Note: This is Part 1 in a series entitled “The Empire of AI.” Look for future articles and subscribe below.

Artificial intelligence is marketed as immaterial—digital, weightless, and clean. Yet behind every model and dataset lies a physical, extractive system that rivals the industrial revolutions of the past. Servers hum in desert data centers. Cobalt miners in the Congo extract the metal that powers GPUs. Data annotators in Nairobi label endless streams of content for a few dollars an hour. The illusion of virtuality conceals a staggering global infrastructure.

AI depends on raw materials, energy, and human labor. Rare earths mined for chips are the new oil. Water cools the machines that “think.” Thousands of humans feed them labeled data, performing cognitive assembly-line work. As Karen Hao describes, this is not innovation in isolation—it’s a planetary supply chain. The system is complex, expanding at breakneck speed and will likely impact us all- if it hasn’t already.

In Chile, lithium extraction for batteries drains aquifers used by local farmers. In Kenya and the Philippines, data annotators handle violent or traumatic material with minimal psychological support. The clean interface of AI hides the messiness of its origin—just as the Industrial Age once hid smog behind factory profits. The battle for these minerals is being fought in politcial arenas and could easily spill into a kinetic event.

Data is the lifeblood of this ecosystem. Yet its provenance is murky. Public web scraping and unauthorized content ingestion blur the line between open information and theft. In security terms, unverified data sources equal compromised integrity. If a model is trained on polluted data, its insights are unreliable.

Recent studies show that a single large-model training run can consume as much energy as 100 American homes use in a year—and millions of liters of water for cooling. The carbon cost of “smart” technology rivals that of heavy industry. These are not abstract sustainability concerns; they’re operational and reputational risks for every company that deploys AI. Electricity rates are skyrocketing across the country and proposed data centers are already facing pushback from the public. Next up, nuclear power. Three Mile Island is slated to come back online and numerous other facilities have been proposed.

Corporate security leaders now face a new frontier: protecting cognitive infrastructure. The AI lifecycle—from mining to model deployment—creates new vulnerabilities: physical sabotage, labor exploitation, data corruption, and environmental backlash. Security programs that ignore these will be blindsided when the next supply disruption comes not from a hacker, but a drought or strike.

Optimists argue that efficiency will solve these problems—renewable energy, improved chips, and automation. Yet history teaches otherwise. Efficiency rarely erases exploitation; it just hides it deeper. The goal isn’t guilt—it’s governance.

Security professionals should begin treating AI ecosystems as supply networks. Ask the same questions you’d ask a logistics manager:

- Where are our data and compute resources sourced?

- Who provides the human labeling?

- What are the single points of failure in infrastructure or energy use?

A “Cognitive Supply Chain Map” can expose dependencies, labor risks, and resilience gaps long before they turn into crises.

AI is not magic—it’s machinery. Understanding its physical and human scaffolding is the first step toward responsible innovation. As Karen Hao writes, “AI didn’t escape history—it inherited it.” Those who secure its supply chain will shape the next industrial revolution with conscience as well as code.