The Eye That Eats What It Watches
The same federal lands being mined and drained to power AI are now being monitored by AI. No one is governing the loop.
Somewhere in the Sonoran Desert, a 33-foot tower is watching.
It runs on solar. It sees three miles in every direction. It classifies what it finds whether its a person, vehicle, animal, then pings a Border Patrol agent’s phone without a human ever reviewing the footage. It was built by a defense tech startup, Anduril. The government bought 300 of them. Most sit on or adjacent to federal public land.
That detail matters more than it looks like it does.
Here is the thing nobody is saying plainly: the same public lands being mined and drained to power artificial intelligence are now being monitored and “saved” by AI conservation tools. That loop, extraction feeding the machine that watches what’s being extracted, is not a metaphor. It is a supply chain. And no one in any agency or administration has ever looked at it whole.
U.S. data centers consumed 176 terawatt-hours of electricity in 2023. The Department of Energy (DOE) projects that figure doubling or tripling by 2028. A meaningful share of that energy comes from coal, gas, and uranium pulled from federal lands. The Colorado River Basin is being tapped to cool data centers in Phoenix and Las Vegas and lost 52 cubic kilometers of water storage since 2002. Tech companies have contracted for over 10 gigawatts of new nuclear capacity in roughly one year. Uranium production surged from 2023 to 2024. Most domestic uranium deposits sit on or near federal land.
Meanwhile, U.S. Geological Survey (USGS) deploys acoustic AI to detect endangered birds in national parks. The Forest Service is building AI-assisted fire behavior tools. The Bureau of Reclamation is using machine learning to manage drought on the Colorado. These are real tools doing real work.
But who is measuring whether the conservation return justifies the resource cost? What is the net ledger? Extraction impact on one side, monitoring benefit on the other. Does anyone have the mandate or the data to calculate it?
I don’t know.
The surveillance question is even less resolved.
Border Patrol’s autonomous towers are already operating across public lands with no comprehensive federal law governing what they collect, how long they keep it, or who they share it with. A December 2024 Government Accountability Office (GAO) report found Customs and Border Protection (CBP) failed all six Fair Information Practice Principles for its surveillance systems. Zero for six. A complete failure. Yosemite is running license plate readers at park entrances, retaining vehicle records indefinitely. The Tohono O’odham Nation is surrounded by hidden Automated License Plate Reader (ALPR) cameras stretching 100 miles north of the border.
These tools were built for one purpose. Management. History suggests they rarely stay there. Standing Rock showed what surveillance infrastructure deployed near energy projects does when it gets redirected toward protesters and indigenous communities. The question isn’t whether that risk exists. It clearly does. The question is what governance looks like before the next Standing Rock, not after.
Look up Surveillance at Standing Rock (2016–2017) if you have not heard of it.
The biggest gap isn’t a missing study or a funding shortfall. It’s institutional.
Data center siting decisions. Where to build, how much water to pull, which grids to tap are made through energy permits, water rights transfers, and utility interconnection agreements. None of those processes require a public land use review. None of them ask what the cumulative impact is on the landscape being resourced. The Trump administration’s AI infrastructure executive order explicitly streamlined that review further. The Biden executive order that tried to impose some integration was revoked before implementation.
So we are left with the following: the most consequential land use decisions of this decade are being made by default, through processes designed for individual projects, with no framework capable of seeing them together.
Some questions worth sitting with:
Who decides how AI infrastructure intersects with public land policy, and is anyone actually deciding, or is it just happening?
When AI is simultaneously degrading and monitoring the same watershed, what does “conservation” even mean in that context?
If surveillance tools deployed on public lands can be redirected against the people who use them like hunters, hikers, activists, tribal members, what accountability exists before that happens, rather than after?
Does the conservation benefit of a wildlife monitoring AI justify the water and energy cost of running it? Has anyone tried to measure that?
And the one underneath all of it: 348 million Americans own these lands. When did they get a vote on any of this?
Back to the tower in the Sonoran Desert.
It watches a landscape sitting south of uranium deposits needed for the nuclear plants powering AI data centers. On the ridge above it, there may be a camera running a machine learning model trying to detect a Sonoran pronghorn. Both tools draw from the same stressed infrastructure. Neither was placed there through any process that considered the other.
Nobody decided the eye that watches the land should eat the land to keep watching.
