The true price of AI, Part 1: Who pays for your free lunch?

In the world of technology there's an old saying: if you're not the buyer, you're the product.

The saying began with social media, where the free service was never truly free; we paid with our attention and data, and advertisers paid for us. Today, the same saying applies to artificial intelligence, except that it doesn't go far enough.

Everyone is the product

An overwhelming majority (95%+) of people use ChatGPT, Gemini, Claude, and other similar apps without paying a single cent. In return, they contribute their work. Every question, every answer rated with a thumbs up or thumbs down, and every correction ("no, that's not what I meant") provides valuable information that helps these tools learn and improve. Most major consumer AI products use your conversations to train future models, unless you disable this feature in the settings.

That's the free version. What surprises me is that this also happens with many paid plans. You pay $20 a month, or $200, and in many cases your conversations continue to feed the system automatically. You are both the buyer and the product.

And even paying customers aren't paying what it costs.

The subsidy

Operating these models is incredibly expensive: data centers, specialized chips, electricity on the scale of small cities. The companies that develop them run enormous deficits, and investors cover the difference. They do this intentionally to corner the market: hook everyone, obtain everyone's data, refine the models, and calculate the profits later.

How big is the difference? According to some estimates, for the heaviest users, a $200 monthly plan can cover usage that would cost more than $14,000 at the companies' own pay-as-you-go rates. This is an extreme case, but it illustrates the magnitude of the subsidy. The real price is approximately 70 times the current price.

Investors are patient, but not indefinitely. At some point, they'll want their ROI. And at that point (which is coming sooner than we think), they'll decide that the models are smart enough that any remaining growth can be achieved through profitable, ordinary business use. When that day comes, the free lunch ends.

To be fair, the cost of running AI does continue to fall, and there are free and open-source models available for anyone to download. Prices may not increase overnight across the board. But the best tools -- the ones companies are using to develop their projects -- are a different story.

A friend's story

A friend of mine runs a marketing company. For years, her business was based on content creation: she employed a large team of writers who were experts in accounting, law, trades, and just about every other imaginable profession. These writers earned thousands of dollars a month, consistently. And my friend's overhead cost was astronomical.

A couple of years ago she started experimenting with AI. She was cautious, asked it to write a few articles, watched her sales and waited for complaints.

Nobody complained. So she made a few more. And still no complaints.

Fast-forward to today and she laid off almost all of her staff and kept only a couple of assistants to work alongside the AI. Now she works just one or two days a week for a few hours each day. And she earns the same revenue as before without the overhead. Her AI bill is around $2,000 a month.

Stories like this are becoming increasingly common. Even I'm thinking about it. I value my human team, but in a business, every asset needs to be efficient and profitable in terms of time and money. If an asset costs you either, it's not an asset at all. And every month, it gets harder and harder to justify keeping them. Mathematically speaking.

The savings are not real

Here's the problem. My friend's savings have been based on a price that isn't the real price.

If we multiply her AI bill by 70, it comes to $140,000 a month. The real savings don't look so great after all. But even still, prices don't have to rise until AI costs more than people. They only have to rise until AI costs a little less than people. Whoever controls the price pockets almost the entire difference.

By then, the writers are gone. They've moved on, changed careers, or simply lost their touch. You can't just make a phone call and rehire a team of experts that were let go years ago.

Star Trek predicted it, as it so often does. In "The Ultimate Computer" (1968), Starfleet installs the M-5, a computer designed to operate the Enterprise with a minimum crew of twenty. Captain Kirk spends the episode pondering a question that feels painfully relevant today: what happens to people when machines can do their jobs? In 1968, it was science fiction. In 2026, it's a line item on everybody's income statement.

What we're really building on

Here's the situation: an entire economy is quietly reorganizing itself around a technology whose true cost no one knows. Companies are downsizing, restructuring workflows, and setting prices based on a discount that exists only because investors haven't yet demanded their money back.

That's no reason to avoid AI. I use it daily and I think it's one of the most extraordinary tools humanity has created. But it's a reason to go in with our eyes open.

In the second part, I will analyze what happens when the subsidy ends: what it means for the skills we are losing, for our children, for who has access to the most powerful tools ever created, and why I believe the answer is ultimately political, not technical.

Continue with Part 2: Dark Age or Enlightenment →