- The five biggest AI infrastructure spenders, Meta, Alphabet, Microsoft, Oracle, and Amazon, have disclosed roughly $1.67 trillion in leases and purchase commitments in their latest filings.
- About $821 billion of that total sits in leases that have not yet commenced, which stay off the balance sheet until the data centers are delivered.
- Moody's warns the recorded liability understates the real exposure, noting uncommenced data-center leases that at one point equaled 113 percent of the five firms' adjusted debt.
Five companies, $1.67 trillion, and most of it not yet on a balance sheet
The scale of the AI infrastructure race is usually measured in annual capital spending, where 2026 guidance across the largest hyperscalers already runs past $700 billion. The larger and less visible number is what these companies have contractually committed to. Across their most recent quarterly filings, Meta, Alphabet, Microsoft, Oracle, and Amazon disclosed about $1.67 trillion in leases and purchase commitments tied to data centers, servers, cloud capacity, and the energy to run them.
The composition matters more than the headline figure. Roughly $848 billion is purchase and construction commitments, and about $821 billion is leases that have been signed but not yet commenced. Per company, the totals are led by Meta at about $420 billion and Alphabet at about $408 billion, followed by Microsoft at roughly $339 billion, Oracle at $273 billion, and Amazon at $229 billion.
| Combined leases and purchase commitments, five hyperscalers | ~$1.67T |
| Signed leases not yet commenced, off balance sheet until delivery | ~$821B |
| Largest individual commitments (Meta, Alphabet) | $420B / $408B |
| Uncommenced leases as a share of the five firms' adjusted debt | 113% |
Why the off-balance-sheet structure is the real story
An uncommenced lease is a real obligation that does not appear as a liability until the facility is handed over, at which point it converts to recognized debt. That accounting treatment is what makes the current buildout hard to read from the balance sheet alone. Moody's Ratings, in a February 2026 analysis, put the uncommenced data-center lease commitments across five hyperscalers at $662 billion, equal to 113 percent of their adjusted debt, and cautioned that reported figures may not capture the full exposure as those leases go live. The mismatch is structural: AI hardware has a useful life of four to six years, while data-center leases run ten to fifteen, and residual value guarantees used to bridge that gap sit off the books until a payout becomes probable.
The accounting liability is unlikely to reflect certain plausible future scenarios.Moody's Ratings, February 2026 analysis of hyperscaler data-center commitments
The takeaway for anyone tracking the AI trade is that the true cost of the buildout is not fully visible in the capital expenditure line that earnings coverage fixates on. It is sitting in commitments that convert to liabilities on a delay, locked in by contracts signed today and paid down over the years it will take the data centers to earn their keep. The bet has already been placed. The balance sheets have simply not caught up to its size.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.