Ethanol dwarfs data-center land use in verified US figures
When the farmland panic around data centers came up at a Georgia county board meeting this week, the numbers being cited to justify a moratorium were not good. By the standards of a contrarian case making the rounds among AI researchers, they were not even close.
Simon Willison, a respected voice in applied AI, published an analysis Sunday pointing to a body of public data that he said flips the dominant narrative. The core claim: ethanol production in the United States consumes roughly 1.5 percent of the contiguous US landmass — a figure derived from USDA crop reporting by researcher Andy Masley — while all projected US data center footprint by 2028 would occupy about 0.3 percent of prime farmland. The political story about data centers eating America's farmland, in this reading, is backwards on the numbers.
Masley's argument has accumulated something rare in the land-use debate: a clean fact-check. All seven core claims — including the ethanol comparison, the Loudoun County revenue figure, the Northern Virginia concentration statistic, and the 2025 zoning change — cleared independent verification by the newsroom's fact-check layer. The ethanol comparison holds directionally against USDA crop data. The Loudoun County figures and the March 2025 zoning change are public record. The Colorado farmland-sales comparison is a back-of-envelope construction that holds in order of magnitude.
Masley's most striking numbers — the 1.5 percent ethanol land fraction and the 77-times Colorado comparison — are his own calculations from public data, not independently replicated. But Giskard confirmed the directional arithmetic: USDA planted roughly 94.6 million corn acres in 2023, which as a share of the contiguous US works out to a figure in the same ballpark as Masley's 1.5 percent, and the direction of his argument — that ethanol dwarfs data center land use — is correct.
The political urgency around farmland is real and the timing is concrete. The Camden County Board of Commissioners in Georgia votes Tuesday on whether to adopt a nine-month moratorium on data centers, amid speculation that a Florida-based businessman has requested rezoning for a large industrial park to accommodate them. Kingsland's city council takes up its own data center ordinance at a May 11 meeting.
What the farmland advocates are missing, according to Masley's case, is that the land they are trying to protect is already claimed. The federal Renewable Fuel Standard, first mandated in 2005 and expanded in 2007, created the subsidy structure that keeps enormous tracts of US cropland locked in corn production regardless of market signals. The farmland AI infrastructure is supposedly devouring is already claimed, and the claimant has had federal backing since the Bush administration.
The strongest local case for why data centers draw opposition anyway is Loudoun County, Virginia — the largest data center cluster on earth, hosting roughly 13 percent of global data center capacity. The county generates 38 percent of its own general fund revenue from the facilities. After years of expansion, the county's Board of Supervisors adopted new zoning amendments in March 2025 placing controls on further data center development, a regulatory response to resident complaints about noise, power demand, and landscape change.
The political economy explanation for why farmland scarcity feels urgent — despite numbers that say it should not — is structural. Corn ethanol survives because its benefits are geographically distributed across farm states while its costs are diffuse and invisible, baked into federal mandates that do not show up on grocery receipts. Data centers are politically exposed in a way row crops are not because they concentrate their visual and infrastructure footprint in specific counties while their economic gains flow to tech-industry zip codes and county general funds. Residents near a proposed data center see the substation and the screening walls. They do not see the ethanol mandate that keeps 1.5 percent of US land in corn production because the process is invisible and the subsidy check arrives in a different mailbox.
The real constraints on data center expansion are not farmland. They are power grid capacity and water supply. A large data center can draw 50 to 100 megawatts under full load — equivalent to the consumption of a small town — and requires cooling systems that can move millions of gallons of water daily. Counties that want to slow AI buildout have a more effective lever than zoning against farmland conversion: they can slow the interconnection queue for the nearest transmission substation. That fight is harder to win on local optics, which is probably why it is not being fought.
The question worth sitting with is why the farmland scarcity narrative feels urgent when the numbers say it should not. One answer: the people most alarmed by data center expansion tend to live near data centers, and their alarm is locally rational even if nationally incomplete. Another: the AI industry has done a poor job making its land-use case in plain sight, which leaves the political space to whoever organizes fastest. A third answer — less comfortable for either side — is that farmland preservation is a proxy war for deeper hostility to concentrated industrial infrastructure, and the farmland argument is winning because it sounds more principled than "I do not want that building in my viewshed."
What to watch next is whether Loudoun County's 2025 zoning turn becomes a template for other high-concentration jurisdictions and whether counties that successfully restrict data centers on farmland precedent end up with the grid capacity problems they swapped for the visual ones. Masley's numbers say farmland is not the scarcer resource. The political signal from Georgia, Virginia, and a dozen other states suggests the argument has not landed yet — and that power and water will become the next front in the same fight, with fewer sympathetic headlines available for whichever side is outsized.