Yesterday something happened that in any other industry would set off alarms. Nvidia announced it will buy $3 billion of Meta stock, and AMD added another billion. Meta, for its part, is one of their biggest customers, spending tens of billions on their chips. Put simply: the suppliers are funding the customer so the customer keeps buying from the suppliers. Wall Street called it a "strategic partnership". I call it something else: money going around in a circle.

It isn't the first time. Over the past year we have seen chipmakers invest in model companies that buy their chips, cloud companies fund startups committed to renting their servers, and GPU leasing deals structured like financial products. Each of these deals makes sense on its own. Together, they paint a picture of an industry in which part of the demand is created by those who profit from it.

The numbers no longer add up

Alphabet has guided to capital spending of $195 billion to $205 billion this year. Microsoft is expected to spend more than $50 billion in a single quarter. Meta, one of the most profitable machines in the history of capitalism, sold $25 billion of stock because the cash it generates is no longer enough for its pace of building. When the richest companies in the world need to raise money to build, that is not a sign of strength. It is a sign that spending is running faster than revenue.

The standard answer is that the revenue will come. Maybe. But the real question is not whether AI is useful, it certainly is, but whether there are enough customers paying enough to justify trillion-dollar infrastructure. Right now, a large share of the cloud giants' AI revenue comes from other AI companies, some of them funded by the same giants. The end customer, the business or person paying out of pocket, is still small relative to the infrastructure being built for them.

Tonight, when Microsoft shows Azure revenue in dollars for the first time and Google reports its cloud backlog, it is worth reading the numbers with a skeptical eye. Don't just ask how much revenue grew, ask where it comes from. A half-trillion-dollar backlog sounds impressive, until you ask how much of it comes from customers the giants themselves fund.

And what about us

Israel has a huge interest in this bubble, if it is one, deflating slowly rather than all at once. The past year's exits, from Wiz to the small startups bought with six employees, rest on the appetite of those same giants. If the giants cut spending, the tap will close on acquisitions first. Israeli founders and investors need to build companies with real customers who pay with their own money, not just one potential buyer in Seattle or Mountain View.

AI will change the world. I have no doubt about that. But the internet changed the world too, and that did not stop the dot-com bubble from bursting in 2000 and wiping out trillions. A revolutionary technology and a healthy business model are two different things. Those who confuse them pay in the end.

Sources and further reading