Plans for new data centers across the U.S. are coming up against strong opposition from lawmakers and communities, but a new proposal argues that sharing the tax windfall with local households could help ease tensions.
The Bitcoin Policy Institute's "data center dividend" model calls for counties to distribute a fixed portion of the property taxes they already collect from data centers to the surrounding households. This would not require a new tax or impose additional costs on operators, according to the report released Wednesday.
After setting aside between 50% and 75% of the revenue for schools, public safety, infrastructure and other services, the think tank estimates that a single 1-gigawatt AI data center could still fund annual payments of between $4,500 and $8,900 per household.
The proposal comes as the number of local data center moratoria has risen from just six in 2024 to 294 by August 2026, according to the report. It also cited a Gallup survey from earlier this year that found over 70% of Americans would oppose an AI data center built in their area.
The paper used Hut 8's River Bend AI campus in West Feliciana Parish, Louisiana, as a prime example of the effect this proposal could have. Its first phase is expected to generate roughly $90 million in annual payments for a town with about 4,000 households. That's more than triple its current tax collections.
If West Feliciana directed just 25% of that to residents, each household could collect approximately $5,600 per year.
Louisiana lawmakers have already authorized local tax authorities to offer property tax credits funded by data center revenue starting next year, though they removed an option that would allow cash payments.
Similar ideas are also being explored elsewhere. West Virginia plans to use half the revenue from large data centers to reduce or eliminate its personal income tax, while Chesterfield County in Virginia, where Google is building three data center campuses, is considering directing future data center revenue toward vehicle tax relief.
