How much is Big Tech really spending to build out AI? A lot of it does not appear where investors usually look. Nikkei Asia reports that five large US technology companies now carry about $3.2 trillion in AI-related obligations that sit off their balance sheets, up roughly 90% in just six months. For anyone trying to judge whether the current AI spending is sustainable, that is the number the headline figures leave out.
The word debt needs care here. These are not bonds or bank loans but future lease payments and multi-year purchase commitments: contracts for Nvidia chips, server deliveries and data-centre capacity that has not come online yet. Under US accounting rules such commitments are disclosed in the footnotes of a filing rather than on the main balance sheet, and staying off it until the rules require recording is entirely legal. The effect is that a cost which is very real to the business is easy for an ordinary reader to miss.
The scale is what makes it notable. An earlier Nikkei count in July put the hidden obligations of five giants at about $1.65 trillion, so the figure has roughly doubled in half a year as AI build-out accelerated. The report says the commitments now equal around 80% of the companies' assets, a ratio that would have looked extreme a year ago. The companies involved are the familiar names behind the models and clouds we cover, including Google and Meta, so this is the money underwriting the infrastructure that today's AI products run on.
The risk analysts point to is straightforward. If demand for AI services falls short of these multi-year bets, today's lease and purchase commitments could turn into write-downs, hurting shareholders and the private-credit investors who financed the construction. A commitment that is cheap to carry while revenue grows becomes a burden the moment it does not.
The honest caveat is about the number itself. The public excerpt does not break the total down company by company or set out exactly how it was calculated, and other analysts have reached different totals using different definitions of what counts, with one bank putting the footnote figure even higher. Treat $3.2 trillion as a well-sourced estimate of a trend rather than an audited line, and watch whether the pace of growth, not just the level, holds in the next set of filings.