The Charts the AI Industry Doesn’t Want You to See

This is what the AI story looks like

May 29, 2026

Over the years, I’ve shared many charts that illustrate how AI and the industry are evolving. But I’ve never put them together. This is the bigger picture that standalone news articles miss. When you see them at once, things start to make sense.

I’ve compiled 11 charts that I consider extremely important: the ones that reveal the ugly side of the story. A chart tells you what the industry wants you to see. Enough charts, and you start to see through the industry.

The format is: title, chart, one-liner explanation with source. In 5 minutes of reading time, you will understand more about where the AI industry is heading than from hundreds of pages of corporate reports.

Here’s the outline:

TEN COMPANIES ARE CARRYING THE ENTIRE ECONOMY
THE CAPABILITIES OF AI HAVE BEEN EXAGGERATED
MOST TOKENS GENERATED BY AI ARE WASTED
AI RELIABILITY IS BARELY IMPROVING
THE AI INDUSTRY RUNS ON CIRCULAR DEALS
PEOPLE ARE NOT USING AI TO ITS FULL POTENTIAL
AMERICANS DON’T WANT DATACENTERS NEARBY
AI USERS BELIEVE AI EVEN WHEN IT’S WRONG
WORKERS DON’T THINK AI IS SAVING THEM TIME
AI IS KILLING THE JOBS OF JUNIOR DEVELOPERS
THE IMPOSSIBLE MATH OF THE AI BOOM

IMPORTANT REMINDER

I’m running a 50% discount offer on annual subs that ends on June 5th. Get yours now and enjoy the discount forever.

If you read the free edition and have been thinking about subscribing, this is the best time. I won’t do another offer anytime soon.

GET YOUR 50% DISCOUNT

1. TEN COMPANIES ARE CARRYING THE ECONOMY

Source. The SP10 vs SP490 between 2021 and 2025. Only 10 companies—most of them involved in AI—have seen significant earnings growth for years. In real terms—inflation-adjusted—the non-AI market is shrinking.

2. THE CAPABILITIES OF AI HAVE BEEN EXAGGERATED

Source. Blue is what AI could theoretically do work-wise. Red is what is actually doing right now. AI’s limitations show up as the gap between red and blue. Laboratory success doesn’t always transfer to the real world.

3. MOST TOKENS GENERATED BY AI ARE WASTED

Source. 82% of the tokens generated by AI are devoted to fixing bugs generated by AI or other frictions. “82% of every dollar never reaches the product.” The output/$ you gain in one place by using AI, you lose in another.

4. AI RELIABILITY IS BARELY IMPROVING

Source. Across 14 frontier models, reliability improved only modestly, if at all, compared to accuracy, which AI companies love to report in exponentials.

5. THE AI INDUSTRY RUNS ON CIRCULAR DEALS

Source (edited and updated). If AI succeeds, it’s fine. But if it falls short of its promise, then AI companies won’t be able to pay the bills and Big Tech loses twice: 1) AI companies don’t buy chips anymore and 2) their stakes lose value.

6. PEOPLE ARE NOT USING AI TO ITS FULL POTENTIAL

Source. Power users use the “thinking capabilities” of AI 7x more than the median paying user. But the median paying user knows how AI works; they use it all the time. They just… don’t find it as useful or reliable as to deploy it further.

7. AMERICANS DON’T WANT DATACENTERS NEARBY

Source. 71% oppose the construction of a datacenter in their local area; 48% are strongly opposed. Only about a quarter favor construction.

8. AI USERS BELIEVE AI EVEN WHEN IT’S WRONG

Source. When AI was wrong, participants followed it ~80% of the time, performing “worse than having no AI at all.” The authors called this effect “cognitive surrender”: people don’t even check AI’s outputs.

9. WORKERS DON’T THINK AI IS SAVING THEM TIME

Source. 40% of non-management employees say AI saves them no time at all; only 2% of executives agree; the people reporting the AI productivity gains are not the people doing the work with AI.

10. AI IS KILLING THE JOBS OF JUNIOR DEVELOPERS

Source. AI is contracting the labor market for junior developers; ~500,000 fewer of them work today than they would if pre-ChatGPT hiring trends had not changed.

11. THE IMPOSSIBLE MATH OF THE AI BOOM

Source. From the FT: “These numbers show that if the hyperscalers continue on the current trajectory, the AI boom will become a story of one of the largest destructions of shareholder value in history.”

Evan Maxwell

May 29

Alberto: I find your charts fascinating, but I have some questions. (consider that I am 83, an analog grandpa in a digital world.). You mention "right vs. wrong responses." My experience tells me that right and wrong are slippery terms. Some responses, like those in the "real world," are often hard to characterize one way or the other. I don't find all that many purely wrong AI bleats, but when I know something about the subject, I may find them incomplete.

I don't automatically get the distinction between reliability and accuracy.

I also see your distinction between the widespread and the power users. But I wonder whether AI needs many users who are just dabbling. Can it be successful as a technology if it only becomes valuable to "power users?" (One of my favorite movie analogies here is Interstellar, where the true believers had to hide the space program from the common citizenry. The true believers win out, but it is a bittersweet victory.)

Drilling down on that thought, it just occurred to me that military and national security uses may justify a huge amount of AI, whether the average citizen wants or likes it.

Thanks for your work. I had my eyes opened by your earlier piece on public opinion polls about AI. Luckily, I've always been a bit of a contrarian.

Guilherme Peixoto

Jul 14

Stock prices are not economy-carrying, right? You could say only 10 companies are responsible for the entire stock valuation on SP500, but it's neither fair nor correct to state it's *economy carrying*. Stock market prices tell you only about corporate wealth and future expectations, it's not real economy (employment, production, etc)

1 reply by Alberto Romero

3 more comments...

The Algorithmic Bridge

Discussion about this post

Ready for more?