This article concisely describes the mechanics of the AI bubble, namely, how much of the projected $3 trillion spend on AI and data centers is in the form of (hidden) debt, also known as SPVs (Special Purpose Vehicles).

Much of that debt uses GPU computer chips as collateral. Meaning that if profits fail to materialize, companies will use their computer chips to settle their debts. But these GPU chips depreciate in value rather quickly, and if the market is suddenly flooded with supply, that will push their value down further.

This is partly explains how the AI bubble "popping" may bring down the entire economy. If AI products fail to deliver meaningful profits, and the trillions of dollars of debt used to finance the AI buildout can't be paid back, it could lead to a cascade of businesses and financial institutions collapsing.

KEY QUOTES:

>A Nikkei study put that hidden figure at $1.65 trillion, up roughly eightfold in four years. It is more than the $1.35 trillion the five report outright.

>The mechanics are simple. A company packages the debt for chips, servers, and power into a separate legal entity, often a joint venture, so the cost never flows through its own accounts. [Special Purpose Vehicle]

>Oracle has $260 billion of future lease commitments that will eventually land on its books. Nvidia carries $119 billion in purchase obligations. Alphabet and Microsoft keep their vehicles off-book too. The scale is the story. Meta’s off-balance-sheet debt alone is about $420 billion, nearly triple its reported debt. Oracle’s has grown roughly thirtyfold in four years.

>The industry is expected to spend more than $3 trillion through 2028 building and equipping AI data centres, much of it financed against the chips inside them.