Big Tech's hidden $3T in AI obligations dwarf reported capex
The AI buildout's bill just got a lot bigger — and most of it was never on the books. A Wall Street Journal analysis of nine top tech companies found roughly $3 trillion in AI-related off-balance-sheet commitments, far exceeding the roughly $600 billion those same companies reported as capital expenditure. The gap tells you something about how this boom is being financed, and how exposed investors may be.
A WSJ analysis found nine top tech companies carry ~$3 trillion in AI-related obligations that never appear on their balance sheets — roughly five times their reported ~$600 billion in capex. These are commitments tied to data-center leases and multi-year chip purchase deals, structured so they don't count as traditional debt or spending. Reporting on the finding is wide: the Financial Times and Nikkei Asia have separately flagged similar "hidden debt," with Nikkei pegging the figure around $1.65 trillion for five US giants, while Bloomberg Tax drew an explicit parallel to the accounting devices that helped bring down Enron.
Why it matters: the off-balance-sheet structure is what let Big Tech spend at scale without spooking earnings — but it also means the industry's real exposure to an AI downturn is far larger than its reported books suggest. If demand softens or financing tightens, those lease and chip commitments don't evaporate; they're contractual. This ties directly into the broader picture we've tracked of a buildout straining at its financial seams — The $1 trillion AI build-out hits a wall cash can't fix. The market is starting to price the risk: investors like Michael Burry have publicly pointed at the exposure. It's the rare AI story where the number that matters isn't a model score — it's an accounting footnote.
Chinese robotics firm RoboScience unveiled REX G1, a wheeled humanoid general-purpose robot it positions as an "embodied-intelligence productivity partner." The company's first robot, REX G1 runs its in-house Visics embodied-AI model and targets logistics, factory, retail, and home settings, pairing 22 degrees of freedom with a claimed ±0.1 mm repeat positioning accuracy. RoboScience's pitch is that the unit adapts to existing warehouse layouts rather than forcing space to be redesigned — it can thread 0.75-meter aisles and work a continuous 0.1-to-2.0-meter vertical range, with hot-swappable batteries that resume work in about five seconds. It's another sign the embodied-AI competition is shifting from demos toward machines engineered for real, continuous shifts, even as the category's funding and product cycles accelerate.
If $3 trillion in hidden AI obligations came due in a downturn, which balance sheets would you worry about first? Tell us in the comments.
