Imagine a restaurant that charges $5 for a steak dinner. Business is booming. The place is packed every night. Competitors are shutting down because they can't match the price. And the whole time, the restaurant is losing $15 on every plate it serves, kept alive by a billionaire investor who believes the losses will eventually work themselves out if they can just get a high enough market share.

This isn't even a metaphor. It's the current state of the AI industry.

• OpenAI spent $2.25 for every dollar it made in 2024, losing $5 billion after revenue (LessWrong).

• In 2025, the losses grew to around $15 billion, with internal projections showing losses getting even worse throughout 2026.

• Most recently, OpenAI decided to shut down Sora despite a $1 billion infusion of cash from Disney, and it seems clear that losses like this will keep adding up.

Across the market, Google is cross-subsidizing its AI with its advertising empire. Microsoft is doing it through Azure. Amazon through AWS. OpenAI is doing it with investment capital, including a $110 billion funding round in early 2026 at a $730 billion valuation. They're even bribing investors with a 17.5% "guaranteed" return.

This is not a functioning market, it's a heavily subsidized price war being fought with other people's money.

So when a business fires its human staff and replaces them with an AI bot for $100 a month, it isn't tapping into some new efficiency that the free market discovered. It's benefiting from a product being sold at an enormous loss, funded primarily by venture capital. Oh yeah, and also taxpayers (I'll show you how in just a second).

The true cost of AI is staggering, and that's before we even talk about the downstream effects on the economy or the electric grid.

In 2024, US data centers consumed 183 terawatt-hours of electricity in more than 4% of everything the country used. By 2030, that number is projected to grow by 133% (Pew Research Center). This isn't abstract. In the PJM electricity market stretching from Illinois to North Carolina, data centers already drove a $9.3 billion price increase in the capacity market. A Carnegie Mellon study found data centers and crypto mining could push the average US electricity bill up 8% by 2030, and over 25% in the most data-center-heavy markets (Pew Research Center).

So while your company saves on payroll, its remaining employees get hammered with higher electricity bills.

Funny how that works.

The energy cost is one thing, but this also presents a huge new strain on the physical grid itself. In July 2024, a single voltage fluctuation in northern Virginia triggered 60 data centers to simultaneously disconnect, creating a 1,500-megawatt power surplus that nearly caused cascading outages (Belfer Center for Science and International Affairs). This is real, physical infrastructure being pushed harder than it was built to handle. These issues will result in huge expenses that have to be paid for by somebody.

(You guessed it, that somebody is you and your fellow taxpayer.)

The economic logic of replacing a $60,000 employee with a $1,200/year software subscription makes sense on paper, but it's only a no-brainer if you don't consider the expense passed to every consumer, the grid infrastructure costs passed to every taxpayer, the reduced consumer spending from higher unemployment, and the eventual price correction when the companies running the AI can no longer absorb losses measured in the tens of billions.

Is the true cost of running an AI bot exactly equivalent to a human? No, it's likely still far lower. But it's not orders of magnitude lower like we're being taught to believe. It's in a ballpark where it would make sense to cut some employees, but not lay off entire departments.

Just imagine this scenario: A company fires 99 of its 100 developers thinking it will save $150k/year for each one. At first, things go fine. Then they get hit with a new bill: Instead of $100/month, Anthropic is raising its rates to $1k/month per license to meet actual cost and demand. Six months later, it's $2k, then $5k, then $10k. Finally, Anthropic is making up for the cost associated with the increased AI usage and the electric bill that comes with it, and the companies now find themselves with 1/10th the savings they initially imagined, all at the cost of losing human proficiency and true control of their product.

I'm not arguing that the entire AI economy is fake or that companies won't find huge value in replacing jobs with bots. Of course they will.

But the real costs are currently hidden to most people, including executives, and nobody seems to be thinking about it or talking about it.

We need to start that conversation and ask the question: Would this whole process be more gradual, and more healthy for the economy, if AI companies were responsible to charge consumers what it actually costs to run their product? I think so. At least it would make management think twice before making cuts they aren't fully prepared to make just to chase the huge dollar difference in payroll.

---

Full disclosure: I did use AI to help me edit this article. I also used AI to generate the cover image. And gosh darn it I'm going to use AI tonight to help me write code for our website and write a sonnet for my wife.

Disclaimer: The opinions expressed in this article are those of the author and do not necessarily reflect the opinions of Not the Bee or any of its affiliates.