You’ve seen the post: a dry Rio Grande riverbed, a hulking data center, and a caption blaming Mark Zuckerberg’s servers for drinking the river dry. Then comes the number: 75 million gallons. Big. Terrifying. Algorithm-ready.
But scary image plus scary number does not automatically equal scary truth.
The Viral Claim Had a Problem
Snopes checked the Meta Los Lunas, New Mexico claim and rated it false: available records did not show Meta’s data center completely dried up the Rio Grande.
The key detail? Meta’s Los Lunas facility buys municipal water from the Village of Los Lunas, supplied by groundwater wells. That is not the same as direct Rio Grande diversion.
The 75 million gallons number was real, but narrower than the viral version suggested. Snopes traced it to Meta’s reported annual Los Lunas withdrawals in 2023. Convert that and the picture shifts: 75 million gallons per year is about 205,000 gallons per day, or roughly 230 acre-feet annually.
Then another number got tossed into the discourse blender: a service agreement allowing capacity up to 3 million gallons per day under maximum-use conditions. But capacity is not typical use. If that maximum ran every day, it would exceed 1 billion gallons per year. Different universe.
That’s how water claims glitch: withdrawal, consumption, capacity, restoration. Same liquid. Very different accusations.
Data Centers Really Can Consume Water
Here’s the part where nobody gets a free pass. Data centers can use serious water, especially for cooling.
The U.S. Department of Energy explains the basic pathway: servers generate heat, cooling systems move that heat, and cooling towers can shed it through evaporation. Evaporation means consumption. That water is not promptly returned to the same system.
A paper in npj Clean Water adds another layer: direct water use at the data center is only one side of the ledger. Indirect water use comes from the electricity powering the facility. Depending on the grid mix and location, power generation can carry its own water footprint.
So yes, data center water consumption is a real issue. But “real issue” is not the same as “this facility drained that river.” To prove that claim, you need a chain: source, volume, timing, and hydrology.
No chain, no river-drain verdict.
Why Los Lunas Still Deserves Scrutiny
New Mexico is not a random wet suburb arguing over sprinklers. Local scarcity is the plot.
In April 2026, the Bureau of Reclamation cited extreme drought, very low snowpack, early snowmelt, low storage, and a third dry year. Reclamation also reported key Rio Grande and Rio Chama reservoirs below 15 percent, with some at deeply stressed levels. Officials warned stretches of the Rio Grande through Albuquerque could dry because of low runoff, limited stored water, and scarce environmental flows.
That backdrop matters. A false viral causation claim can sit next to a valid public concern.
Groundwater is not magic basement water, either. In some places, pumping aquifers can affect connected streams over time. Snopes reported New Mexico officials said the village’s total water use, including Meta’s supply, is included in regional rights modeling. But those officials did not have facility-level water records, because Meta buys through the village system.
Meta says it supports more than 60 restoration projects, including Rio Grande watershed work it says restores over 172 million gallons yearly. It also says its water-positive target restores 200 percent of consumption in high-stress regions and 100 percent in medium-stress regions.
Good. Now audit it. Where are the projects? When do they restore water? Who measures them? Are they in the same basin, same season, and comparable to actual consumption?
A sustainability page is a lead, not a verdict.
The 4-Number Test
Before sharing the next data center panic post, demand four numbers.
First: source. Did the water come from a river diversion, municipal groundwater, reclaimed wastewater, imported supply, or the electricity chain? For Los Lunas, the reported direct source was municipal groundwater, not direct Rio Grande pumping.
Second: unit. Are we talking withdrawals, consumption, permitted capacity, peak demand, replenishment, or offset claims? Withdrawn water can return. Consumed water usually does not return soon enough to help the same system. Capacity may never be used.
Third: timeframe. Per day and per year are not interchangeable. Los Lunas reportedly withdrew about 67 million gallons in 2024, lower than 2023’s 75 million. The 3-million-gallon daily figure was maximum-use capacity, not proof of typical daily consumption.
Fourth: local scarcity. The same gallon means something different in a wet basin than in a drought-stressed watershed. Ask: same aquifer? Same season? Same communities? Same reservoir system?
The verdict on the viral claim: false as stated. The evidence does not show Meta’s Los Lunas data center drained the Rio Grande.
The verdict on the concern: valid to audit. Data centers bring jobs, tax deals, power demand, water demand, and long-term infrastructure promises. All of that belongs in daylight.
Misinformation loves one huge number. Reality usually arrives as a ledger, a map, a meter, and one annoyingly local context.
So keep the checklist: source, unit, timeframe, scarcity. Four numbers before the outrage. Four numbers before the absolution.
