The Cost of AI: Who Really Pays?

What is the real cost?
Who pays it?
Is it worth it?
Who gets to decide?

A cursor blinks in an empty field. The interface is clean, minimal, visually elegant. A question is typed and within seconds the LLM answers, sounding fluent and confident, assembled from what feels like the sum of human knowledge. It is seamless, impressive, and it hides almost everything that made it possible. AI generated content hides its footprint. The cost does not disappear. It is simply moved somewhere the LLM end user cannot see.

Somewhere, a data center the size of several city blocks is consuming water around the clock to prevent servers from overheating. Larger data centers require up to 5 million gallons every day, the equivalent of a city of 50,000 people. Training a single AI model like GPT-3 consumed enough water to supply the daily needs of 5,000 to 7,000 people. Newer models like Google's Gemini consumed over a million litres per training cycle. Without mitigation, global data center water consumption could reach 28 billion litres per day by 2050.

Meanwhile, somewhere in Nairobi or Manila or Kampala, workers are paid as little as two dollars an hour to sit at laptops provided by the AI companies to label each piece of disturbing content so the model knows what not to generate. Working in shifts reviewing hundreds of horrific images or text descriptions per day, many develop symptoms consistent with PTSD and mental health support available to the labelers is minimal to nonexistent. The LLM responds, and nowhere in that transaction is this human cost visible to the end user making the request. Investigations by Time magazine, CBS 60 Minutes, and SOMO have documented this workforce across Kenya, the Philippines, India, and Uganda, where workers describe the conditions as modern day slavery. In April 2026 a Kenyan data labeler who spent years reviewing explicit content became secretary general of the newly formed Data Labelers Association, drawing a direct line between colonial exploitation and what LLM technology companies are doing today.

Two Systems of Value

Once again humanity is faced with two fundamentally different systems of value. The intrinsic stewardship system and the extractive ownership system.

The intrinsic one assigns value through a symbiotic relationship to all other things. Coexistence is valued. Everything in nature is recognized as an integral part of the planet systems that enable life. A story has worth because it is a lived experience that shares knowledge. An artwork has worth because the process in which it was created gives it meaning.

The extractive one assigns value only to what something yields. Knowledge, nature, creative work, and human labor become resources to be acquired and converted into profit, valued only through utility, control, and exchange. Within this framework what cannot be monetized and owned is not counted or recognized as valuable.

The Pattern Is Not New

Extractive systems do not present themselves as extractive. They present themselves as progressive, beneficial, and inevitable. Colonization claimed to value human development. It was presented as civilization and framed conquest and conversion as the natural order, all the while stealing land and forcefully dismantling the cultures and communities of the people they colonized. Medieval England operated on a collectively managed commons beginning in the 12th century. Then those lands were enclosed, seized and fenced for private profit, and the languages, land practices, and belief systems of Celtic, Gaelic, and indigenous British cultures were not recognized as knowledge. The pattern continued in North America with Manifest Destiny, where Indigenous nations were displaced, their cultures silenced, and their children taken to institutions designed to eliminate indigenous knowledge. The Industrial Revolution was also framed as progress, while polluting the environment and dismantling pastoral life.

We began experiencing Shifting Baseline Syndrome, accepting a diminished environment as normal, without knowing what existed before. The meadows, rivers, birdsong, and biodiversity that previous generations knew as ordinary become invisible to those who never experienced them. The baseline quietly shifts. The intrinsic value of what was lost was never counted. It still isn't.

The Familiar Ask

The extractive system has always needed a human face to make the ask. It's the colonial invader, the slave trader, the factory boss, the corporation representative, the big tech recruiter. Compliance is always framed as opportunity and it is framed as inescapable. It comes with the expectation that those of us with knowledge, with craft and talent, with time to do the work, or proximity to something real, will hand it over to those with capital. There appears to be no other choice.

Extractive tools arrive with the tools of the moment: social media platforms designed to connect us to family and friends, while quietly collecting our data, mapping our relationships, and converting our attention and behavior into profit.

Big tech companies make the same promises with AI, presenting it not only as faster, cheaper, more efficient, but as democratizing creativity, empowering individuals, and building the future, while systematically extracting from creators, replicating their work without consent, and replacing authors' and artists’ labor with automated systems trained on what was taken from them at the cost of the environment and the lives of data labelers.

Yet, 94% of chief executives say they will continue investing in AI at current or higher levels even if the investments do not pay off, while 95% of organizations reported zero return on investment in generative AI projects in 2025. The stock is still going up. That is all the extractive system requires, profit.

The Choice

Choosing not to hand over what you know, not to train the model, not to take the paid task, not to document your own replacement, is not passivity. It is agency. People in need may not have that luxury, but those that can, must. Control over knowledge can challenge more visible forms of power.

Artists, writers, musicians, journalists, photographers, record labels, and film studios are withholding through the courts, filing over 70 copyright infringement lawsuits against AI companies and counting. The data workers organizing in Kenya are withholding compliance. The creatives who choose to create without relying on AI are making the same choice.

The most powerful response available is often the simplest. Withhold. Stop using the extractive tools.