The AI economy is full of financial gimmicks.
Tech journalists need to start moonlighting as finance journalists.
(Side Note: Last week I read Mark Zuckerberg’s new Superintelligence memo and offered my reactions in a piece for Tech Policy Press — “Mark Zuckerberg is Out of Ideas.” If you enjoy it when I get mad and heckle the tech barons, I think you’ll like this one.)
I wish every tech journalist was assigned John Kay’s Other People’s Money as required reading. The book is a decade old, but it could hardly be more timely. I do not believe it is possible to effectively report on the current AI boom without thinking hard about how tech has become finance, and finance has become an elaborate house of cards.
Consider just a few recent headlines:
Mira Murati’s new AI startup, Thinking Machines, is now valued at $12 billion. The company launched in February 2025. It does not yet have any revenues or a product.
A few months ago, Elon Musk sold
TwitterX.com for $33 billion to his own company, X.ai, in an all-stock transfer. X.ai (which makesmechahitlerGrok) is pursuing a valuation of $200 billion in its next funding round. Which will surely be soon, since the company is an utter cash furnace.“[Spending on AI infrastructure] contributed more to growth in the U.S. economy in the past two quarters than all of consumer spending.” Paul Kedrosky notes that “one explanation for the U.S. economy’s ongoing strength, despite tariffs, is that spending on IT infrastructure is so big that it’s acting as a sort of private-sector stimulus program.” (H/T Brian Merchant)
Earlier this year, Sam Altman held a press conference with Donald Trump and Masayoshi Son, announcing a $500 billion AI infrastructure project called “Stargate.” The Wall Street Journal reports that they have run into some major developmental roadblocks, and have massively scaled back the project.
A question that every tech journalist on the AI beat ought to be repeatedly asking is “how much of this money is actually real?”
When Elon Musk pays himself $33 billion in stock from one of his companies to acquire another one of his companies, that isn’t real money. It’s an accounting gimmick.
When early investors give Mira Murati $2 billion in exchange for 1/6th of her company, the company is nominally valued at $12 billion. But that’s also an accounting gimmick. Murati has $2 billion. Once she has spent that money, she will have to go out and ask for some more. She still may not even have a product at that point. All she has is vibes.
And yet, as we learn from John Kay’s book, the vast majority of financial activity is effectively just accounting gimmicks. (Read Kay’s “The Parable of the Ox,” which was published by FT in 2012 and later became the prologue of his book.)
Kay estimates that only 3 per cent of the financial activity in Britain’s financial sector (circa 2015) involved banks lending money to firms and individuals engaged in the production of goods and services. The other 97% was banks taking the other side of financial bets with other banks.
Let’s say that two large financial institutions make a $1 million bet on a coinflip proposition. (The coin is a quarter. It is worth $0.25.) One bettor will win a million and the other will lose a million. They make these bets routinely, and it all pretty much averages out in the end. Should we consider this transaction to represent $2 million in economic activity (since 2 players are each risking a mil), or $0 in economic activity (since it all evens out), or $0.25, since they did make use of a quarter after all? The simple, correct answer is “Well, it depends.” But the more accurate answer is “let’s call it zero for tax purposes and $2 million for other purposes, to suit the interests of the people making these bets.”
It is not clear, at least from the reporting I have seen, how much of these economy-boosting capital expenditures represent actual concrete being poured/material infrastructure being built. At least some of the capital expenditure is tech firms paying other tech firms for cloud computing costs, using the money that the first tech firm invested to begin with.
If Microsoft or Google nominally invests a billion in a new AI startup, and that AI startup then turns around and pays Microsoft/Google a billion in cloud computing costs, is that $2 billion in economic activity, or $0, or some third number? I suspect we will eventually realize that the answer varies according to whatever suits the interests of the people making these bets.
[Note: Ed Zitron has been doing an admirable job of digging into these capital expenditures and accounting gimmicks. His latest, “The Haters Guide to the AI Bubble” is a 55 minute read, but it’s absolutely worth it.]
There is a lot of accounting gimmickry at play here. It’s reminiscent of Global Crossing. It’s reminiscent of Enron. And the crucial difference between the dotcom crash (which was at least as much a telecom investment story as it was a Pets.com story) and today is that telecom circa 2000 did not represent more economic activity than all consumer spending.
Tech reporters keep getting lost in the headline numbers, insisting that “AI isn’t a bubble” because OpenAI and Anthropic have revenues and use-cases. But the telecom industry had revenues and use-cases 25 years ago too. It was still a bubble, propped up by reporting that took all that financial gimmickry at face value, along with investing that presumed the bubble would never pop.
The same thing happened in the 2008 great financial crisis. There was the underlying economic activity — people building and buying houses — and then there was the financial gimmickry. It turned out that the finance gamesmanship far outpaced the actual economic behavior, and things got out of hand, and then the whole house of cards came crashing down.
In retrospect, journalists on the housing beat in 2007-08 were doing finance journalism whether they realized it or not. They largely didn’t realize it, and that meant they weren’t covering the most important story.
AI, today, is finance. It’s other things too. But the financial component is too big — and too full of strange gimmicks, financial gamesmanship, and self-dealing — to be ignored.
Tesla and crypto, both on the periphery of the AI bubble, are even more striking examples. Tesla has a declining business making cars and batteries and a trillion dollar valuation based on the hope that 10 or 20 (partly) self-driving cars in Austin are the precursor of a robotaxi business that will dominate the world. Crypto is as big as NVIDIA and utterly useless.
These examples are helpful because the absence of any real value is obvious, implying that financial market valuations bear no relation to actual economic activity. That shows how much weight should be placed on the market valuations of AI stocks
This is just GOLD: "When early investors give Mira Murati $2 billion in exchange for 1/6th of her company, the company is nominally valued at $12 billion. But that’s also an accounting gimmick. Murati has $2 billion. Once she has spent that money, she will have to go out and ask for some more. She still may not even have a product at that point. All she has is vibes."