The AI Class War: Who Profits, Who Pays, and Who Gets Left Behind?

Someone is paying for your free chatbot. It's probably not who you think.

Big Tech is on track to spend over $700 billion on AI infrastructure in 2026. Meanwhile, workers in Nairobi label its training data for around $2 an hour, and households in 13 US states absorb billions in higher power bills. A clear-eyed look at who wins, who pays, and what a fairer deal could look like.

Artificial intelligence is often described as "weightless": software in the cloud. But follow the money and the materials, and a very physical economy appears, with clear winners and clear losers. Let's map it.

Who profits

The four largest US tech companies (Amazon, Alphabet, Microsoft and Meta) are expected to spend roughly $700 billion or more on capital expenditure in 2026, most of it on AI data centers and chips. That's about 77% more than the roughly $410 billion they spent in 2025, making it the largest corporate investment wave in history. Their businesses are growing too: Statista reported Microsoft's AI business up 123% year on year in mid-2026, and Amazon Web Services growing at its fastest pace in 18 quarters.

The gains flow mainly to shareholders, executives, chipmakers and a relatively small number of highly paid engineers. Economist Thomas Piketty has documented how, when returns on capital outpace wider economic growth, wealth concentrates. AI looks set to accelerate that dynamic.

Who pays: the invisible workforce

AI models don't learn on their own. Humans label images, rate chatbot answers and filter violent or abusive content so that the rest of us never see it. Researchers Mary Gray and Siddharth Suri call this "ghost work": essential labour designed to be invisible.

The pay gap is stark:

• A TIME investigation found Kenyan workers labelling toxic content for ChatGPT took home about $1.32 to $2 an hour, while the outsourcing firm was reportedly billing around $12.50.

• A 2026 Kenyan government draft policy puts local AI data workers' pay at $1.46 to $3.74 an hour, against $21 to $27 for US moderators doing comparable work.

• In the Philippines, more than 10,000 people worked on Scale AI's Remotasks platform, where the Washington Post reported delayed, reduced, or cancelled payments.

Meanwhile, industry analysts at SemiAnalysis estimate each frontier AI lab spends on the order of $1 billion a year on human training data. The money exists; it just doesn't reach the people doing the most gruelling work. Many describe lasting trauma from reviewing graphic material.

There's some progress: Kenya is drafting minimum pay standards and mental health protections for AI data workers, and a Data Labelers Association has formed to organise workers.

Who pays: communities and ratepayers

Electricity. The International Energy Agency estimates a typical hyperscale data center uses as much electricity as 100,000 households. Meta's planned Hyperion campus in Louisiana will need at least 5 gigawatts, about three times the electricity use of New Orleans. In the PJM grid region spanning 13 states and Washington, D.C., the market's independent monitor attributed about $9.3 billion in one year of higher capacity costs to data center demand, recovered through everyone's electricity bills.

Water. Many facilities use large volumes of water for cooling, often in regions already under drought stress, such as the US Southwest.

Tax breaks. Virginia's data center tax exemptions are now valued at about $1.6 billion a year. That's revenue public services don't receive.

E-waste. AI runs on rapidly upgraded hardware. The UN's Global E-waste Monitor found the world generated 62 million tonnes of e-waste in 2022, with less than a quarter formally recycled. A 2024 study in Nature Computational Science projected that generative AI alone could add 1.2 to 5 million tonnes by 2030. Much of the world's e-waste ends up in lower-income countries, where informal recyclers, often including women and children, handle toxic materials.

Who gets left behind

Workers in routine white-collar roles are the most exposed. In NBC's September 2026 poll, 70% of Americans said AI is reducing job opportunities, and among 18- to 29-year-olds it was 73%. Entry-level jobs in customer service, admin, writing and basic coding are the rungs young people traditionally climb, and they're the ones being automated first. Women are heavily represented in many clerical and administrative roles, which means the impact isn't gender-neutral.

Is it really a "class war"?

That's a strong phrase, so let's test it honestly. AI also creates value widely: free tools, medical breakthroughs, productivity gains, and new jobs in areas like data centers and AI safety. Not every cost is a scandal, and some fears are overstated.

But economists Daron Acemoglu and Simon Johnson, in Power and Progress, show that whether a technology shares its benefits broadly has never been automatic. It depends on power: labour laws, taxes, regulation, and who gets a seat at the table. In the early industrial revolution, productivity soared for decades while workers' wages barely moved, until unions and reforms changed the balance.

So "class war" may overstate intent, but it describes the distribution accurately: the gains are concentrated, while many of the costs are spread across people with the least say.

What a fairer deal looks like

• Large data centers paying the full cost of the grid upgrades they require, rather than passing them to households.

• Enforceable minimum pay and mental health protections for data workers worldwide.

• Transparency on water and energy use, and conditions on tax breaks.

• Retraining programmes funded by the companies benefiting most.

If you knew the person who made your AI tool safe earned $2 an hour, would you use it differently, or just try not to think about it?

Sources: CNBC, Futurum and Statista (2026 capex); TIME (2023); Outsource Accelerator on Kenya's draft AI policy (July 2026); HeroHunt citing SemiAnalysis and the Washington Post; Consumer Reports (March 2026); IEEFA (PJM); Cardinal News via Introl (Virginia tax exemption); UN Global E-waste Monitor 2024; Wang et al., Nature Computational Science (2024); NBC News (Sept 2026).

Follow Femonomic for data-driven stories about who really pays for progress.