The hidden costs of our digital revolution

I ask ChatGPT to help me draft an email, summarize a report, outline a blog post, more times a week than I’d like to admit. It feels clean. Efficient. Almost magical.

What I don’t see is the trail my question leaves behind. A bottle of water, vanishing somewhere in a data center. A worker in Kenya, earning less than $2 an hour, spending his day labeling disturbing content so my AI assistant would know what not to say.

I’ve been sold a story about AI that’s only half true. Yes, it’s revolutionary. Yes, it’s changing everything. But the other half, the one about exploitation and environmental cost, stays carefully tucked behind corporate PR and the promise of effortless productivity.

Here’s what I found when I went looking.

The water problem

More than 160 new AI data centers have gone up across the US in the last three years. That’s a 70% increase. And here’s the part that stopped me: most of them sit in places already struggling with water shortages.

We’re building the AI revolution in the exact places that can least afford to share their water.

Nobody agrees on how much water a single AI query costs. The estimates range from a fraction of a teaspoon to a full 16-ounce bottle, depending on whether you count only the cooling towers or the power plants upstream too. The companies publish the low number. Independent researchers publish the high one. That gap is the story: we’re arguing about the cost of something nobody is required to measure. What’s less disputed is the scale — ChatGPT alone uses as much electricity as 180,000 American homes. Microsoft’s water use jumped 34% in one year: 1.7 billion gallons. Google guzzled 5.56 billion, enough to fill 8,400 Olympic pools.

In The Dalles, Oregon, a small town on the Columbia River, Google’s data centers pull more than a quarter of the city’s entire water supply. Farmers, Native American tribes, and environmentalists have been fighting for years just to find out how much water is disappearing. The city sued to keep that number secret.

This isn’t only happening here. People have protested in Chile and Uruguay over data centers that would tap the same reservoirs they drink from. In Spain, marchers carried a sign that said it plainly: “Your cloud is drying my river.”

Tech companies knew this was coming. Microsoft promised to be water-positive by 2030. Instead, its emissions are up 30% since 2020.

The workers behind the curtain

The other cost is human, and it’s just as hidden.

Data labelers sit in offices across Kenya, Venezuela, the Philippines, and India, teaching AI to recognize faces, objects, emotions. “This is a TV. This is a microwave. If it looks like this, it’s white. If it looks like this, it’s Black.”

Kenyan workers training ChatGPT were paid between $1.32 and $2 an hour. American workers doing the same task made $10 to $25. OpenAI paid the outsourcing company $12.50 an hour per worker. The workers saw $2. Nobody’s explained where the other $10.50 went.

The conditions are brutal, too: overcrowded rooms, no job security, no healthcare, payments that vanish without explanation. These workers spend their shifts processing the worst the internet has to offer: violence, hate speech, abuse, while being told to stay completely neutral. Last year, 97 of them wrote an open letter to the president. Their words: “Our working conditions amount to modern day slavery.”

Many of them didn’t even know who they were really working for. They thought they were labeling data for a platform called Remotasks. They were actually working for a subsidiary of Scale AI, a company that contracts for the biggest names in tech.

Why we keep choosing this path

Here’s what I keep coming back to: none of this is inevitable. It’s a choice, and it’s one the market keeps making because the incentives point that way.

Cloud-based AI is a subscription. Every query is a billable moment. The bigger and thirstier the model, the more expensive the infrastructure, and the higher the wall against anyone trying to compete. There are no regulations requiring these companies to disclose their water or energy use, so there’s no real pressure to change.

It’s not that the people running these companies are evil. They’re responding to incentives that reward scale over sustainability, and nobody’s pushing back yet.

But there’s a live counterexample worth paying attention to. Chinese AI companies, working under chip restrictions, have been forced to build for efficiency instead of throwing more compute at every problem. DeepSeek-R1 runs at a fraction of the cost of comparable Western models, and its larger model activates only a small slice of its own parameters per query instead of firing up the whole thing. Other companies are pushing even further, training frontier models on what Andrej Karpathy called “making it look easy” with a “joke of a budget.”

The efficiency breakthrough is really just AI catching up to a vision that was always possible. Apple’s 2020 chip could already do 11 trillion AI operations a second. The hardware to run AI locally, on your own phone, has existed for years. We just haven’t built the business model around it, because a model that runs on your phone doesn’t generate a monthly cloud bill.

What intention looks like

I don’t think the answer is refusing AI. I use it. I’m not interested in pretending otherwise.

But I am interested in what changes when each question I ask doesn’t carry an invisible cost: a bottle of water, a few cents flowing to someone paid $2 an hour on the other side of the world. When that cost disappears from view, it gets easier to ask whether I’m using the tool for something that actually matters, instead of asking it to validate an opinion I already had, or fill a silence I didn’t want to sit in.

The technology to do this differently already exists. The demand for it already exists. What’s missing is the collective decision to ask for it.

The water disappearing with every AI query, the workers earning $2 an hour to make our assistants smarter: these were never inevitable costs of progress. They’re choices. And the more of us who know that, the harder those choices get to keep making in the dark.