<br>minimum standards, but global uniformity is difficult
    section Technical and Definition Challenges
        2026-2028 : Metric Standards Controversy<br>How to define and track<br>"automated labor" becomes the focus
        2028-2030 : Collection Technology Development<br>Potential emergence of blockchain or<br>AI-assisted tax audit systems
    section Economic and Social Challenges
        2026- : Tax Burden Shifting Controversy<br>Companies may pass costs to<br>consumers or supply chain partners
        Ongoing : Distribution Justice Debate<br>Endless debate on whether tax revenue should fund<br>universal welfare or specific workforce transitions

Insights for Taiwan’s Tech Industry: Are We Future Taxation Targets or Rule-Makers?

Direct answer: Taiwan is in a unique and critical position. As an irreplaceable global supplier of AI hardware, especially advanced chips, we are highly vulnerable to being targeted in discussions on ‘compute taxes’ or ‘hardware taxes.’ Simultaneously, Taiwan is an active adopter of AI technology, facing pressure from employment structure transformation. Therefore, Taiwan must not be a passive follower but actively engage in international dialogue, leveraging its critical role in hardware manufacturing to influence rule-making, while preemptively planning domestic strategies to turn potential impacts into opportunities for industrial upgrading and societal strengthening.

First, we must soberly recognize the risks. If ‘AI compute consumption tax’ becomes a mainstream proposal, cloud giants (e.g., AWS, Google Cloud, Microsoft Azure) purchasing large quantities of NVIDIA or AMD chips for model training will face direct cost increases. This may lead to two effects: 1) Rising prices for cloud AI services, suppressing demand; 2) Cloud giants pressuring chip suppliers for lower prices, squeezing profit margins for manufacturers like TSMC. Another possibility is that the tax point is set directly at the ‘AI chip’ factory exit. This would expose Taiwan’s semiconductor industry, which accounts for over 15% of Taiwan’s GDP, directly to policy risks.

According to data from the Industrial Technology Research Institute’s Industrial Economics and Knowledge Center, the global AI chip market size exceeded $200 billion in 2025, with Taiwan holding over 90% market share in advanced processes. Any tax targeting AI hardware will transmit through the supply chain, ultimately affecting Taiwan’s exports and economic growth. Therefore, the government and industry associations (e.g., TSIA) must immediately initiate research to simulate impacts under different taxation scenarios and inject Taiwan’s perspectives and data into global discussions through overseas missions and international think tanks. Our argument could be: Taxing basic compute is like taxing electricity, potentially hindering all beneficial applications including medical research and climate modeling; taxation should more precisely target ‘final substitutional applications.’

Second, Taiwan must also prepare internally. The Directorate-General of Budget, Accounting and Statistics should develop more detailed models to assess AI’s impact on employment and wages across various sectors. Taiwan’s social insurance (labor insurance, health insurance) also heavily relies on wage taxes. We need to consider: if regular positions in manufacturing and services decrease, are there alternative tax bases? For example, is it possible to levy some form of social impact fee on corporate ‘automation equipment investment’? Or, following OpenAI’s suggestion, design stronger incentives to encourage companies to use productivity gains for wage increases or investment in skill retraining for existing employees?

The table below outlines specific scenarios Taiwan may face in the global wave of AI taxes and strategic options: