Senator Wyden’s AI Tax Could Push Data Center Tax Rates Above 100 Percent

Adam N. Michel

Senate Finance Committee Ranking Member Ron Wyden (D‑OR) recently proposed a new federal tax on data centers. The proposal would tax gross data center revenue “at a low single-digit rate.” Because it taxes revenue rather than net profits, the proposal risks imposing effective tax rates exceeding 100 percent.

Such a tax could end new data center development and risk stripping the US of its status as the global leader in AI and other cloud-based technologies. 

The fundamental problem with taxing revenue rather than profits is that the tax ignores whether a firm can pay the tax. Revenue tells policymakers nothing about profitability. Grocery stores, for example, generate hundreds of billions of dollars in sales each year while typically earning profit margins of only 1 to 3 percent.

Because Wyden’s AI tax proposal does not scale with profitability, it would impose the highest effective tax rates on the least profitable data centers. 

To illustrate this problem, I added Wyden’s proposed tax to illustrative data center tax burdens estimated by the Tax Foundation’s Jared Walczak. He modeled a stand-alone $1 billion data center in 12 jurisdictions. The model firm has $220 million in revenue and earns a 15 percent net pretax margin. For an established operation, Walczak estimates total federal, state, and local taxes ranging from 24 percent of profits in Charlotte, North Carolina, to 80 percent in Santa Clara County, California. Most of that burden is not corporate income tax but real property taxes, tangible personal property taxes, and taxes on electricity.

Under Walczak’s specifications, an additional 3 percent federal tax on data center revenue would increase taxes in each jurisdiction by $6.6 million. In Santa Clara County, that low single-digit tax would push the total effective tax rate to 100 percent. Every dollar of pretax profit goes to the government. Figure 1 shows estimated effective tax rates as a percentage of profits with and without the Wyden proposal. 

The distortion becomes even more pronounced for lower-margin data centers. If we assume profit margins are only 10 percent, four of the 12 jurisdictions face tax rates above 100 percent, and eight would face rates above 90 percent. Table 1 shows the minimum profit margin a firm must surpass to keep any after-tax profit under the Wyden AI tax at different tax rates. 

The burden of a gross tax is inversely proportional to profitability: the lower the margin, the higher the effective tax rate. Wyden’s own justification is that the tax will fall most heavily on the major “hyperscalers.” But if the largest firms in the industry are also the most profitable, a revenue tax does the opposite. It burdens lower-margin operators more heavily. 

The case for gross taxation is often that profit is a lagging indicator of economic success. Something real is happening that the income tax is not capturing, and taxing the revenue is a good enough proxy. But that argument overlooks why tax systems generally tax profits rather than receipts. Profit is what’s left over after the cost of production. Taxing that residual is less destructive than taxing the underlying activity itself. Taxes on net profits still distort economic activity, but they are significantly less damaging than gross taxes. Administrative convenience is neither a coherent nor a sufficient economic rationale for taxing revenue. 

The income tax is already built to capture the gains from AI. If investments in data centers pay off, the returns will show up as corporate profits and higher firm values. Those profits will be taxed by state and federal corporate income taxes and again as capital gains when investors sell.

A new revenue tax, instead, taxes investment before policymakers know whether the investment will succeed. At the potentially high effective tax rates contemplated by Senator Wyden, the tax could prevent those investments—and the economic gains they might have produced—from happening at all.