So. Apparently, to justify $1.5 trillion a year of AI infrastructure spending, the AI industry needs to generate $6 trillion in annual revenue by 2031. It is all in last Tuesday's Bain & Company annual Global Technology Report.

For scale: that is more than the GDP of any country on earth except the United States and China. It is also roughly the annual revenue of the entire global oil and gas industry.

Now, how does one arrive at such a number?

Well... kinda not so complicated.

You start with the spending. Five to six and a half trillion dollars going into data centres by 2030, with annual AI infrastructure spend reaching 1.5 trillion. Then you apply a rule of thumb from the cloud industry, that capex should be about a quarter of revenue. Then you divide. 

It is a "what needs to be true" analysis. Kind of: "here is what we built, the world now owes us the demand." To be fair to Bain, they put a caveat around the calculation themselves: the more important question, their report admits, may be whether enough economic value can be created to justify it.

And where will the 6 trillion come from? Existing AI services, everything that makes money today, consumer subscriptions, enterprise tools, all of it, tops out at 1.8 trillion in Bain's own math. The remaining 4.2 trillion, seventy percent of the target, is assigned to products that do not exist yet. Robotics. Autonomy. Drug discovery.

The lead author, David Crawford, is honest about this too, to his credit.

Now. Your friends at Meme Guide to Management have seen this document before. Three times, actually.

In the 1840s, British railway prospectuses promised investors dividends around ten percent. The Economist, then a young magazine, warned in real time that the math could not work. By 1852, the network had nearly quintupled to 7,300 miles of track, and according to the economist Andrew Odlyzko, who studied the mania, it produced one-fourth of the revenue the builders expected.

His explanation: everyone inside the boom was paid by its continuation.

In the late 1990s, the ridiculous assumption was a demand number: internet traffic, the industry said, doubles every 100 days. The claim came from WorldCom, and in 1998 the US government repeated it in an official report. One mathematician at AT&T Labs did the actual division. His name was... Andrew Odlyzko. Same man.

He found traffic doubled once a year, not every hundred days. The difference between 1000 percent and 100 percent turned out to be the difference between boom and bust: half a trillion dollars spent, 80 million miles of fiber laid, and by 2002 only 2.7 percent of it carrying any light at all.

WorldCom became the largest bankruptcy in history.

And in March 2000, Barron's published "Burning Up," an audit of 200 dot-coms counting exactly how many months of cash each had left. Half had less than a year. It ran ten days after the NASDAQ's all-time peak and was correct in nearly every particular.

In 2007, the US Congressional Budget Office projected healthcare spending growth and found that it would eventually exceed 100 percent of GDP.

So, a small toolkit for your own decks, before you laugh too hard at anyone else's.

First, find the plug variable: in every plan there is one number that was solved for rather than evidenced. Demand in 1999. Revenue in 2026. "Synergies" in your last merger model.

Second, ask the 35 percent question: what fraction of this plan can its author actually see, and what is filed under "other"? And third, run the Stein test.

The economist Herbert Stein left us one law: "If something cannot go on forever, it will stop." Data centre costs are doubling every 12 to 16 months. That is a sentence with only one possible ending, and Stein wrote it in 1976. 

None of this means the technology is fake.

It means two things are true at once: the technology is real, and the business case is unfinished.

A manager who can hold both without a panic face is worth more right now than most of the 4.2 trillion.

Talk soon,
Sultanbek