How America’s tax code subsidises artificial intelligence

The United States has a tax-base problem hiding in plain sight. Its fiscal system is built around labour income and artificial intelligence is beginning to reduce the economy’s dependence on labour. Paul Krake shows how the tax code makes human employment more expensive than the software that may replace it. In 2025, the US economy generated roughly $13 trillion in wages and salaries, compared with about $4.1 trillion in corporate profits. Labour remains the larger pool of income, but the direction of travel should concern policymakers. Profits are growing faster than wages, corporate margins are expanding and AI allows companies to increase output without adding workers at the same rate. Yet the government taxes the slower-growing income pool heavily, while taxing rapidly expanding corporate profits leniently and with greater scope for deductions, deferral and avoidance. The federal revenue mix exposes that dependence. In fiscal 2025, individual income taxes generated 51 per cent of total federal revenue, and payroll taxes accounted for another 33 per cent, meaning taxes on individuals and employment accounted for 84 per cent of federal revenue. Corporate income taxes contributed less than 9 per cent, while customs duties and tariffs supplied under 4 per cent. The figures describe a fiscal system overwhelmingly dependent on people remaining employed and earning taxable income. Federal corporate income-tax receipts were about $452 billion in fiscal 2025, compared with approximately $4.4 trillion from individual income taxes and social-insurance contributions. As AI shifts income from wages to profits, the government risks becoming increasingly reliant on the part of the economy growing more slowly. The tax bias against human labour Payroll taxes also distort the decision between hiring a person and deploying software. For most employees below the Social Security wage ceiling, the combined Social Security and Medicare burden is 15.3 per cent of wages, divided between the employer and employee. The employer-side charge alone adds 7.65 per cent to the wage bill before benefits and other employment costs. Consider a company deciding how to perform a role that costs $100,000 in wages. Before paying for healthcare, training or administration, the employer owes an additional $7,650 in payroll taxes. The employee also has $7,650 withheld to help fund Social Security and Medicare. Now assume the company can buy an AI system for $100,000 that performs enough of the work to make the position unnecessary. The company pays the invoice but avoids the employer payroll tax attached to the job. That is not a neutral tax system. Human cognition arrives with a payroll-tax obligation, while artificial cognition does not. Companies already have operational reasons to adopt AI, including speed, consistency, scalability and lower marginal cost. The tax code adds another by making labour more expensive before any comparison of productivity is made. The defence of the current system is that AI is not tax-free. Vendors, employees, shareholders and data centres all generate other taxes. That is true, but it does not address the decision facing the company substituting software for a person. At the point of replacement, the company eliminates both the worker and the employer contribution associated with that position. American social insurance remains financed through employment. Social Security and Medicare depend on payrolls continuing to expand alongside the economy. But AI creates the possibility that output and profits rise while employment grows much more slowly. The result could be a healthy-looking economy with a steadily weakening capacity to finance the commitments attached to work. If software permanently absorbs a task, the associated position may not return just because demand improves. This separates AI displacement from a conventional recession, in which employment usually recovers as economic activity rebounds. A company that has redesigned a workflow around software has little reason to recreate the previous cost structure. There are no cyclical solutions The fiscal consequences follow directly. Companies deploy AI to reduce costs and protect margins. Labour intensity falls while profits rise, but the tax system captures the increase in profits less effectively than it captured the wages they replaced. Deficits can therefore widen even when aggregate growth remains respectable, while spending pressures increase as displaced workers require support. Monetary policy cannot solve this problem. Lower interest rates can support demand, asset prices and financial stability, but they cannot persuade a company to rehire workers whose functions have been absorbed into software. The Federal Reserve was designed to manage cyclical weakness in a labour-driven economy. It has no instrument capable of correcting a structural shift in the source of taxable income. Policy should begin with neutrality between human and artificial cognition. This does not require a crude “robot tax.” It requires recognition that a tax system built for a labour-heavy economy cannot finance a capital-heavy one by continuing to place most social obligations on employment. The cleanest reform would reduce the tax penalty on labour while increasing the effective burden on corporate profits and distributed capital income. Personal income-tax rates and employer payroll charges could fall if corporate profits and dividends were taxed more consistently. Social insurance also needs a broader funding base so that its solvency does not depend on human labour retaining its historic share of production. Capital gains require separate treatment because realisation is discretionary and not directly tied to current-year production. The central employment distortion lies between the automatic taxation of wages and the lighter effective taxation of corporate profits. That is where reform should begin. The objective is not to punish successful companies or discourage productivity. AI will strengthen many American firms and reinforce the country’s technological advantages. The objective is to prevent the fiscal system from favouring the input-replacing workers over the workers themselves. The political consequences of failing to adjust will not arrive as a technical debate over tax neutrality. They will emerge as anger toward the firms and investors capturing the gains from automation. A society can tolerate high profits and rapid technological change when the gains are broadly taxed and recycled through legitimate institutions. It will struggle to sustain a system in which the burden of financing the state remains concentrated on people whose economic position is being weakened by the technology. A labour-based tax system cannot indefinitely finance an increasingly capital-heavy economy. The United States does not need to stop AI adoption, which would be impossible and self-defeating. It needs to redesign the tax base so that the financing of government follows where income is increasingly being generated. This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment. Image credit: Trive Studios ID provided by Shutterstock.