- A WSJ analysis of company filings reveals that Alphabet, Amazon, Meta, and Microsoft have collectively accumulated over $2.4 trillion in off-balance-sheet AI spending commitments — including $1.52 trillion in purchase commitments and $904 billion in data-center leases not yet started — obligations that are entirely absent from the quarterly capex figures that markets typically focus on.
- The on-balance-sheet picture already looks enormous ($248 billion in lease liabilities and $356 billion in long-term debt), but the off-balance-sheet figures dwarf it: Alphabet alone has $811 billion in leases not started, Meta has $347 billion in purchase commitments, and Microsoft has $329 billion in purchase commitments — suggesting the AI infrastructure buildout is far more front-loaded and locked-in than public disclosures imply.
- These hidden commitments matter for capital markets because they represent future cash outflows that will require financing — additional bond issuance, revolving credit, or equity — and because they signal that the AI capex cycle has extraordinary momentum that cannot easily be unwound even if the competitive or economic environment changes.
- For investors trying to assess the true financial commitment these companies have made to AI, quarterly capex disclosures are materially incomplete: the real number is roughly $3 trillion larger than the balance-sheet figures that appear in standard earnings presentations.
What Happened?
Each quarter, investors scrutinize the capital expenditure figures that big tech companies report as the primary signal of their AI commitment. But a WSJ analysis of regulatory filings shows those numbers severely understate the true financial obligations these companies have locked in. The gap lies in two categories of off-balance-sheet commitments: purchase commitments for chips, hardware, and services (totaling $1.52 trillion across the four companies), and data-center lease agreements that haven’t yet started ($904 billion). Neither category appears in headline capex. Together, these obligations add more than $2.4 trillion to the already-enormous on-balance-sheet figures — a number that is roughly $3 trillion higher than what earnings coverage typically conveys.
Why It Matters?
The scale and lock-in nature of these commitments has several important implications. First, for bond markets: as these companies draw on credit facilities and issue debt to fund future outlays, they add to the long-duration supply that is already pressuring Treasury yields — precisely the dynamic that Bank of America warned about this week. Second, for competitive dynamics: these commitments are contractual, meaning even if AI demand disappoints or model efficiency improves faster than expected, the capex will flow through regardless. Third, for earnings transparency: investors making decisions based on reported capex are working with an incomplete picture of how capital-intensive these businesses have become. The $3 trillion gap is not a rounding error — it is the hidden load-bearing structure of the AI economy.
What’s Next?
As more of these off-balance-sheet commitments convert into active spending and balance-sheet debt, the financing pressure on credit markets will intensify — reinforcing the “reverse crowding out” dynamic already visible in Treasury yield data. Investors and analysts will increasingly be forced to look beyond headline capex to understand the true capital commitments at play. For regulators and accounting standard-setters, the size of these off-balance-sheet obligations may eventually prompt questions about whether current disclosure frameworks are adequate for an era in which a handful of companies are making multi-trillion-dollar infrastructure bets. And for the companies themselves, the commitments signal that regardless of near-term AI revenue trajectories, they are deeply and durably invested in building the infrastructure layer of the AI economy.
Source: The Wall Street Journal
