From 2000 to 2024, US farmers voluntarily sold an area of land equivalent to the entire state of Colorado, which is 77 times the projected land use for datacenters in 2028—the claim that the AI industry is "grabbing land" may be severely exaggerated.
What this is
Andy Masley presented an overlooked set of numbers this week: over the past 24 years, US farmers have sold a total area of farmland roughly the size of Colorado, an amount 77 times the estimated footprint of all US datacenters by 2028. More crucially, food production on the remaining land didn't fall but instead rose to record highs; the US food supply has not been affected at all.
His target is a popular recent narrative surrounding datacenter construction: Hyperscalers (giants like AWS, Microsoft, and Google that build massive-scale datacenters) are grabbing land everywhere to build datacenters, encroaching on farmland and threatening food security. Masley's core judgment is this: when a farmer in Loudoun County, Virginia, sells a few acres of mediocre hayfield to a cloud provider at ten times its agricultural valuation, and public opinion claims "farmland is running out," this is more an intuitive panic than a data-backed conclusion.
Industry view
We note that Masley's argument holds up at the macro level. The total volume of US farmland is massive, datacenters account for an extremely small fraction, and agricultural efficiency continues to improve, allowing the same amount of land to feed more people. Amplifying a localized phenomenon into a systemic crisis is indeed a common narrative bias.
However, the opposing voices are equally worth hearing. First, total volume does not equal structure: datacenter site selection is concentrated in specific regions (like Loudoun County and Idaho), and the land prices, water usage, and power grid pressures on local communities are real. Farmers may profit from selling land, but residents who stay may bear the infrastructure costs. Second, water consumption is trickier than land—the water used for datacenter cooling is already a focal point of controversy in some arid regions, meaning land area is not the only dimension. Third, the premise of "farmers voluntarily selling land" can itself be questioned: when giants offer ten times the price, how much room for choice does this "voluntariness" really have under capital pressure? This is a question of political economy.
Impact on regular people
For enterprise IT: Datacenter land costs are unlikely to rise significantly due to a "farmland defense war," but approval delays and community resistance in specific regions may increase project uncertainty. Site selection logic needs to factor in community relations.
For individual careers: This debate does not directly impact employment, but the signal it reflects is worth noting—the regional clustering effect of datacenters is reshaping the employment structure of some tier-2 and tier-3 cities, and the distribution of infrastructure and operations jobs will shift accordingly.
For the consumer market: In the short term, cloud service prices will not rise due to land costs; in the medium to long term, if water and grid constraints force datacenters to migrate to remote areas, latency and bandwidth costs could indirectly be passed down to end-user services.