Beyond Marathons and Backflips, China’s Robots Face a Commercial Test

Big tech companies are making massive, multi-trillion-dollar bets on artificial intelligence infrastructure, but a substantial portion of these future financial obligations are not reflected on their balance sheets. A Wall Street Journal analysis revealed that nine top tech firms have around $3 trillion in off-balance-sheet commitments, largely tied to AI development and capacity. These commitments are expanding more rapidly than traditional capital expenditures, which totaled about $600 billion in the past reported year.

Companies like Alphabet, Meta Platforms, Oracle, and others are basing these huge investments on assumptions about future demand for AI computing power and the availability of AI hardware. Their strategy relies on generating sufficient future revenue from widespread consumer and business adoption of AI to cover these obligations. However, if these technological and demand forecasts are inaccurate, these deals to secure future capacity could transform into a significant financial burden for the companies and their investors.

Meta's "Hyperion" data center project in Louisiana, spanning an area equivalent to 1,700 football fields, exemplifies how large obligations can remain off-balance sheet. Meta initially agreed to lease the facility for four years starting in 2029, with options to extend for up to two decades. The company guaranteed bondholders would be compensated if it didn't occupy the space for the full term. Since Meta doesn't consider payments under this guarantee probable, it hasn't recorded a liability. Under accounting rules, these lease obligations will stay off-balance sheet until payments begin. Meta's aggregate initial lease commitment for Hyperion is approximately $12.3 billion, with total disclosed obligations for uncommenced leases reaching $347 billion as of June.

Across the companies analyzed by the Journal, promises for payments under these uncommenced leases amounted to $1.2 trillion in off-balance-sheet obligations, a fourfold increase from the previous year. In addition to Meta, the analysis included commitments from Alphabet, Amazon.com, Microsoft, Oracle, Nvidia, Broadcom, SpaceX, and Advanced Micro Devices. These companies often secure long-term contracts for hardware like Nvidia chips and memory modules well in advance to guarantee production, contributing to a staggering $1.9 trillion in purchase obligations examined.