Gates wants to change AI's cost advantage
Bill Gates wants governments to tax AI tokens and robots, putting a possible future cost on the same AI use that Microsoft and Amazon are spending heavily to capture.
In an essay on AI policy choices, Gates argued that employers pay payroll taxes when they hire people, while robots can usually be written off immediately as business expenses. He wants a tax structure that slows labor replacement and helps fund retraining and a stronger safety net.
Microsoft and Amazon need AI usage to keep rising
Microsoft has direct exposure because it is selling AI through Azure and Microsoft 365 Copilot. Azure annual revenue passed $100 billion, and Copilot had more than 30 million paid seats at the end of fiscal 2026.
Microsoft also added $115.9 billion of property and equipment during fiscal 2026, up from $64.6 billion a year earlier. That spending builds capacity before the company knows how much customers will ultimately pay for AI work.
Amazon faces the same basic equation through AWS. AWS generated $42.2 billion of second-quarter sales and $16.6 billion of operating income. Amazon said its AWS AI business exceeded a $25 billion annual revenue run rate.
AI infrastructure is already reshaping Amazon's cash flow. Property and equipment purchases rose by $66.1 billion from the prior year, primarily due to AI investment, and trailing-12-month free cash flow was a $7.6 billion outflow.
Neither company's total revenue is the key exposure. The issue is whether customers keep increasing paid AI use after any new tax cost. That demand supports cloud revenue, data-center utilization, and the eventual return on today's infrastructure spending.
A token tax would create a pricing decision
Gates did not propose a finished tax plan. A U.S. federal proposal defining taxable AI tokens or robots, a rate, exemptions, and the legal payer would be the key next test. He also wrote that domestic and international institutions to manage AI will take years to build.
That makes this a policy risk with a longer timeline than the next earnings cycle, not evidence of an imminent revenue hit for Microsoft or Amazon.
The mechanism would matter if lawmakers reached broad cloud AI processing. Customers could pay the tax directly, providers could absorb some of it through lower prices, or the burden could be shared. Higher customer costs could slow usage growth. Provider absorption could reduce the margin on a service built to justify large infrastructure outlays.
The strongest countercase is Gates' own limit on the idea. He called for targeted taxes that do not slow beneficial uses such as medicine and education. A rule aimed mainly at physical robots or a narrow group of high-risk uses would leave mainstream cloud AI demand largely intact.