It just happened. And we haven’t figured out what to do about it.
Thank you for reading! I highlight threats to public lands and the energy industry’s impact. I believe clean energy is the future, and ALL energy projects should prioritize private land first to keep wild places wild. When energy extraction is needed on public lands all projects must restore the land after extraction. Public lands are unique and once lost, they’re gone forever.
Sources
Energy / Federal Lands Extraction
Lawrence Berkeley National Laboratory: 2024 United States Data Center Energy Usage Report (LBNL-2001637), December 2024. Direct source for the 176 TWh figure and 2028 projections.
International Energy Agency: Energy and AI, April 2025. Source for AI’s share of data center power (5–15% today, 35–50% by 2030) and natural gas supplying 40%+ of U.S. data center electricity.
Bureau of Land Management: Oil and Gas: About. Source for federal lands supplying ~41% of U.S. coal, ~15% of oil, ~9% of gas.
U.S. Energy Information Administration: Domestic Uranium Production Report, Annual. Source for uranium production surge (50,000 lbs in 2023 to 677,000 lbs in 2024).
U.S. Department of the Interior / USGS: Final 2025 List of Critical Minerals. Source for uranium’s addition to the list and the expanded 60-mineral inventory.
Water
Geophysical Research Letters (Abdelmohsen et al., 2025): Declining Freshwater Availability in the Colorado River Basin Threatens Sustainability of Its Critical Groundwater Supplies. Source for 52 km³ water storage loss since 2002 and the explicit data center citation.
LBNL 2024 Report (same as above): Source for 17 billion gallons direct consumption and 211 billion gallons indirect.
OPB: As Google’s Water Demands Grow, The Dalles Aims to Pull More from Mount Hood Forest, January 2026. Source for Mount Hood National Forest transfer lobbying and Google’s share of The Dalles water supply.
Reporters Committee for Freedom of the Press: Oregon’s City of The Dalles Agrees to Reveal Google’s Local Water Usage. Source for the trade secret fight and Google paying $153,000 in legal fees.
Ceres: Water Impacts from Data Centers, September 2025. Source for 32% of U.S. data centers in high water stress areas and Phoenix 900% consumption projection.
AI Conservation Tools
USGS: Artificial Intelligence in the USGS Ecosystems Mission Area: Fish and Wildlife. Source for SeeOtter, BirdNET deployment, waterfowl counting, Klamath fish ID, tegu detection.
USFWS: 2024 Range-wide Indiana Bat and Northern Long-eared Bat Survey Guidelines. Source for formal incorporation of acoustic AI.
Upstream Tech: Integrating and Evaluating AI Forecasts for Bureau of Reclamation. Source for HydroForecast deployment on western basins.
MDPI / AI journal: AI for Wildfire Management: From Prediction to Detection, Simulation, and Impact Analysis, 2025. Source for the “operational use remains limited” finding.
Surveillance
GAO: GAO-25-107302, December 2024. Source for CBP failing all six Fair Information Practice Principles.
Electronic Frontier Foundation: Customs & Border Protection Fails Baseline Privacy Requirements for Surveillance Technology, December 2024. Corroborates GAO findings, maps tower deployments.
Arizona Mirror: Hidden in Plain Sight: Surveillance at the Arizona Border, February 2026. Source for the ALPR network on Tohono O’odham land.
NPS: License Plate Reader Pilot, Yosemite National Park. Direct agency source for the Yosemite LPR program details.
The Intercept / Grist: Leaked Documents Detail Standing Rock Surveillance and How TigerSwan Pitched Its Pipeline Playbook After Standing Rock. Source for Standing Rock surveillance precedent and TigerSwan operations.
In These Times: How Border Patrol Occupied the Tohono O’odham Nation. Source for Ned Norris Jr. congressional testimony.
Governance / Policy
White & Case: Trump Administration Issues Executive Order to Streamline Data Center Development. Analysis of EO 14318 and NEPA streamlining provisions.
Morgan Lewis: President Biden’s Executive Order: Accelerating AI Infrastructure Development on Federal Lands. Analysis of EO 14141 and its revocation.
Davis Graham: The Trump Administration’s Progress to Site Data Centers on Federal Lands: Initial Steps but Work Remains. Source for DOE’s four selected federal sites and BLM’s lack of identified parcels.